This post may not be for everyone; it consists of a paper I wrote about mRNA vaccines. This was my first time writing a scientific paper on my own and so the jargon may be too technical and convoluted for most to understand. But if you want an in-depth history of vaccine technologies and mRNA vaccines, and are willing to put up with my crazy writing, then please take a read and let me know what you think.

Abstract

Vaccines are one of the most impactful biomedical developments in history. They prevent between 3.5–5 million deaths every year and have helped eradicate diseases such as smallpox and polio. Recently, vaccines have significantly reduced the severity of symptoms and mortality caused by SARS-CoV-2. Despite incredible advances in vaccine technology, including the most recent development and commercialization of messenger RNA (mRNA) based vaccines for SARS-Cov-2, there remain several limitations. Vaccine induced immunity is often limited in its breadth, resulting in poor protection to rapidly mutating viruses, and its durability, requiring patients to receive a complex booster regime. Furthermore, many vaccines require complex storage conditions relying on the cold chain, preventing equitable global access in regions that lack the infrastructure. This review will discuss recent advances in vaccine technologies and highlight mRNA vaccine strategies that seek to overcome current technological challenges to create next generation vaccines that are more biomimetic, durable, and potent.

Keywords

Vaccines, mRNA, Immunology, Lipid nanoparticles, Vaccine technology

Introduction

Vaccines are a biological product intended to safely stimulate and prepare the immune system against infection or disease. They leverage the ability of the mammalian innate and adaptive immune systems to recognize, respond to, and remember pathogens, and in countries with high vaccination rates, many diseases responsible for high childhood mortality have essentially disappeared. These successes notwithstanding, the recent COVID-19 pandemic demonstrated that infectious diseases with high incidence or fatality rates pose a substantial threat to both public health and the global economy. Thus, it has become imperative to be able to respond to such threats more rapidly. While strategies such as social distancing, quarantine, and lockdown control the spread of an emerging pathogen, they are logistically challenging and hard to enforce long-term. Globally accessible and potent vaccination is ultimately the best approach to limit disease transmission and establish herd immunity.

The earliest documented practice of variolation – intentionally exposing healthy people to a virus – took place in the 16th century when people attempted to prevent illness by exposing healthy people to smallpox pus or scabs (1). In 1774, Benjamin Jesty made a breakthrough, testing his hypothesis that infection with cowpox could protect a person from smallpox. Shortly after, in 1796, Dr. Edward Jenner observed that the milkmaids who had cowpox lesions were immune against smallpox infection (1-2). Jenner’s following work evaluating immune protection in James Phipps and other children laid the foundation for modern vaccinology, a term coined from the Latin word vacca for cow (3).

Over the following two centuries, advancements in biomedical technology and immunology enabled the design of live-attenuated, whole-inactivated, sub-unit, and mRNA vaccines, among other platforms. For example, the evolution of cell culture led to the creation of the polio vaccine, and soon afterward vaccines for measles, mumps, rubella, and varicella were developed (4). The introduction of recombinant DNA and whole-genome sequencing techniques were major milestones in vaccine development, giving researchers the tools to develop new vaccines against pathogens, something not possible before (1). Other important milestones in vaccine research are the development of recombinant viral vector vaccines (5), virus-like particle vaccines (6), protein-based vaccines (7), and toxoid vaccines (5-8). Most recently, an important milestone for next-generation vaccines was the development of mRNA vaccines, which garnered increased attention after their rapid development and approval for the COVID-19 pandemic. The facile scalability and expedited approval can be attributed to mRNA vaccines’ unique safety profile, ease of fabrication, and ability to harness the patient’s own cellular machinery to express the desired vaccine antigen, the protein or peptide against which an immune response should be generated.

We are currently in the era of mRNA vaccinations. mRNA, which was discovered during pioneering studies between 1947–1961 (9), is a transient intermediate between genes and proteins (9). Efforts in the early 1990s showed that in vitro transcribed (IVT) mRNA vaccines can induce the production of proteins in animal models, with epitope presentation proving effective (10-11). In vitro transcription is a reaction in which a linearized DNA plasmid containing the gene of interest is transcribed to the mRNA sequence. Since the proof-of-concept animal studies in the early 1990s, numerous strategies have been explored to ameliorate the instability and immunogenicity of IVT mRNA (12). The true importance and efficacy of this vaccine technology, however, was realized when mRNA vaccines were developed and approved for the COVID-19 pandemic. They were developed in a record-breaking time of less than a year and their widespread vaccination to millions of people helped to control the COVID-19 outbreak. The development, approval, and manufacturing capabilities demonstrated by the makers of these vaccines has validated the mRNA platform as a safe and effective tool for vaccination. The goal of this review is to provide an overview of mRNA vaccines, focusing on their pharmacology, types, the unique immune responses they generate, mechanisms, benefits, and drawbacks.

Overview of Immune Response to mRNA vaccines

Vaccination is intended to mimic real infection as closely as possible without exposing the vaccine recipient to undue risks. There are two main mechanisms of the immune response: humoral and cellular immunity. In humoral immunity, B-lymphocytes cells (B cells) produce antibodies that bind to antigens, neutralizing them or preventing foreign substances (i.e viruses, bacteria, etc) from entering or damaging cells. Cellular immunity, on the other hand, relies on the activation of phagocytes, such as macrophages and natural killer cells, antigen-specific cytotoxic T-lymphocytes (CD8+ T cells), and the release of various cytokines in response to an antigen.

After a vaccine is administered, antigens are processed by a diverse population of immune cells known as antigen-presenting cells (APCs), which includes dendritic cells, macrophages, Langerhans cells, and B cells. Traditional vaccines, such as subunit or live attenuated vaccines, deliver whole proteins or peptides as their antigens while mRNA vaccines introduce a piece of mRNA that corresponds to a viral protein, acting as its antigen. The mechanism of antigen processing has important consequences on how immunity is induced. Vaccine antigens that are produced in or enter the cytoplasm (eg. live attenuated viruses) are displayed on class I major histocompatibility complexes (MHC-I) via the endogenous antigen-processing pathway (13). These antigens are then recognized by T cell receptors (TCR) of a particular subset: cytotoxic T cells (14). On the other hand, antigens that enter cells via phagocytosis are displayed on class II MHC (MHC-II) by the exogenous antigen-processing pathway (13). These antigens are recognized by a different subset of T lymphocytes: a naïve form of T helper cells (CD4+ T cells) (15).

mRNA vaccination, once administered intramuscularly, leads to potential adaptive immune system activation by the following pathways: (i) transfection of muscle and skin cells, activating immune cells and helping prime CD8+ T cells (16); (ii) transfection of tissue-resident immune cells such as the dendritic cells (DCs), macrophages, and Langerhans cells at the injection site, initiating priming and activation of not only T cells but also B cells (17); and (iii) transport to secondary lymphoid tissues, such as the lymph nodes and the spleen (17-18). After leveraging the translational machinery, such as the ribosome, of the host cell, the mRNA introduced by the vaccines is translated into proteins. The resulting translated proteins are processed and presented on either MHC-I or II. mRNA vaccines can then leverage both the endogenous and exogenous antigen-processing pathways. The endogenous antigen-processing pathway is used after proteasomes degrade cytoplasmic proteins, thus generating antigenic peptide epitopes that are loaded onto MHC-I molecules. MHC class I then can present these peptides on the surface of the cell for CD8+ T cells (19). This helps in establishing cellular immunity to the antigen expressed from the mRNA. Alternatively, the activation of APCs can result in the exogenous antigen-processing pathway, where secreted exogenous proteins are taken up by APCs, either residing in the tissue or draining lymph nodes, and then are processed and presented on MHC-II molecules (19).

The antigen is co-presented with macromolecular structures associated with pathogens. These pathogen associated molecular patterns (PAMPs) are recognized by pattern recognition receptors (PRRs), such as Toll-like receptors (TLRs) (20-21). The inclusion and design of these immunostimulatory molecules, often referred to as adjuvants, have become an active area of research, which will be more thoroughly considered in the following section. Briefly, the interaction between these adjuvants and PRRs on APCs triggers key intracellular signaling events that promote phagocytosis, maturation, and secretion of cytokines (22).

Activated APCs displaying vaccine antigens on MHCs then migrate to secondary lymphoid organs (eg., draining lymph nodes or spleen). Here APCs encounter naïve T cells, T cells that have not undergone activation, in T cell zones (23-24). The interaction between APCs and naïve T cells through MHC-TCR binding leads to the differentiation and proliferation of naïve T cells into effector cells.

In response to MHC-TCR binding, and cues from cytokines, CD4+ T-helper cells differentiate into two subsets of effector T-helper cells: T helper 1 (Th1) and T helper 2 (Th2) cells. Th1 and Th2 cells are primarily responsible for cellular and humoral immunity respectively. Th1 cells produce Interferon-gamma (IFN-γ) as their cytokine in order to stimulate the activation and expansion of CD8+ T cells. Th1 cells are further involved in the production of IgG1 and IgG3 antibodies by B cells (15). CD8+ T cells differentiate into cytotoxic killer T cells following TCR/MHC-I interaction and help from Th1 cells (e.g., INF-γ). Development of cytotoxic killer T cells following vaccination is important because they can recognize and eliminate infected cells. In addition to the effector cells that are generated in response to the presentation and recognition of vaccine protein antigens, both CD4+ and CD8+ T cells also differentiate into memory cells, which are critical in responding and expanding the clonal pool upon subsequent encounter with the same pathogen.

In promoting humoral immunity, Th2 cells secrete IL-4, IL-5, and IL-13 as their signature cytokines to promote the development, maturation, and differentiation of B cells into memory B cells (MBCs) and antibody-secreting plasma cells (Figure 1).

A number of additional T cell subclasses exist, each of which have niche roles and interactions in mediating immune responses to vaccines. For example, T follicular helper cells (Tfh) and Th17 cells are two additional subtypes of CD4+ cells that are essential for the generation of high-affinity antibodies and mucosal immunity. Tfh cells regulate B cell affinity maturation (somatic hypermutation), selection of high-affinity germinal center (GC) B cells, and the duration of GC reactions (14)(25). The quality of these processes ultimately determines the durability and potency of the humoral immune response through the generation of antibodies (by B cells) with high affinity. Antibodies are large (150 kDA MW) proteins that uniquely bind to epitopes on the antigens against which a patient is vaccinated. Having high quantities of circulating antibodies, or a robust memory response which can quickly generate new antibodies, is essential to preventing infection upon natural viral exposure. These circulating or memory-produced antibodies are capable of opsonizing viral particles, thereby preventing viruses from entering and infecting cells, or binding to secreted proteins, resulting in their degradation.

Durable GC reactions favor the differentiation of GC B cells into high-affinity MBCs and antibody-secreting long-lived plasma cells (LLPCs) (26)(27). MBCs are important in vaccine-induced immunity and for providing protection, as they can rapidly expand, and differentiate into short-lived antibody-secreting plasma cells upon antigen re-encounter (28). LLPCs migrate from the draining lymph nodes to the bone marrow, where they continue to produce antibodies for several months to decades. LLPCs are terminally differentiated, and, unlike MBCs, do not need reactivation or re-exposure to antigens. The high levels of neutralizing antibodies produced by LLPCs provide protection against reinfection.

B cells are also able to recognize and respond to vaccine antigens before T cells are engaged. Following vaccine administration, B cells detect and internalize antigens, and upon PRR activation, differentiate into plasmablasts, short-lived antibody-secreting cells, that produce the initial wave of antibodies.

The above aspects of the immune response help guide the development of vaccines and vaccine technologies to meet their intended use. For instance, the development of vaccines against intracellular infectious agents like viruses requires technologies or platforms that promote the endogenous antigen processing pathway or cross-presentation to induce potent cytotoxic T cell responses, which are essential for eliminating intracellular pathogens. Additionally, identifying vaccine technologies and/or adjuvants that effectively promote Tfh cell and GC responses, as well as generation of LLPCs, is vital for developing effective vaccines against both current and emerging infectious diseases.

Vaccine design

The most important component of vaccines are one or more protein antigens, either directly derived from the pathogen of interest or biomanufactured (15). Polysaccharide antigens also exist and can also induce protective immune responses. They have been the basis of vaccines like pneumonia and meningitis caused by Streptococcus pneumoniae (29). These antigens are responsible for inducing immune responses that provide protection. Other vaccine components may include preservatives (eg. thimerosal, phenol, formaldehyde) to prevent contamination by bacteria or fungi; stabilizers (eg. gelatin, lactose, sorbitol) to keep the vaccine potent during transportation and storage, especially under temperature variations; and excipients (eg. aluminum salts) to ensure vaccine consistency and efficacy (15). Adjuvants are often added to increase potency, reduce the dose of vaccine which must be administered, and increase efficacy in some populations (e.g., infants, elderly, and immunocompromised individuals)(30)(Figure 2).

Adjuvants primarily function by promoting the generation of APCs and co-stimulatory signals through the activation of APCs. After antigen peptides are taken up and processed, APCS present them on their surface via MHCs. Co-stimulatory signals include co-stimulatory molecules (​​e.g., CD40, CD80, CD86) expressed on the surface of APCs and secreted inflammatory cytokines (e.g., IL-6, IL-10, IL-12, and TNF-α) (31). The production of these two signals can strongly enhance the activation of naïve T cells, thereby boosting the adaptive immune response (32).

There are a few adjuvants that are commonly used in licensed vaccines: aluminum adjuvants (33), MF59 (34), AS03 (35), AS04, CpG ODN 1018, and AS01 (36) are classical human vaccine adjuvants (37)(31). Collectively, these adjuvants act to stimulate pathways that would be stimulated by danger-signals coming from a virus during natural infection. For mRNA vaccines however, the mRNA molecule serves as both an immunogen and adjuvant, due to the intrinsic immunostimulatory properties of mRNA.

Immune responses to mRNA vaccines greatly rely on the delivery system, the immunogenicity of the encoded antigen, and the longevity and subcellular localization of antigen expression. Intramuscular and intradermal administration of mRNA vaccines is highly immunogenic and induces local cytokine and chemokine production that initiates prompt recruitment of neutrophils, monocytes, and other cells to prime the immune response. Injection of mRNa encapsulated in lipid nanoparticles (LNPs) has been shown to induce robust infiltration of neutrophils, monocytes, and dendritic cells as well as the activation of pro-inflammatory cytokines and chemokines in mice (38)(39). LNP formulations are made up four components: ionizable (for example, 1,2-dilinoleyl oxy-N,N-dimethyl-3-aminopropane (40)) or cationic (for example, 1,2-di-O-octadecenyl-3-trimethylammonium-propane (41) and Dimethyldioctadecylammonium bromide (42)) lipids, phospholipids (for example, phosphatidylcholine and phosphatidylethanolamine), cholesterol, and polyethylene glycol (PEG) lipids. These lipids can improve nanoparticle properties, such as particle stability, delivery efficacy, tolerability and biodistribution (43)(12). The most common design of LNPs involves PEG lipids to avoid immune recognition, extend their blood circulation, and improve LNP during synthesis and storage (44-45).

mRNA vaccines have been shown to promote Th1-type immune responses, and potent inductions of antigen-specific germinal center and T-follicular helper cells responses (38)(46-47). Previous research indicates that adjuvant activity of LNPs relies on the ionizable lipid component and IL-6 cytokine induction, rather than MyD88- or MAVS-dependent sensing of LNPs (38). Compared to inactivated or recombinant vaccines, mRNA vaccines using LNP adjuvants induce a stronger and more sustained immune response. This is likely due to the type and amount of cytokines released by the LNP adjuvants as well as prolonged antigen presence in the body up to ten days after intramuscular and intradermal injections (48), allowing for an extended period of immune activation.

mRNA vaccines

mRNA is a single-stranded ribonucleic acid that is transcribed using DNA as a template and can carry genetic information to guide protein synthesis. The idea of mRNA-based therapeutics emerged over three decades ago when Dimitriadis (49), Malone et al. (41), and Wolff et al. (11) provided the first evidence that mRNA produced in vivo and IVT could be introduced into cells and animals to produce proteins. Major limitations such as potent inflammation and reduced in vivo translation due to mRNA short half-life were quickly recognized. Inflammation-mediated inhibition of protein production, instability of the mRNA molecules, easy degradation by enzymes, and difficulties in delivering mRNA into cells effectively further limited the potential clinical and therapeutic application of the platform (50). Overcoming these shortcomings significantly improved the platform, enabling the successful development of vaccines and adjuvants. Martinon et al. and Conry et al. (51) showed that mRNA loaded into liposomes elicited antigen-specific cytotoxic T lymphocytes and humoral responses, paving the way for mRNA vaccine development and early human trials (Figure 3).

Over the past decade, major technological innovation and research investment have enabled mRNA to become a promising therapeutic tool. Recent advances that were critical for the development of a safe and potent mRNA vaccine platform include the incorporation of modified nucleosides into mRNA (52-53) and the removal of contaminants using purification chromatography (54-55), which were both pioneered by Kariko and Weissman. Improvements in sequence engineering and codon optimization (56), innovations in cap moieties and capping strategies (57), and the evolution of potent and relatively safe delivery systems such as lipid nanoparticles (LNPs) (58) have also significantly advanced the development and regulatory approval of mRNA-based vaccines. Nucleoside modification and elimination of double-stranded RNA contaminants generated during IVT have improved the intrinsic adjuvant effect of the IVT mRNA, improved tolerability, and boosted antigen expression by several folds (59)(52). New methods for adding caps to mRNA molecules have improved how much usable mRNA is produced and reduced its detection by the body's immune system sensors (eg., RIG-I and MDA5), again resulting in improved translation, enhanced safety, and reduced costs. Two mRNA vaccines are currently FDA-approved and are both for the SARS-Cov-2 pathogen (COMIRNATY® and SpikeVax®)(Figure 4).

Pharmacology of mRNA vaccines

mRNA structure is composed of 5’cap, untranslated regions (UTRs), and the poly(A) tail, and the open reading frame (ORF) encoding target antigen (60)(Figure 5).

The 5′ end of the mRNA features a 7-methylguanosine (m7G) cap, connected to the first nucleotide by a triphosphate bridge (m7GpppN). m7GpppN is called a 5’ cap, which is a protective structure which protects RNA from exonuclease cleavage, regulates pre-mRNA splicing, and initiates mRNA translation and nuclear export of the mRNA to the cytoplasm (17)(61). The 5’ cap structure can enhance the translation of mRNA. Combining a 5’ cap with eukaryotic translation initiation factor 4e is the key to effective translation (62). The cap is also one of the determinants of mRNA degradation as it regulates the degradation of mRNA, in combination with scavenger decapping enzymes Dcp1, Dcp2, and DcpS (63).

Although UTRs do not encode the desired antigen or a protein, they play a critical role in regulating mRNA expression and stability. These regions are situated between the ORF and the 5′ and 3′ ends (17). They also facilitate mRNA recognition by ribosomes and contribute to post-transcriptional modification of mRNA (64). Inclusion of naturally occurring sequences, such as those derived from alpha- and beta-globins, have been widely used to design mRNA constructs for vaccines (65-66). For example, Zeng et al. developed de novo 5′ UTR sequences based on the guanine–cytosine content and length for mRNA vaccine development. (17)(67).

The IVT mRNA has a polyadenylated section at its 3′ end which is known as the poly(A) tail. The poly(A) tail regulates the stability, translation effectiveness, and lifespan of the mRNA (68). Since the tail size affects the degradation of mRNA, the incorporation of poly(A) tails is desirable in the production of mRNA vaccines and therapeutics with longer half-life. For example, the addition of approximately 100 nucleotides to the poly(A) tail can result in the production of mRNA with desired prolongation of degradation (69)(17).

Natural mRNA contains ATP, CTP, GTP, and UTP as the four basic nucleotides. Although non-modified mRNA has its own advantages, modified nucleotides are beneficial because they can avoid the recognition of IVT mRNA by the innate immune system, avoiding any undesirable immune responses, and improving the translation efficiency of the mRNA (70). Nucleotides that are modified after post-transcriptional modification of mRNA molecules, such as pseudouridine and 5-methylcytidine, can be utilized in the IVT transcription of the mRNA (71). Andries et al. demonstrated that mRNAs containing the N(1)-methyl-pseudouridine (m1Ψ) modification outperformed the pseudouridine (Ψ)-modified mRNA platform by providing up to approximately 44-fold higher and 13-fold higher reporter gene expression upon transfection into cell lines or mice, respectively (72).

Advantages of mRNA vaccines

The use of mRNA has several beneficial features over other vaccines platforms such as subunit, inactivated and live attenuated virus vaccines. Firstly, mRNA is a non-infectious, non-integrating platform, meaning that there is no potential risk of infection or insertional mutagenesis. Additionally, mRNA is degraded by normal cellular processes, and its in vivo half-life can be regulated through various modification and delivery methods (73)(52)(74). The inherent immunogenicity of the mRNA can be down-modulated to further increase the safety profile (73)(75). Secondly, various modifications can make mRNA more stable and highly translatable. For example, efficient in vivo delivery can be achieved by enclosing mRNA within carrier molecules, allowing rapid uptake and expression in the cytoplasm (74)(76). Anti-vector immunity is also avoided because mRNA is the minimal genetic vector, meaning that mRNA vaccines can be administered repeatedly. Third, mRNA vaccines have the potential for rapid, inexpensive and scalable manufacturing, mainly owing to the high yield of in vivo transcription reactions and the fact that mRNA vaccines are produced in a cell-free environment (77). For example, a 5 L bioreactor can produce a million doses of mRNA vaccine in a single reaction (17). Finally, mRNA vaccines have the provision to code for multiple antigens, thus strengthening the immune response against some resilient pathogens (78).

Classes of mRNA vaccines

mRNA vaccines can be divided into three major categories: conventional mRNA, self-amplifying mRNA (SAM), and circular RNA.

Conventional IVT mRNAs are relatively simple in their architecture and are manufactured at high yield using a cell-free template-directed enzymatic synthesis (14). SAM is designed to include viral-derived molecular machines, such as alphavirus enzymes and conserved sequence elements, allowing it to replicate itself within cells after delivery (79). Typical SAM architecture is constructed from an expression cassette that includes a subgenomic promoter and the target antigen. This cassette is then placed between sequences that encode for alphavirus-derived nonstructural proteins and a poly adenosine tail. SAM vaccines are designed with special sequences that include proteins (eg., nsP1-4) which form a RNA-dependent RNA polymerase complex. This complex recognizes specific design elements in the vaccine and replicates the mRNA within the cell cytoplasm. This replication leads to efficient and prolonged transcription and protein expression. However, SAMs are quite large (6,000-12,000 nucleotides) and their manufacturing is more complex and challenging compared with conventional mRNA vaccines due to low yield, difficulty in purification, and are more prone to autocatalysis and physical degradation (80).

Most currently used SAM vaccines are based on an alphavirus genome, where the genes encoding the RNA replication machinery are intact but the genes encoding the structural proteins are replaced with the antigen of interest (77). SAM format cannot easily incorporate modified nucleosides due to interference between the RNA-dependent RNA polymerase and the nucleoside modified sequences, leading to less mRNA being made in target cells (81). The detection of unmodified nucleosides in SAMs by endosomal and cytoplasmic sensors triggers a strong type I interferon response, which poses a challenge for clinical use. However, vaccine dosage of SAMs is significantly lower – about 100 times less – than conventional mRNA vaccines and therefore could offer disease protection with fewer side effects. Another advantage of SAM vaccines is that they create their own adjuvants in the form of dsRNA structures, replication intermediates and other motifs, which may boost their potency. However, the intrinsic nature of these PAMPs may make it difficult to modulate the inflammatory profile or reactogenicity of SAM vaccines. Additionally, SAM vaccines have stricter size constraints than mRNAs that do not encode replicon genes, and the immunogenicity of the replication proteins may theoretically limit repeated use (77). Early experiments developed by the Imperial College and Acuitas Therapeutics used a SAM administered at extremely low doses (10 ng, prime boost), in mice. The experiment showed potent cell and antibody responses in mice (82) and is now being tested at doses 300-1,000x lower than those used in approved nucleoside modified mRNA vaccines (58).

Circular RNA is a class of non-coding single-stranded RNAs found in eukaryotic cells. Circular RNAs are created through a special process called back-splicing, where the ends of the RNA strand join together instead of lining up end-to-end. Due to their stability, long life, low immunogenicity, and translatability, engineering circular RNA has become a vaccine of high interest. Circular RNAs have been engineered to enable protein expression through the addition of internal ribosomal entry sites and/or the incorporation of specific nucleoside modifications in the 5’ UTR (83). This novel platform has proven effective at producing stable and potent translation in eukaryotic cells (84) due to the RNA being more durable and less susceptible to breakdown. Recent studies have suggested that circular RNAs can evade intracellular immune sensors such as RIG-I without nucleoside modifications (83). Qu et al. showed that circular RNA generates potent, antigen-specific CD4+ and CD8+ cellular and humoral immune responses in mice against SARS-CoV-2 and its emerging variants, providing proof of concept for vaccine applications (85). While circular RNA have great potential to become the next generation of RNA-based vaccine platforms, they are at an early stage in terms of design, synthesis, purification, delivery, and application. In general, five pivotal challenges require attention: designing circular RNAs with low immunogenicity and high antigenic yields, improving the circularization efficiency of linear RNA precursors, adequately purifying to avoid contaminants, establishing suitable delivery systems and enabling disease therapeutic applications. For information on methods to specifically improve circular RNA, the reader is encouraged to read this excellent review paper (86).

Limitations and risks of mRNA vaccines

Despite all of their advantages, mRNA vaccines generally face shortcomings with their duration of antibody response, safety, dependence on ultra-low cold chain transport, and high reactogenicity. mRNA vaccines have been shown to elicit potent germinal center immunogenic reactions and T helper cell induction in preclinical studies against HIV-1, SARS-CoV-2 (87-88), Zika virus, and influenza virus (89)(17). Despite these promising results, the duration of the antibody response varies highly from antigen to antigen.

mRNA is inherently unstable in vivo and antigen expression levels are significantly more transient than observed with DNA vaccination, typically only persisting for several days (90). This is at least in part due to viral RNA sensing molecules such as TLRs 3, 7, and 8 as well as Retinoic Acid-Inducible Gene I (RIG-I). Activation of these innate immune sensors triggers a signaling cascade that enhances RNA clearance and also leads to transfected cell apoptosis, limiting antigen production (91-92). Although these events are proinflammatory, it may reduce a potential adaptive immune response due to the suppression of antigen expression.

Consequently, the mRNA delivered in clinically approved SARS-CoV-2 vaccines has been designed to extend the pharmacological half-life of the spike protein. Some reports suggest that introducing highly modified genetic material to patients elicits an immune response that does not adequately resemble natural infection. In fact, some studies demonstrate that the differences in the elicited immune response may be deleterious. Engineering of mRNA vaccine enriches guanine and cytosine content, increased guanine and cytosine content can result in accumulation of G-quadruplex, which could have effects on post-transcriptional regulation via microRNAs, and that defect in post-transcriptional regulation could favor a greater expression of oncogenes, resulting in cancer and neurological diseases.

These distinct differences in cellular expansion and phenotype can be linked to a downregulation of INF I response, which in turn may make patients more susceptible to other infectious diseases or compromise their anti-cancer surveillance. For example, in a study examining the efficacy of OVA-LNP vaccination to induce anti-cancer response against B16-OVA, so and so demonstrated that likelihood of long term survivors was directly related to the degree of modification of the mRNA uridine. They further demonstrated that increasing uridine modification was demonstrably worse for eliciting INF I responses, clearly indicating that mRNA engineering strategies remain nascent and warrant further investigation in the context of globally accessible vaccine products.

In this regard, in the case of SARS-CoV-2 BNT162b2 mRNA vaccine, unlike the immune response induced by natural SARS-CoV-2 infection, where a robust interferon response is observed, those vaccinated with BNT162b2 mRNA vaccines developed a robust adaptive immune response which was restricted only to memory cells, i.e., an alternative route of immune response that bypassed the IFN mediated pathways

Overall, mRNA vaccines have promising safety profiles as supported by clinical trials and real population data. However, there have been some safety incidents that indicate the need for further optimization of mRNA and its components. For instance, CureVac’s protamine-based rabies vaccine, CV7201, caused adverse effects in 78% of participants (93). In Phase I trials of Moderna’s influenza H10N8 vaccine, adverse events were observed from the 400 μg, causing them to use a lower dose of up to 100 μg (94). Furthermore, mild anaphylactic reactions have been seen in 4.7 per million COVID-19 vaccinations, with 2.5 per million vaccinations with the Moderna vaccine and 2.2 per million with Pfizer–BioNTech vaccine (95).

Current mRNA vaccines require improvement, especially in reducing reactogenicity, despite their success against COVID-19. While the reactogenicity may be deemed acceptable for a few doses during a pandemic, there is a demand for mRNA vaccine formulations with milder reactogenicity for repeated boosting against COVID-19 and broader applications in other infectious diseases. Systemic reactogenicity is more common after the second dose of vaccines than the first dose in humans (96), suggesting the involvement of adaptive immunity in reactogenicity. Moreover, enhanced reactogenicity coincided with higher levels of systemic IFN-γ secretion after the second dose in humans and mice (97). In a mouse study, CD4+ and CD8+ T cells were the primary producers of IFN-γ in the second dose, and the administration of an IFN-γ receptor neutralizing antibody dampened the activation of macrophages, monocytes, and dendritic cells. These findings suggest a close relationship between reactogenicity and immunogenicity. However, an IFN-γ receptor neutralizing antibody minimally affected the efficiency of humoral and cellular immunity induction in mice. Furthermore, the intensity of systemic reactogenicity showed limited correlation with the efficacy of humoral and cellular immunity induction in humans (97-98). These results indicate the potential to separate reactogenicity and immunogenicity. Elucidating innate immune signaling pathways responsible for reactogenicity and immunogenicity could benefit the development of effective vaccines with reduced reactogenicity. mRNA delivery systems that efficiently accumulate in the lymph nodes with minimal systemic leakage might mitigate systemic adverse effects while also maintaining effectiveness (97)(99-100).

Affordable and easy access to vaccines is the greatest challenge in achieving prevalent protection against infectious diseases. This access is made further difficult due to the cold storage requirements and stability of mRNA vaccines. mRNA is highly susceptible to RNase enzymes, which degrade it easily (101). Therefore, it requires extreme sterile conditions of an RNase-free environment in preparation, storage, and administration. All equipment used for these three stages must be sterile. On a more local level, portable and reusable Arktek freezers were key in supplying vaccines to over 400,000 people during the 2014-2016 Ebola virus outbreak in West Africa. However, vaccinating billions of people such as during the COVID-19 pandemic requires thermostable vaccines. Two SARS-CoV-2 vaccine candidates were reported to be thermostable in preclinical studies at room temperature (102-103).

Ongoing work

The development of potent and biodegradable lipids, as well as new formulations for the effective delivery of mRNA in vivo will likely address many of these shortcomings. Researchers at the Novartis Institute used the ionizable lipid DLinDMA as the main component of LNPs for a SAM vaccine targeting the respiratory syncytial virus fusion glycoprotein. It was prepared through ethanol dilution and the particle size ranged from 79 to 121 nm. This vaccine technique with an encapsulation rate of 85%-98% elicited a broad, potent, and protective immune response in mice (104). Hoerr et al. combined IVT-mRNA encoding galactosidase with a polycationic peptide (protamine) and then encapsulated the complex in liposomes. The resulting lipopolysaccharide protected the IVT-mRNA for longer and also demonstrated protein expression and subsequent immune response (105). A final example showcasing the potential of delivery vehicles was Geall et al. showcased that nonviral delivery of a 9-kb SAM RNA vaccine encapsulated within an LNP significantly increased immunogenicity compared to delivery of unformulated RNA. The formulation triggered a broad, strong, and protective immune response, comparable to viral delivery methods but without the drawbacks of viral vectors (104).

Exogenous mRNA is inherently immunostimulatory, as it is recognized by a variety of cell surface, endosomal and cytosolic innate immune receptors. This feature is advantageous for vaccination because it may provide adjuvant activity to drive DC maturation and thus elicit robust T and B cell immune responses. However, innate immune sensing of mRNA has also been associated with the inhibition of antigen expression and may negatively affect the immune response (77). Studies over the past decade have shown that the immunostimulatory profile of mRNA can be shaped. Karikó et al. showed that mRNA containing pseudouridines have a higher translational capacity than modified mRNA when tested in mammalian cells and lysates or when administered intravenously into mice at 0.015–0.15 mg/kg doses. The delivered mRNA and the encoded protein could be detected in the spleen at 1, 4, and 24 hours after the injection, where both products were at significantly higher levels when pseudouridine-containing mRNA was administered. These findings indicate that nucleoside modification is an effective approach to enhance stability and translational capacity of mRNA while diminishing its immunogenicity in vivo (52).

Currently, altering mRNA vaccines pharmacokinetic properties by prolonging the translation of antigenic mRNA has emerged as a promising tool to enhance antibody response (90). The use of modified nucleotides such as 5-methyl cytosine or pseudouridine can limit recognition by RIG-I and enables transfected cells to express antigen for significantly longer timeframes (92)(52). Additionally, incorporation of RNA capping analogues have been reported to reduce innate sensing and increase the duration of antigen production in vivo, leading to enhanced immune responses to the encoded antigen (106). Recent clinical trials from Moderna using intramuscular delivery of nucleoside modified RNA show that mRNA vaccines are safe and capable of inducing effective humoral responses against H10N8 and H7N9 influenza strains in humans. In this trial, virus-neutralizing antibody responses were sustained for over 6 months after immunization (94). In studies of recombinant protein vaccines, the exposure kinetics of antigens and adjuvants drastically impact vaccination outcomes, with sustained exposure leading to enhanced antibody production (107-108). Therefore, modulating the kinetics in mRNA vaccines could improve their efficacy. To achieve this, several reports have attempted to control the release of mRNA nanoparticles from hydrogels. For example, mRNA complexed with a lipid-based commercial reagent was loaded onto a poly (2-hydroxyethyl methacrylate)-based porous scaffold (109). After subcutaneous injection, this scaffold retained mRNA for more than 3 days and enhanced protein expression efficiency compared to bolus mRNA injection.

To improve the stability of mRNA vaccines and thus accessibility of mRNA vaccines, mRNA modification of both the coding and non-coding regions can be effective to improve the stability, immunogenicity, and translation efficiency of mRNAs. For example, replacement of uridine with pseudouridine in the coding region reduced degradation by RNase in a nucleoside-modified approach (110). Both Moderna and Pfizer modified the mRNA encoding viral spike protein by replacing natural residues with two consecutive prolines at amino acid positions K986 and V987 for their COVID-19 vaccines (111). The modification allowed stabilization of the spike on the virus particle in its ‘pre-fusion’ conformation. Modifications of the non-coding regulatory regions also increase resistance against degradation by RNase and exonuclease. Studies have also indicated that nucleoside modifications have increased stability of mRNA vaccines. For example, substitution with m1Ψ, m6A, and s2U in mRNA molecules suppresses the degradation of RNA by RNase L (112). Unmodified mRNA can activate RNA-dependent protein kinase, which is one of four kinases that phosphorylate eIF2, a translation initiation factor, leading to the repression of translation (113-114)(86). Therefore, modifying mRNA is necessary to improve vaccine stability. Besides nucleoside modification, common approaches, such as 5′-cap modification and elongation of poly(A) tail, have been applied to optimize mRNA vaccines (115-116). Locked nucleic acid (LNA) caps have also been investigated, in which the ribose is locked in an C3′-endo conformation by a bridging methylene group between the 2′ oxygen and 4′ carbon. Although LNAs have primarily been used in oligonucleotides, mRNAs capped by an LNA analog have recently been demonstrated to have increased translational efficiency and stability (117). Finally, a novel continuous freeze-drying technique based on spin-freezing is being explored as a storage technique for mRNA vaccines. This method has several advantages compared to classical batch freeze-drying including a much shorter drying time and improved process and product quality controlling. Meulewaeter et al. studied the stability of optimally lyophilized mRNA LNPs at 4 °C, 22 °C, and 37 °C and found that transfection properties of lyophilized mRNA LNPs were maintained during at least 12 weeks. This study is one of the first that demonstrates that optimally lyophilized mRNA LNPs can be safely stored at higher temperatures for months without losing their transfection properties (118).

Having discussed mRNA vaccines successes and limitations, I turn now to a consideration of how to design next generation mRNA candidates. Ideal vaccines would be safe, have perfect efficacy, achieve long-lasting immunity, and be low cost. Furthermore, they would mimic the endogenous immune response as much as possible in both space and time. This requires sustained locoregional delivery of both the antigen (or antigen encoding sequence) and co-stimulatory molecules to drive the desired immune response. In natural immunity, for example when we are naturally exposed to a virus, the antigen and immunostimulatory molecules (DAMPS, PAMPs, etc) are co-presented to the immune system, demonstrating that mounting both cellular immunity (driven largely by interferon responses and cellular cytotoxicity) and humoral immunity requires a tightly orchestrated immune cascade….

One proposal would be to functionalize the liposomal nanoparticles in which mRNA vaccines are currently delivered with co-stimulatory antibodies such as anti-CD28 and anti-OX40. Alternatively, co-delivery with adjuvants could rationally skew the immune response toward either a Th1 or Th2 type.

Moreover, many groups have been exploring the use of biomaterials to improve vaccine responses. Sustained exposure to mRNA LNPs may, in fact, generate longer lasting germinal centers, in turn allowing for more somatic hypermutation and affinity maturation of B-cells, resulting in better quality antibodies that provide more durable and broad protection against highly mutagenic viruses.

Another approach could utilize a biomaterial scaffold loaded with specific chemokines or cytokines to create an immunological niche in which cells interact with the mRNA loaded liposomal nanoparticles. The hypothesis here is that by changing the context in which cells uptake mRNA and express the encoded protein could alter the downstream immunity generated.

Conclusion

Vaccines continue to be critically important technologies for the protection of public health. They have a long and storied history from early biological and immunological theory in the later 1700s to present day. Decades of research into mRNA design and its delivery have made mRNA vaccines an astonishing tool for combating pandemics and infectious diseases. The recent COVID-19 pandemic has especially catalyzed the development of mRNA based therapeutics. It is evident that mRNA technology has the potential for the development of more effective vaccines against persistent and challenging pathogens. mRNA vaccines offer several advantages, including the ability to induce strong and sustained immune responses, reduced risk of insertional mutagenesis, and the potential for rapid adaptation to emerging pathogens. Nevertheless, advancement in mRNA delivery technologies will be required for more effective, safer, and cold-chain-free mRNA vaccines, having the capacity to vaccinate billions of populations across boundaries. The further incorporation of modified nucleosides, advanced purification methods, and innovative delivery systems like lipid nanoparticles have significantly enhanced the efficacy and safety of mRNA vaccines. These advancements not only address some of the inherent limitations of the vaccine platform but also open up new avenues for rapid, scalable, and cost-effective vaccine production.

Looking forward, continued research and development in vaccine technologies are essential to further improve their efficacy, safety, and accessibility. Addressing challenges that will come with developing mRNA vaccine platforms will be crucial in creating next-generation vaccines that are more biomimetic, durable, and potent, ultimately ensuring better preparedness for future infectious disease outbreaks and contributing to global health security.

Abbreviations

mRNA: messenger RNA, IVT: in vitro transcribed, B cells: B-lymphocytes cells, CD8+ T cells: cytotoxic T-lymphocytes, APCs: antigen-presenting cells, MHC-I: class I major histocompatibility complexes, TCR: T cell receptors, MHC-II: class II major histocompatibility complexes, CD4+ T cells: T helper cells, DCs: dendritic cell, PAMPs: pathogen associated molecular patterns, PRRs: pattern recognition receptors, TLRs: Toll-like receptors, Th1: T helper 1 cell, Th2: T helper 2 cell, IFN-γ: Interferon-gamma, MBCs: memory B cells, Tfh: T follicular helper cells, GC: germinal center, LLPCs: long-lived plasma cells, LNPs: lipid nanoparticles, PEG: polyethylene glycol, UTRs: untranslated regions, ORF: open reading frame, m7G: 7-methylguanosine, SAM: self-amplifying mRNA, RIG-I: Retinoic Acid-Inducible Gene I, LNA: Locked nucleic acid

If you truly made it to the end. I can’t thank you enough for the time. If you have any feedback it would be much appreciated. Until next time!

References

1. Saleh A, Qamar S, Tekin A, Singh R, Kashyap R. Vaccine Development Throughout History. Cureus, 13: 2021.

2. Pindyck T, Tate JE, Parashar UD. A decade of experience with rotavirus vaccination in the United States – vaccine uptake, effectiveness, and impact. Expert Rev Vaccines, 17: 593-606, 2018.

3. Sánchez-Sampedro L, Perdiguero B, Mejías-Pérez E, García-Arriaza J, Pilato MD, Esteban M. The Evolution of Poxvirus Vaccines. Viruses, 7: 1726-1803, 2015.

4. Delany I, Rappuoli R, Gregorio ED. Vaccines for the 21st century. EMBO Molecular Medicine, 6: 708-720, 2014.

5. Travieso T, Li J, Mahesh S, Blasi M. The use of viral vectors in vaccine development. Nature Vaccines, 7: 75, 2022.

6. Mohsen Mo, Bachmann MF. Virus-like particle vaccinology, from bench to bedside. Nature Cellular & Molecular Immunology volume, 19: 993-1011, 2022.

7. Suryawanshi YR. An overview of protein-based SARS-CoV-2 vaccines. Vaccine, 41: 6174-6193, 2023.

8. Gupta S, Pellett S. Recent Developments in Vaccine Design: From Live Vaccines to Recombinant Toxin Vaccines. Toxins (Basel), 9: 563, 2023.

9. Cobb M. Who discovered messenger RNA?. Cell Press, 25: R526-R532, 2015.

10. Jirikowski GF, Sanna PP, Maciejewski-Lenoir Dominique, Bloom FE. Reversal of Diabetes Insipidus in Brattleboro Rats: Intrahypothalamic Injection of Vasopressin mRNA. Science, 255: 996-998, 1992.

11. Wolff JA, Malone RW, Williams P, Chong W, Acsadi G, Jani A, Felgner PL. Direct gene transfer into mouse muscle in vivo. Science, 247: 1465-1468, 1990.

12. Hou X, Zaks T, Langer R, Dong Y. Lipid nanoparticles for mRNA delivery. Nature Reviews Materials, 6: 1078-1094, 2021.

13. Embgenbroich M, Burgdorf S. Current Concepts of Antigen Cross-Presentation. Front Immunol, 9: 2018.

14. Ghattas M, Dwivedi G, Lavertu M, Alameh MG. Vaccine Technologies and Platforms for Infectious Diseases: Current Progress, Challenges, and Opportunities. Vaccines (Basel), 9: 2021

15. Pollard AJ, Bijker ME. A guide to vaccinology: from basic principles to new developments. Nature Reviews Immunology, 2: 83-100, 2021.

16. Lazzaro S, Giovani C, Mangiavacchi S, Magini D, Maione D, Baudner B, Geall AJ, Gregorio ED, D’Oro U, Buonsanti C. CD8 T-cell priming upon mRNA vaccination is restricted to bone-marrow-derived antigen-presenting cells and may involve antigen transfer from myocytes. Immunology, 2: 312-326, 2015.

17. Gote V, Bolla PK , Kommineni N, Butreddy A,Nukala PK, Palakurthi SS, Khan W. A Comprehensive Review of mRNA Vaccines. Int J Mol Sci, 24: 2700, 2023.

18. Kim J, Eygeris Y, Gupta M, Sahay G. Self-assembled mRNA vaccines. Adv Drug Deliv Rev, 170: 83-112, 2021.

19. Rijkers GT, Weterings N, Obregon-Henao A, Lepolder M, Dutt TS, van Overveld FJ, Henao-Tamay M. Antigen Presentation of mRNA-Based and Virus-Vectored SARS-CoV-2 Vaccines. Vaccines (Basel), 8: 848, 2021.

20. Amarante-Mendes, GP, Adjemian S, Branco LM, Zanetti LC, Weinlich R, Bortoluci KR. Pattern recognition receptors and the host cell death molecular machinery. Frontiers in Immunology, 9: 2379, 2018.

21. Li D, Wu M. Pattern recognition receptors in health and diseases. Signal Transduction and Targeted Therapy, 6: 291, 2021.

22. Mogensen TH. Pathogen Recognition and Inflammatory Signaling in Innate Immune Defenses. Clin Microbiol Rev, 2: 240-273, 2009.

23. Elena Cano, Luz R; Damaris, H; Lopera, E., Autoimmunity: From Bench to Bedside. El Rosario University Press, Bogota, Colombia, 2013.

24. Chauveau A, Pirgova G, Cheng H, De Martin A, Zhou FY, Wideman S, Rittscher J, Ludewig B, Arnon TL. Visualization of T Cell Migration in the Spleen Reveals a Network of Perivascular Pathways that Guide Entry into T Zones. Immunity, 5: 794-804, 2020.

25. Crotty S. T follicular helper cell biology: A decade of discovery and diseases. Immunity, 5: 1132-1148, 2019

26. Griffiths GM, Berek C, Kaartinen M, Milstein C. Somatic mutation and the maturation of immune response to 2-phenyl oxazolone. Nature, 312: 271-275, 1984.

27. McKean D, Huppi K, Bell M, Staudt L, Gerhard W, Weigert M. Generation of antibody diversity in the immune response of BALB/c mice to influenza virus hemagglutinin. PNAS, 10: 3180-3184, 1984.

28. Akkaya M, Kwak K, Pierce SK. B cell memory: building two walls of protection against pathogens. Nature Reviews Immunology, 20: 229-238, 2020

29. Robbins JB, Schneerson R, Szu SC, Fattom A, Yang Y, Lagergard T, Chu C, Sørensen US. Prevention of invasive bacterial diseases by immunization with polysaccharide-protein conjugates. Current Topics in Microbiology and Immunology, 146: 169-180, 1989.

30. Coffman RL, Sher A, Seder RA. Vaccine Adjuvants: Putting Innate Immunity to Work. Immunity, 33: 492-503, 2010.

31. Z T, C Y, Jiang Y, He X, Wei Y, Yu Y, Tian X. Vaccine adjuvants: mechanisms and platforms. Signal Transduction and Targeted Therapy, 8: 283, 2023.

32. Janeway C. Immunogenicity signals 1,2,3... and 0. Immunology Today, 10: 283-286, 1989

33. Ravi Danielsson R, Eriksson H. Aluminium adjuvants in vaccines – A way to modulate the immune response. Seminars in Cell & Developmental Biology, 115: 3-9, 2021.

34. Koa E, Kang S. Immunology and efficacy of MF59-adjuvanted vaccines. Hum Vaccin Immunother, 12: 3041-3045, 2018.

35. Grigoryan L, Feng Y, Bellusci L, Lai L, Wali B, Ellis M, Yuan M, Arunachalam PS, Hu M, Kowli S, Gupta S, Maysel-Auslender S, Maecker HT, Samaha H, Rouphael N, Wilson IA, Moreno AC, Suthar MS, Khurana S, Pillet S, Charland N, Ward BJ, Pulendran B. AS03 adjuvant enhances the magnitude, persistence, and clonal breadth of memory B cell responses to a plant-based COVID-19 vaccine in humans. Sci Immunol, 9: 2024.

36. Coccia M, Collignon C, Hervé C, Chalon A, Welsby I, Detienne S, et al. Cellular and molecular synergy in AS01-adjuvanted vaccines results in an early IFNγ response promoting vaccine immunogenicity. npj Vaccines, 2: 25, 2017.

37. Xie C, Yao R, Xia X. The advances of adjuvants in mRNA vaccines. npj Vaccines, 8: 162, 2023.

38. Alameh M, Tombácz I, Bettini E, Lederer K, Ndeupen S, Sittplangkoon C, Wilmore JR, Gaudette BT, et al. Lipid nanoparticles enhance the efficacy of mRNA and protein subunit vaccines by inducing robust T follicular helper cell and humoral responses. Immunity, 12: 2877–2892, 2021.

39. Ndeupen S, Qin Z, Jacobsen S, Bouteau A, Estanbouli H, Igyártó BZ. The mRNA-LNP platform's lipid nanoparticle component used in preclinical vaccine studies is highly inflammatory. iScience, 12: 103479, 2021.

40. Heyes J, Palmer L, Bremner K, MacLachlan I. Cationic lipid saturation influences intracellular delivery of encapsulated nucleic acids. Journal of Controlled Release, 107: 276-287, 2005.

41. Malone RW, Felgner PL, Verma IM. Cationic liposome-mediated RNA transfection. PNAS, 16: 6077-6081, 1989.

42. Loug G, Anderluzzi G, Schmidt ST, Woods S, Gallorini S, Brazzoli M, et al. Delivery of self-amplifying mRNA vaccines by cationic lipid nanoparticles: The impact of cationic lipid selection. Journal of Controlled Release, 325: 370-379, 2020

43. Hajj KA, Whitehead KA. Tools for translation: non-viral materials for therapeutic mRNA delivery. Nature Reviews Materials, 2: 17056, 2017.

44. Kulkarni JA, Witzigmann D, Leung J, Tam YYC, Cullis PR. On the role of helper lipids in lipid nanoparticle formulations of siRNA. Nanoscale, 11: 21733-21739, 2019

45. Waggoner LE, Miyasaki KF, Kwon EJ. Analysis of PEG-lipid anchor length on lipid nanoparticle pharmacokinetics and activity in a mouse model of traumatic brain injury. Biomater. Sci, 11: 4238-4253, 2023.

46. Lindgren G, Ols S, Liang F, Thompson EA, Lin A, Hellgren F, Bahl K, John S, Yuzhakov O, Hassett KJ, Brito LA, et al. Induction of Robust B Cell Responses after Influenza mRNA Vaccination Is Accompanied by Circulating Hemagglutinin-Specific ICOS+ PD-1+ CXCR3+ T Follicular Helper Cells. Front Immunol, 8: 1539, 2017.

47. Pardi N, Secreto AJ, Shan X, Debonera F, Glover J, Yi Y, Muramatsu H, Ni H, Mui BL, Tam YK, et al. Administration of nucleoside-modified mRNA encoding broadly neutralizing antibody protects humanized mice from HIV-1 challenge. Nature Communications, 18: 14630, 2017.

48. Pardi N, Tuyishime S, Muramatsu H, Kariko K, Mui BL, Tam YK, Madden TD, Hope MJ, Weissman D. Expression kinetics of nucleoside-modified mRNA delivered in lipid nanoparticles to mice by various routes. J Control Release, 217: 345-351, 2015.

49. Dimitriadis GJ. Translation of rabbit globin mRNA introduced by liposomes into mouse lymphocytes. Nature, 274: 923–924, 1978.

50. Alameh, MG. Weissman, D. Pardi, N., Messenger RNA-Based Vaccines Against Infectious Diseases, in Yu D; Petsch B (eds), mRNA Vaccines. Current Topics in Microbiology and Immunology. vol 440., Springer International Publishing, Switzerland, AG, pp 111–145, 2020.

51. Conry RM, LoBuglio AF, Wright M, Sumerel L, Pike MJ, Johanning F, Benjamin R, Lu D, Curiel DT. Characterization of a messenger RNA polynucleotide vaccine vector. Cancer Res, 7: 1397- 400, 1995.

52. Karikó K, Muramatsu H, Welsh FA, Ludwig J, Kato H, Akira S, Weissman D. Incorporation of Pseudouridine Into mRNA Yields Superior Nonimmunogenic Vector With Increased Translational Capacity and Biological Stability. Molecular Therapy, 16: 1833-1840, 2008.

53. Karikó K, Buckstein M, Ni H, Weissman D. Suppression of RNA recognition by Toll-like receptors: the impact of nucleoside modification and the evolutionary origin of RNA. Immunity, 2; 165-175, 2005.

54. Karikó K, Muramatsu H, Ludwig J, Weissman D. Generating the optimal mRNA for therapy: HPLC purification eliminates immune activation and improves translation of nucleoside-modified, protein-encoding mRNA. Nucleic Acids Research, 39: 142, 2011.

55. Baiersdörfer M, Boros G, Muramatsu H, Mahiny A, Vlatkovic I, Sahin U,Karikó K. A Facile Method for the Removal of dsRNA Contaminant from In Vitro-Transcribed mRNA. Molecular Therapy - Nucleic Acids, 15: 26-35, 2019.

56. Asrani KH, Farelli JD, Stahley MR, Miller RL, Cheng CJ, Subramanian RR, Browna JM. Optimization of mRNA untranslated regions for improved expression of therapeutic mRNA. RNA Biology, 6: 756–762, 2018.

57. Trotman JB, Schoenberg DR. A Recap of RNA Recapping. Wiley Interdiscip Rev RNA, 10: 1504, 2018.

58. Chaudhary N, Weissman D, Whitehead KA. mRNA vaccines for infectious diseases: principles, delivery and clinical translation. Nature Reviews Drug Discovery, 20: 817-838, 2021.

59. Weissman, D. Pardi, N. Muramatsu, H. Karikó, K., HPLC Purification of In Vitro Transcribed Long RNA, In: Rabinovich, P. (eds) Synthetic Messenger RNA and Cell Metabolism Modulation. Human Press, Methods in Molecular Biology, Totowa, NJ, pp 43-54, 2012.

60. Sahin U, Karikó K, Türeci Ö. mRNA-based therapeutics — developing a new class of drugs. Nature Review Drug Discovery, 13: 759-780, 2014.

61. Ramanathan A, Robb GB, Chan S. mRNA capping: biological functions and applications. Nucleic Acids Research, 16: 7511- 7526, 2016.

62. Yang L, Tang L,Zhang M, Liu C. Recent Advances in the Molecular Design and Delivery Technology of mRNA for Vaccination Against Infectious Diseases. Frontiers Immunology, 13: 2022.

63. Li Y, Kiledjian M. Regulation of mRNA decapping. Wiley Interdisciplinary Review RNA, 1: 253-265, 2010.

64. Chatterjee S, Pal JK. Role of 5'- and 3'-untranslated regions of mRNAs in human diseases. Biology of the Cell, 101: 251-262, 2008.

65. Carralot JP, Probst J, Hoerr I, Scheel B, Teufel R, Jung G, Rammensee HG, Pascolo S. Polarization of immunity induced by direct injection of naked sequence-stabilized mRNA vaccines. Cellular and Molecular Life Sciences, 61: 2418–2424, 2004.

66. Babendure JR,1, Babendure JL,Ding J, Tsien RY. Control of mammalian translation by mRNA structure near caps. RNA, 12: 851-861, 2006.

67. Zeng C, Hou X, Yan J, Zhang C, Li W, Zhao W, Du S, Dong Y. Leveraging mRNA Sequences and Nanoparticles to Deliver SARS-CoV-2 Antigens In Vivo. Advanced Materials, 32: 2020

68. Gallie DR. The cap and poly(A) tail function synergistically to regulate mRNA translational efficiency. Genes and Development, 5: 2108-2116, 1991.

69. Godiska R, Mead D, Dhodda V, Wu C, Hochstein R, Karsi A, Usdin K, Entezam A, Ravin N. Linear plasmid vector for cloning of repetitive or unstable sequences in Escherichia coli. Nucleic Acids Research, 38: 88, 2010.

70. Anderson BR, Muramatsu H, Nallagatla SR, Bevilacqua PC, Sansing LH, Weissman D, Karikó K. Incorporation of pseudouridine into mRNA enhances translation by diminishing PKR activation. Nucleic Acids Research, 38: 5884-5892, 2010.

71. Karikó K, Weissman D. Naturally occurring nucleoside modifications suppress the immunostimulatory activity of RNA: implication for therapeutic RNA development. Current Opinion in Drug Discovery and Development, 10: 523-532, 2007.

72. Andries O, Cafferty SMc, De Smedt SC, Weiss R, Sanders NN, Kitada T. N(1)-methylpseudouridine-incorporated mRNA outperforms pseudouridine-incorporated mRNA by providing enhanced protein expression and reduced immunogenicity in mammalian cell lines and mice. Journal of Controlled Release, 217: 337-344, 2015.

73. Thess A, Grund S, Mui BL, Hope MJ, Baumhof P, Fotin-Mleczek M, Schlake T. Sequence-engineered mRNA Without Chemical Nucleoside Modifications Enables an Effective Protein Therapy in Large Animals. Molecular Therapy, 23: 1456-1464, 2015.

74. Guan S, Rosenecker J. Nanotechnologies in delivery of mRNA therapeutics using nonviral vector-based delivery systems. Nature Gene Therapy, 24: 133-143, 2017.

75. Karikó K, Muramatsu H, Ludwig J, Weissman D. Generating the optimal mRNA for therapy: HPLC purification eliminates immune activation and improves translation of nucleoside-modified, protein-encoding mRNA. Nucleic Acids Research, 39: 142, 2011.

76. Kauffman KJ, Webber MJ, Anderson DG. Materials for non-viral intracellular delivery of messenger RNA therapeutics. Journal of Controlled Release, 240: 227-234, 2016.

77. Pardi N, Hogan MJ, Porter FW, Weissman D. mRNA vaccines — a new era in vaccinology. Nature Reviews Drug Discovery, 17: 261-279, 2018.

78. Freyn AW,da Silva JR, Rosado VC, Bliss CM, Pine M, Mui BL, Tam YK, Madden TD, Ferreira LCS, Weissman D, et al. A Multi-Targeting, Nucleoside-Modified mRNA Influenza Virus Vaccine Provides Broad Protection in Mice. Molecular Therapy, 28: 1569-1584, 2020.

79. Magini D, Giovani C, Mangiavacchi S, Maccari S, Cecchi R, Ulmer JB, De Gregorio E, Geall AJ, Brazzoli M, Bertholet S. Self-Amplifying mRNA Vaccines Expressing Multiple Conserved Influenza Antigens Confer Protection against Homologous and Heterosubtypic Viral Challenge. PLOS One, 11: 2016.

80. Bloom K, van den Berg F, Arbuthnot P. Self-amplifying RNA vaccines for infectious diseases. Nature Gene Therapy, 28: 117-120, 2021.

81. Maruggi G, Zhang C, Li J, Ulmer JB, Yu D. mRNA as a Transformative Technology for Vaccine Development to Control Infectious Diseases. Molecular Therapy, 10. 757-772, 2019.

82. McKay PF, Hu K, Blakney AK, Samnuan K, Brown JC, Penn R, Zhou J, Bouton CR, Rogers P, Polra K, Lin PJC, Barbosa C, Tam YK, Barclay WS, Shattock RJ. Self-amplifying RNA SARS-CoV-2 lipid nanoparticle vaccine candidate induces high neutralizing antibody titers in mice. Nature Communications, 11: 2020.

83. Kristensen LS, Andersen MS, Stagsted LVW, Ebbesen KK, Hansen TB, Kjems J. The biogenesis, biology and characterization of circular RNAs. Nature Reviews Genetics, 20: 675-691, 2019.

84. Zhou W, Cai Z, Liu J, Wang D, Ju H, Xu R. Circular RNA: metabolism, functions and interactions with proteins. Molecular Cancer, 19: 172, 2020.

85. Qu S, Yang X, Li X, Wang J, Gao Y, Shang R, Sun W, Dou K, Li H. Circular RNA: A new star of noncoding RNAs. Cancer Letters, 365: 141-148, 2015.

86. Niu D, Wu Y, Lian J. Circular RNA vaccine in disease prevention and treatment. Nature Signal Transduction and Targeted Therapy, 8: 2023.

87. Corbett KS, Flynn B, Foulds KE, Francica JR, Boyoglu-Barnum S, Werner AP, Flach B, O’Connell S, Bock KW, et al. Evaluation of the mRNA-1273 Vaccine against SARS-CoV-2 in Nonhuman Primates. The New England Journal of Medicine, 383: 1544-1555, 2020.

88. Lederer K,1, Castaño D, Atria DG, Oguin TH, Wang S, Manzoni TB , Muramatsu H, Hogan MJ, Amanat F, Cherubin P, et al. SARS-CoV-2 mRNA Vaccines Foster Potent Antigen-Specific Germinal Center Responses Associated with Neutralizing Antibody Generation. Immunity, 53: 1281-1295, 2020.

89. Pardi N, Hogan MJ, Naradikian MS, Parkhouse K, Cain DW, Jones L, Moody MA, Verkerke HP, Myles A, Willis E, et al. Nucleoside-modified mRNA vaccines induce potent T follicular helper and germinal center B cell responses. Journal of Experimental Medicine, 215: 1571–1588, 2018.

90. Irvine DJ, Aung A, Silva M. Controlling timing and location in vaccines. Advanced Drug Delivery Reviews, 158: 91–115, 2020.

91. Besch R, Poeck H, Hohenauer T, Senft D, Häcker G, Berking C, Hornung V, Endres S, Ruzicka T, Rothenfusser S, Hartmann G. Proapoptotic signaling induced by RIG-I and MDA-5 results in type I interferon–independent apoptosis in human melanoma cells. The Journal of Clinical Investigation, 119: 2399–2411, 2009.

92. Anderson BR, Muramatsu H, Jha BK,3 Silverman RH, Weissman D, Karikó K. Nucleoside modifications in RNA limit activation of 2′-5′-oligoadenylate synthetase and increase resistance to cleavage by RNase L. Nucleic Acids Research, 39: 9329- 9338, 2011.

93. Alberer M, Gnad-Vogt U, Hong HS, Mehr KT, Backert L, Finak G, Gottardo R, Bica MA, Garofano A, Koch SD, Fotin-Mleczek M, Hoerr I, Clemens R, Sonnenburg F. Safety and immunogenicity of a mRNA rabies vaccine in healthy adults: an open-label, non-randomised, prospective, first-in-human phase 1 clinical trial. The Lancet, 390: 1511-1520, 2017.

94. Feldman RA, Fuhr R, Smolenov I, Ribeiro AM, Panther L, Watson M, Senn JJ, Smith M, Almarsson A, et al. mRNA vaccines against H10N8 and H7N9 influenza viruses of pandemic potential are immunogenic and well tolerated in healthy adults in phase 1 randomized clinical trials. Vaccine, 37: 3326-3334, 2019.

95. Shimabukuro TT, Cole M, Su JR. Reports of Anaphylaxis After Receipt of mRNA COVID-19 Vaccines in the US—December 14, 2020-January 18, 2021. JAMA Insights, 325: 1101-1102, 2021.

96. Baden LR, Sahly HM, Essink B, Kotloff K, Frey S, Novak R, Diemert D, Spector SA, Rouphael N, Creech CB, McGettigan J, Khetan S, et al. Efficacy and Safety of the mRNA-1273 SARS-CoV-2 Vaccine. The New England Journal of Medicine, 384: 403-416, 2021.

97. Mochida Y, Uchida S. mRNA vaccine designs for optimal adjuvanticity and delivery. RNA Biology, 21: 1-27, 2024.

98. Debes AK, Xiao S, Colantuoni E, Egbert ER, Caturegli P, Gadala A, Milstone AM. Association of Vaccine Type and Prior SARS-CoV-2 Infection With Symptoms and Antibody Measurements Following Vaccination Among Health Care Workers. JAMA Internal Medicine, 181: 1660-1662, 2021.

99. Yang R, Deng Y, Huang B, Huang L, Lin A, Li Y, Wang W, Liu J, Lu S, Zhan Z, Wang Y, Ruhan A, Wen W, Niu P, et al. A core-shell structured COVID-19 mRNA vaccine with favorable biodistribution pattern and promising immunity. Nature Signal Transduction and Targeted Therapy, 6: 2021.

100. Han X, Alameh M, Butowska K, Knox JJ, Lundgreen K, Ghattas M, Gong N, Xue L, Xu Y, Lavertu M, et al. Adjuvant lipidoid-substituted lipid nanoparticles augment the immunogenicity of SARS-CoV-2 mRNA vaccines. Nature Nanotechnology, 18: 1105–1114, 2023.

101. Pascolo S. Synthetic Messenger RNA-Based Vaccines: From Scorn to Hype. Viruses, 13: 270, 2021.

102. Zhang N, Li X, Deng Y, Zhao H, Huang Y, Yang G, Huang W, Gao P, Zhou C, et al. A Thermostable mRNA Vaccine against COVID-19. Cell, 182: 1271–1283, 2020.

103. Qi Y, Fox CB. Development of thermostable vaccine adjuvants. Expert Review of Vaccines, 20: 497-517, 2021.

104. Geall AJ, Verma A, Otten GR, Shaw CA, Hekele A, Banerjee K, Cu Y, Beard CW, Brito LA, Krucker T, O'Hagan DT, Singh M, Mason PW, et al. Nonviral delivery of self-amplifying RNA vaccines. PNAS, 109. 14604–14609, 2012.

105. Hoerr I, Obst R, Rammensee HG, Jung G. In vivo application of RNA leads to induction of specific cytotoxic T lymphocytes and antibodies. European Journal of Immunology, 30: 1-7, 2000.

106. Kuhn AN, Diken M, Kreiter S, Selmi A, Kowalska J, Jemielity J, Darzynkiewicz E, Huber C, Türec Oi, Sahin U. Phosphorothioate cap analogs increase stability and translational efficiency of RNA vaccines in immature dendritic cells and induce superior immune responses in vivo. Nature Gene Therapy, 17: 961-971, 2010.

107. Tam HH, Melo MB, Kang M, Pelet JM, Ruda VM, Foley MH, Hu JK, Kumari S, Crampton J, Baldeon AD, Sanders RW, et al. Sustained antigen availability during germinal center initiation enhances antibody responses to vaccination. PNAS, 113: 6639-6648, 2016.

108. Cirelli KM, Carnathan DG, Nogal B, Martin JT, Rodriguez OL, Upadhyay AA, Enemuo CA, Gebru EH, Choe Y, Viviano F, Nakao C, et al. Slow Delivery Immunization Enhances HIV Neutralizing Antibody and Germinal Center Responses via Modulation of Immunodominance. Cell, 117: 1153-1171, 2019.

109. Chen R, Zhang H, Yan J, Bryers JD. Scaffold-mediated delivery for non-viral mRNA vaccines. Nature Gene Therapy, 25: 556-567, 2018.

110. Rein Verbeke R, Lentacker I, De Smedt SC, Dewitte H. Three decades of messenger RNA vaccine development. Nano today, 28: 2019.

111. Uddin MN, Roni MA. Challenges of Storage and Stability of mRNA-Based COVID-19 Vaccines. Vaccines (Basel), 9: 1033, 2021.

112. Anderson BR, Muramatsu H, Jha BK, Silverman RH, Weissman D, Karikó K. Nucleoside modifications in RNA limit activation of 2′-5′-oligoadenylate synthetase and increase resistance to cleavage by RNase L. Nucleic Acids Research, 39: 9329–9338, 2011.

113. Sonenberg N, Hinnebusch AG. Regulation of Translation Initiation in Eukaryotes: Mechanisms and Biological Targets. Cell, 136: 731-745, 2009.

114. Anderson BR, Muramatsu H, Nallagatla SR, Bevilacqua PC, Sansing LH, Weissman D, Karikó K. Incorporation of pseudouridine into mRNA enhances translation by diminishing PKR activation. Nucleic Acids Research, 38: 5884- 5892, 2010.

115. Kim SC, Sekhon SS, Shin W, Ahn G, Cho B, Ahn J, Kim Y. Modifications of mRNA vaccine structural elements for improving mRNA stability and translation efficiency. Molecular and Cellular Toxicology, 18; 1-8, 2021.

116. Warminski M, Mamot A, Depaix A, Kowalska J, Jemielity J. Chemical Modifications of mRNA Ends for Therapeutic Applications. Accounts of Chemical Research, 56: 2814-2826, 2023.

117. Senthilvelan A, Vonderfecht T, Shanmugasundaram M, Pal I, Potter J, Kore AR. Trinucleotide Cap Analogue Bearing a Locked Nucleic Acid Moiety: Synthesis, mRNA Modification, and Translation for Therapeutic Applications. Organic Letters, 23: 4133-4136, 2021.

118. Meulewaeter S, Nuytten G,MHY, De Smedt SC, Cullis PR, De Beer T, Lentacker I, Verbeke R. Continuous freeze-drying of messenger RNA lipid nanoparticles enables storage at higher temperatures. Journal of Controlled Release, 357: 149-160, 2023.

Understanding protein structure is integral to giving us clues to a proteins function. Yet, the question of what a protein does inside a living cell is not as simple as determining the structure. Imagine an analysis of an uncharacterized protein’s structure and amino acid sequence suggests it may act as a protein kinase. Simply knowing that a protein adds phosphates does not reveal how it functions in a living organism. Additional information is required to understand the context in which the biochemical activity is used. What are its protein targets? In which tissues is it active? Which biological pathways does it influence? What role does it have in the growth or development of the organism?

This article will discuss methods to determine protein structure, which is certainly still important and no easy feat, as well as its function. A lot of the information I got in this article was from a conversation I had with Professor Lars Konermann at The University of Western Ontario, whos lab focused on protein folding, protein interactions with drugs and other molecules, and the mechanisms of protein aggregation (clumps of improperly folded or unfolded proteins that can form both inside and outside of cells). A lot of his work involves using experimental techniques to determine protein structure. Finally, another cool branch of his lab explores the mechanisms of electrospray ionization: the transition of proteins and other biological molecules from solution into the gas phase. We will discuss why that is important later.

Starting with the amino acid sequence of a protein, one can often predict which secondary structural elements, the two main ones being alpha helices and beta sheets, will be present in the protein. Yet, it is still a challenge to reliably deduce the three-dimensional folded structure of a protein from its amino acid sequence. The main techniques for a long time has been experimental methods such as X-ray crystallography and NMR spectroscopy, yet computational methods, particularly deep learning-based approaches like AlphaFold2, AlphaFold3 and RoseTTAFold (an extra paper on RoseTTAFold), have recently achieved remarkable success in predicting structures from sequences and much more — notably protein interactions, the structure of assemblies of proteins, and even designing molecules from scratch.

Experimental methods: x-ray crystallography, NMR spectroscopy, and cryo electrode microscopy.

While these computational methods are now revolutionary, they would not be possible without the initial, hard work from experimental methods such as X-ray crystallography. In short, to use this technique, the crystallographer obtains protein crystals, records the diffraction pattern formed by x-rays passed through the crystals, and then interprets the data using a computer. The result is a atomic-resolution model of a protein. Lets examine this in more detail

Proteins must be first crystallized in order to collect data on them. Protein crystallization involves transitioning proteins from a dissolved state in a solution into a solid, highly ordered, repeating structure. In a bit more detail, the concentrated protein solution is subjected to a wide variety of crystallization conditions: temperature, pressure, solvent type, solute concentration, cooling rate. Many different conditions are tried in parallel because we don’t know beforehand which conditions are needed for obtaining crystals for a given protein.

Proteins have hydrophilic and hydrophobic side chains. Water soluble proteins fold in a way that the hydrophobic ones (the water hiding ones) are hidden inside. The hydrophilic chains are on the outside which makes the protein pretty stable and easy to work with. They even crystalize quite well. Even if some proteins are soluble, such as membrane proteins, it can still be very hard to crystallize them because their structure is dependent on the membrane that they are attached to. However, if you want to look at a protein, you want to have just the protein in the sample, essentially purifying it. This conflict means simulating membrane proteins or just membrane themselves is an active area. Here is one particularly good example.

The Nobel Prize in 1988, was the first membrane protein to be crystallized ever. Ever since there has been slow and steady progress. Even computational models struggle to accurately model membrane proteins because they don’t understand that the protein is attached to the membrane and they also rely on evolutionary data, which is limited for membrane proteins. There is a pretty cool solution outlining crystallizing membrane proteins using lipidic mesophases. Here is another good overview of membrane protein solutions if you are interested.

X-rays, like light, are a form of electromagnetic radiation, but they have a much shorter wavelength, typically around 0.1 nanometers — the diameter of a hydrogen atom. If a narrow parallel beam of x-rays is directed at a sample of a crystal, most of the x-rays pass straight through it. A small fraction, however, interacts with the crystal and experience a change in some of there properties: position, amplitude, and phase. We call these changes “diffraction,” and all the diffracted x-rays are collected in a diffraction pattern. The crystal is then rotated sightly and a new diffraction pattern is obtained. This process is repeated through 360 degrees along one axis (typically rotations through a smaller angle on another axis are also recorded to avoid blind spots) until the instrument has recorded a diffraction pattern for each position. This makes sense, if you saw my hand at all of its orientations, you would confidentially be able to produce a 3D structure of it.

These changes by the diffracted x-rays obey a mathematical concept called the Fourier transform. To get a model of our protein structure, we feed back the diffracted x-ray data into the Fourier transform, and reciprocate it (raise it to the power of minus one).

To reciprocate the Fourier transform and deduce a a three-dimensional map of all the electron densities (where electrons are) in crystal — or an electron density map — we require the following information:

1. The relative position of the diffracted waves.

2. The relative amplitudes of the diffracted waves.

3. The relative phases of the diffracted waves.

We can directly measure the first two using an X-ray detector. All the diffracted X-rays hit the detector with a definite position and produce a signal proportional to their amplitude. But we lose all the phase information. Nor can we measure it directly or indirectly.

This is called the phase problem.

The two most common ways of overcoming the phase problem in protein crystallography are Molecular Replacement and SAD phasing.

Molecular Replacement

We can take the phases from another structure that resembles the one we are trying to solve. Combine them with the symmetry of the new crystal, and do the inverse Fourier transform. With some luck, we get an electron density map with interpretable features.

SAD (Single-wavelength Anomalous Dispersion) Phasing

It exploits two properties of diffraction patterns:

1. Every single atom contributes to every bit of the diffraction pattern.

2. Atoms with lots of electrons (”heavy atoms”) make the greatest contribution to the diffraction pattern.

These” heavy atoms” cause small, measurable differences in the diffraction pattern that depend on the phase, so we can deduce the missing phase information.

Once we have the electron density map. Since the electrons orbit atomic nuclei or exist between them in bonds, the electron density map is, roughly a ball-and-stick model of the protein structure.

Interpreting this map—translating its contours into a three-dimensional structure—is a complicated procedure that requires knowledge of the protein’s amino acid sequence of. Largely by trial and error, the sequence and the electron-density map are correlated by a computer to give the best possible fit.

What Information Do We Get from Protein Crystallography?

In a successful case, you get an atom-by-atom model of the protein structure and a set of statistics that tell you roughly how accurate it is. Depending on the quality of the data, you may get some bonus information, such as the water molecules that form part of the quaternary structure. Or you may be able to tell if a disulfide bond has formed or if an amino acid is modified.

You’ll usually be able to see if the protein is bound to a ligand, but accurately building the ligand into the model might be difficult. The granularity and information-richness of a structure largely depend on the resolution of the diffraction data.

Despite this great data, these static structures don’t tell you anything about folding or dynamics. Normally, proteins are not static in the body because they are constantly being pushed around by water molecules in our body. Knowing dynamics and folding is important because we need to know how it would interact with different molecules and how it will look like in the body when molecules are colliding and the structure is being slightly altered.

Nuclear Magnetic Resonance (NMR) Spectroscopy

NMR spectroscopy has been widely used for many years to analyze the structure of small molecules, and is increasingly applied to the study of small proteins. Unlike x-ray crystallography, NMR does not depend on having a crystalline sample; it simply requires a small volume of concentrated protein solution that is placed in a strong magnetic field. Thus, NMR is especially useful when a protein of interest has resisted attempts at crystallization, a common problem for many membrane proteins. Because NMR studies are performed in solution, this method also offers a convenient means of monitoring changes in protein structure, for example during protein folding or when a substrate binds to the protein. NMR is also used widely to investigate molecules other than proteins and is valuable, for example, as a method to determine the three-dimensional structures of RNA molecules and the complex carbohydrate side chains of glycoproteins.

NMR exploits the magnetic properties (particularly the property spin) of certain atomic nuclei, most often hydrogen-1 but also carbon-13 and nitrogen-15, to figure out how atoms are arranged in a molecule. In response to pulses of electromagnetic radiation, the spin of the atoms, which is normally aligned along the strong magnetic field, can be changed to a misaligned excited state.

When the excited nuclei return to their aligned state, they emit radiation, which can be measured and displayed as a spectrum. The nature of the emitted radiation depends on the environment of each hydrogen nucleus, such as the atom type (backbone vs side chain), neighboring atoms and bonds ( if one nucleus is excited, it influences the absorption and emission of radiation by other nuclei that lie close to it), and the secondary structure it is apart of (α-helix, β-sheet, loops). Because chemical shifts are sensitive to structure, they act like a "fingerprint" for each atom. If we know which shifts belong to which atoms, and we measure how spins interact (through bonds or through space), we can piece together:

• Connectivity (which atoms are bonded)

• Distances (which atoms are close in space)

• Angles (bond angles from coupling constants)

Combining all this and the knowledge of the amino acid sequence makes it possible to produce the three-dimensional structure of the protein.

Cryo electrode microscopy (cryo-EM)

Cryo-EM combines the power of electron microscopes to see tiny things and the power of cryogenic temperatures to preserve their natural shapes and structures. (Think freezing, freezing cold, around −183 °C or nearly –300 °F: more than twice as cold as the coldest surface temperature ever recorded on Earth). Arguably the most exciting entry point to structural biology in recent years, today’s cryo-EM is known for the remarkable and ever-increasing precision with which it “solves” molecular structures, earning cryo-EM’s early pioneers the 2017 Nobel Prize in Chemistry.

Cryo-EM can capture flexible, wiggly proteins not suited to x-ray crystallography, and it can illuminate big molecules like intact viruses, giving it an advantage over NMR, which is limited to small molecules.

In summary, cryo-EM freezes many copies of a delicate sample into a glassy state, hits them with an electron beam to create numerous 2D shadows, and then, using some sort of software, groups these 2D shadows to create a 3D model of the sample.

Quantum mechanics tells us that fundamental bits of matter like the electron are both particles and waves. Cryo-EM takes advantage of the fact that the electron has a very tiny wavelength – much shorter than wavelengths of light – so it can make clear images of equally tiny things. However, an intense electron beam can damage or destroy a delicate sample. Thus, the samples are frozen to reduce the radiation damage caused by the electron beams. The freezing method also must be incredibly specific: doesn’t form ice crystals, which happens when water freezes slowly and would damage the delicate molecular structures, and that’s fast enough to capture that natural state.

Cryo-EM gets around both of those problems by freezing samples into a glass-like state.

First, researchers extract the specimens they want to study – lets say a protein molecules in this case — and suspend them in water, where they float freely. Next, they deposit a drop of water containing thousands of copies of the protein across a grid of tiny holes in a carbon net.

A robotic arm plunges the grid into liquid ethane that’s been cooled in a bath of liquid nitrogen, and the water and its cargo of proteins instantly freeze into a glassy state.

When the sample is flash frozen, the proteins trapped in the ice are ideally arranged in random orientations, so that we’ll be able to image their structures from all angles. Placed into a transmission electron microscope and shot with an electron beam, these random orientations appear as 2D shadows on a detector, producing many thousands of images of different views of the particles.

Because the intensity of the electron beam is set as low as possible to limit the inevitable structural damage to the sample, all of the images we have now collected appear grainy. This means we have a considerable amount of “noisy” images — images with unwanted background “noise” relative to the meaningful “signal” of the image.

To improve this “signal-to-noise ratio,” we pick out the best 2D images of our proteins before computer software groups them together by orientation (up, down, sideways, and so on). We can now build a representative image — an average of all the 2D images at said orientation (a 2D class average) — for every orientation that has a much improved signal-to-noise ratio.

This array of 2D images is combined again into a 3D reconstruction of the molecule. Researchers can move and rotate this reconstruction in a computer to look at the protein from all sides. If you want to learn more about cryo-em, check out these two cool papers: 1, 2.

Cryogenic electron tomography (cryo-ET)

The cryo-EM technolgy is of course evolving, and one cool example is a development of a form of cryo-EM called called cryogenic electron tomography. Cryo-ET is the only structural biology method available today that can image everything from molecules to cells to tissues to small organisms in their near-native context. Whereas cryo-EM takes many frames of many particles on the grid, cryo-ET takes many frames of the same sample as it is tilted under the microscope to capture different parts of its structure. Computational steps align and merge these images, forming a tomogram, or a 3D picture that can illuminate how molecules carry out their work inside the cell.

Yet, solving structures using experimental methods requires extremely specialized training, a high degree of skill, and a lot of luck. The cost of solving a new, unique structure is on the order of $100,000. Given the difficulty of solving an experimental structure, and considering the rate at which new protein sequences are discovered, it has become clear that with today’s technology, we will not solve structures for all the new proteins being identified and sequenced.

Finding alternative ways to predict a protein structure becomes more and more important as this gap

Sequence Similarity Can Provide Clues About Protein Function

Because amino acid sequence determines protein structure and structure dictates biochemical function, proteins that share a similar amino acid sequence usually perform similar biochemical functions, even when they are found in distantly related organisms. Since there is ample data on protein and nucleic acid sequences, the function of a gene—and its encoded protein—can often be predicted by simply comparing its sequence with those of previously characterized genes. At present, determining what a newly discovered protein does therefore usually begins with a search for previously identified proteins that are similar in their amino acid sequences. This is in fact one of the first steps AlphaFold does. It compares and aligns multiple proteins, DNA, or RNA that have similar sequences with the target protein’s — the one with the 3D structure we want to predict — sequence through a process called MSA (multiple sequence alignment). Predictions that emerge from sequence analysis are often only a tool to direct further experimental investigations. It is definitely not the determining step for a proteins function.

Fusion proteins to analyze protein function and track proteins in living cells

The location of a protein within the cell often suggests something about its function. Proteins that travel from the cytoplasm to the nucleus when a cell is exposed to a growth factor (literally just something that can stimulate a cell to grow or proliferate), for example, may have a role in regulating gene expression — the process where a gene's instructions are used to create a functional product, usually a protein — in response to that factor.

A protein often contains short amino acid sequences that determine its location in a cell. Most nuclear proteins, for example, contain one or more specific short sequences of amino acids that serve as signals for their import into the nucleus after their synthesis in the cytosol. These special regions of the protein can be identified by fusing them to an easily detectable protein that lacks such regions and then following the behavior of this protein in a cell.

Another common strategy used both to follow proteins in cells and to purify them rapidly is epitope tagging. In this case, a fusion protein is produced that contains the entire protein being analyzed plus a short peptide of 8 to 12 amino acids (an “epitope”) that can be recognized by a commercially available antibody. The fusion protein can therefore be specifically detected, even in the presence of a large excess of the normal protein, using the anti-epitope antibody and a labeled secondary antibody that can be monitored by light or electron microscopy.

Large numbers of proteins are also being tracked in living cells by using a fluorescent marker called green fluorescent protein (GFP). Tagging proteins with GFP is as simple as attaching the gene for GFP to one end of the gene that encodes for a protein of interest. In most cases, the resulting GFP fusion protein behaves in the same way as the original protein, and its movement can be monitored by following its fluorescence inside the cell by fluorescence microscopy.

GFP and its color variants can be used to see protein-protein interactions. The method works by attaching a different glowing tag to each protein. These tags are chosen so that the light given off by the first tag can be absorbed by the second one. If the proteins come very close together—within a few nanometers—the energy from the first tag transfers to the second. This process is called fluorescence resonance energy transfer (FRET). Scientists detect this by shining light on the first tag and checking if the second one glows. Using different versions of GFP, researchers can monitor the interactions of any two protein molecules inside a living cell.

Protein affinity chromatography and co-immunoprecipitation allow the identification of associated proteins

Because most proteins in the cell function as part of a complex with other proteins, an important way to begin to characterize their biological roles is to identify their binding partners. If an uncharacterized protein binds to a protein whose role in the cell is understood, its function is likely to be related. For example, if a protein is found to be part of the proteasome complex, it is likely to be involved somehow in degrading damaged or misfolded proteins.

Protein affinity chromatography is one method that can be used to isolate and identify proteins that interact physically. To capture interacting proteins, a target protein is attached to polymer beads that are packed into a column. When a mixture of cellular proteins is passed through, only the ones that naturally bind to the target protein will stick to the beads; the rest wash away. Later, the bound proteins can be released (eluted) from the beads and identified using tools like mass spectrometry.

Perhaps the simplest method for identifying proteins that bind to one another tightly is co-immunoprecipitation. In this approach, an antibody is used to grab a specific protein. The antibody is attached to a solid support so it can pull the protein—and anything stuck to it—out of a solution. If another protein is tightly bound to the target protein, it will come along too. This technique helps researchers identify groups of proteins that work together inside cells, even if their interactions are brief, such as when cells respond to signals.

Co-immunoprecipitation techniques require having a highly specific antibody against a known cellular protein target, which is not always available. One way to overcome this requirement is to use recombinant DNA techniques to add an epitope tag or to fuse the target protein to a well-characterized marker protein, such as GST (glutathione S-transferase). Because antibodies for these tags and markers are easy to get, researchers can use them to pull out the tagged protein along with any partners it’s bound to. In the case of GST, you may not even need antibodies—GST naturally sticks to glutathione, so the fusion protein and its partners can be captured using beads coated with glutathione.

Methods such as co-immunoprecipitation and affinity chromatography allow the physical isolation of interacting proteins. A successful isolation yields a protein whose identity must then be ascertained by mass spectrometry, and whose gene must be retrieved and cloned before further studies characterizing its activity—or the nature of the protein-protein interaction—can be performed.

Other techniques allow the simultaneous isolation of interacting proteins along with the genes that encode them. Two-hybrid screening uses a reporter gene to detect the physical interaction of a pair of proteins inside yeast cells. The system is designed so that when two proteins interact, they bring together two separate halves of a gene activator. This activator switches on a “reporter gene,” signaling that the two proteins have bound to each other.

This method works because gene activator proteins are modular—they usually have one part that binds DNA and another part that turns on transcription. Scientists take advantage of this by attaching the DNA-binding half to a “bait” protein (the protein they want to study). This bait protein then anchors at the control region of a reporter gene in the yeast nucleus.

To search for partners, scientists create a library of potential “prey” proteins by linking random DNA fragments to the activation half of the activator protein. Each yeast cell is given a different prey construct. If the prey protein binds to the bait, the DNA-binding half and activation half are brought together, turning on the reporter gene. The yeast cells that light up this reporter are then isolated, and researchers can sequence the DNA to identify which prey protein was interacting with the bait.

If this process sounded confusing imagine you’re trying to figure out which pairs of puzzle pieces fit together. But instead of looking at the shapes, you set up a lamp that only turns on when the right two pieces snap together.

• The bait protein is like the first puzzle piece you already know and place into the lamp system.

• The prey proteins are all the other puzzle pieces you want to test.

• If a prey piece clicks perfectly with the bait, the lamp lights up — just like the reporter gene switching on.

By seeing which puzzle piece made the lamp turn on, you know which protein interacts with your bait, and you can then look up the “label” (the gene) for that puzzle piece.

Two-hybrid screening is relatively simple to use in the laboratory. Although the protein-protein interactions occur in the nucleus of yeast cels, proteins from every part of the cell and from any organism can be studied in this way.

A variation called the reverse two-hybrid system is used to find mutations or chemicals that block two proteins from binding. Instead of using a reporter gene, this system uses a gene that actually kills the yeast cell if the bait and prey proteins bind. That way, only cells where the proteins fail to interact — because an engineered mutation or a test compound prevents them from doing so — will survive. Eliminating a particular molecular interaction can reveal something about the role of the participating proteins in the cell. In addition, compounds that selectively interrupt protein interactions can be medically useful: a drug that prevents a virus from binding to its receptor protein on human cells could help people to avoid infections, for example.

Another method for finding protein–protein interactions uses viruses that infect the bacterium E. coli (called bacteriophages, or “phages”). In this approach, the DNA encodingor the protein of interest—or a smaller peptide fragment of this protein—is fused with a gene encoding one of the proteins that forms the viral coat. When the virus infects E. coli it replicates, producing new viral particles that display the hybrid protein on their outer surface. These phages, carrying the protein of interest on their coats, can then be used to search through a large collection of proteins to identify binding partners.

The most powerful use of phage display is to screen huge collections of proteins or peptides to see which ones bind to a chosen target. To do this, scientists first create a library of phages, each carrying a different fusion protein—similar to the prey library used in the two-hybrid system. This phage library is then tested against a purified target protein. For example, the phages can be passed through a column that contains the target protein fixed in place. Any phages displaying a protein or peptide that sticks tightly to the target will stay behind in the column, while the rest wash away. These bound phages can then be released (eluted) by adding extra target protein.

The selected phages, which carry the DNA encoding the interacting protein or peptide, are collected and allowed to multiply inside E. coli. Researchers then recover the DNA from these phages and determine its nucleotide sequence to identify the binding protein or peptide. A similar version of this method has been used to find peptides that specifically attach to the inner lining of blood vessels in human tumors. These peptides are now being tested as a way to deliver anti-cancer drugs directly to tumors.

DNA footprinting reveals the sites where proteins bind on a DNA molecule

So far we have concentrated on examining protein-protein interactions. But some proteins act by binding to DNA. Most of these proteins, by binding to regulatory DNA sequences usually located outside the coding regions (regions which encode for RNA) of a gene, have a central role in determining which genes are active in a particular cell.

In analyzing how such a protein functions, it is important to identify the specific nucleotide sequences to which it binds. A method used for this purpose is called DNA footprinting. First, a pure DNA fragment is prepared and labeled at one end with radioactive phosphorus (32P). This DNA is then cut randomly at single sites using either a nuclease enzyme or a chemical that makes single-stranded breaks. After heating (denaturing) the DNA to separate its two strands, the labeled strand fragments are separated on a gel and visualized by autoradiography.

The key step is comparing two samples: one with a DNA-binding protein present and one without. If the protein is bound, it shields the nucleotides at its binding site and prevents those bonds from being cut. As a result, the DNA fragments that would normally end within the binding site are missing from the gel. This absence shows up as a clear gap in the banding pattern, called a “footprint.” Similar approaches can also be applied to map where proteins bind on RNA.

Computational Methods

Computational methods, while not perfect, have been revolutionary in quickening our pace of determining protein structures and uncovering use cases for proteins. While they won’t ever replace experimental methods, that doesn’t take away from their importance.

AlphaFold models are already useful for many practical aspects of structural biology: to design better protein expression experiments; to solve experimental structures faster, especially for X‐ray crystallography; to overcome tedious model building steps in experimental crystallographic and cryo‐EM maps; and to interpret lower‐resolution cryo‐EM maps.

Above all, the analysis of the models themselves can generate new and testable hypotheses about protein function. This actually is the great joy of structural biology: to derive mechanistic insight from a protein structure. In a sense, AI provides structural biologists with a new technique, bringing the fun of structure‐gazing without the effort of experimental work.

To better understand the enormous advance that AlphaFold and RoseTTAfold have achieved, it is worthwhile to look at the problem. Protein folding involves rearranging a linear sequence of amino acids in space to a physiologically preferred low‐energy state. Predicting the correct structure from the amino acid sequence alone and determining how the amino acid sequence dictates this 3D structure was deemed an intractable problem. The degrees of freedom in the peptide bonds create an astronomically high number of possible structures: going through all the possibilities would take longer than the age of the universe, even for a small protein (Levinthal’s Paradox).

The “protein folding problem” consists of three closely related puzzles:

1. What is the folding code?

2. What is the folding mechanism?

3. Can we predict the structure of a protein from its amino acid sequence?

Computational predictions were developed to circumvent testing protein structures one by one (sequential sampling). Over the past 40 years, these methods have steadily improved by comparing protein sequences to the growing number of protein structures solved through the aforementioned experimental methods — all freely available in the Protein Data Bank (PDB)

In 1994, the scientific community began a bi-annual competition (CASP) to test how well these computational tools could predict protein structures. The challenge was to predict the shape of proteins that had just been experimentally solved but not yet released to the public. Over the decades, gradual improvements were made, but everything changed with the arrival of AlphaFold. Its first version, launched only three years ago, applied artificial intelligence to the problem and was already a major breakthrough. By 2020, a redesigned version of AlphaFold achieved near-perfect predictions, creating a seismic shift not only in structural biology but also across many fields of science.

I already wrote an article outlining how AlphaFold works but “in short” (I enjoy discussing this stuff):

The user inputs a protein sequence. The principle used in protein folding models is that proteins with similar functions often have similar structures and thus sequences. AlphaFold2 thus aligns this sequence with multiple similar protein sequences and builds a multiple sequence alignment (MSA).

Sequences from various species are aligned in a 2D table format, placing corresponding amino aid residues — the parts of the amino acid that remain after two or more amino acids combine to form a peptide — in the same column and species in different rows. Most known protein sequences (about 70%) have at least one part that’s similar to a protein with a known structure. This is why MSA is crucial for predicting protein structure.

A high-quality MSA is essential for AlphaFold2 to produce an accurate prediction of protein structure. A diverse and deep MSA, with hundreds or thousands of sequences in the alignment, will help AlphaFold2 to identify co-evolutionary signals, conserved regions (parts that stay the same across different species), understand how mutations might affect the protein, and use all of these clues to accurately predict the protein’s 3D structure. A shallow MSA, with only tens of sequences and low variability among them will struggle to identify these patterns and is the most common reason for failing, non-confident and inaccurate AlphaFold2 predictions. Co-evolution, or covariation, is a process that allows the protein to preserve its structure even as it evolves. If two amino acids are in close contact, a mutation in one will be followed by a mutation in the other. Conversely, if two regions of a protein are changing and evolving independently from each other, it is likely that they are not in direct contact

AlphaFold2 also uses the MSA to predict and generate a set of pair representations modeling the interactions between every pair of amino acid residues based on the secondary structure of the protein, regardless of distance.

The MSA and pair representations are passed through AlphaFold2’s neural network (Evoformer) to interpret and refine both the MSA and the pair representations. AlphaFold2’s structure module takes both the refined pair representation and the original sequence (from the MSA) from the Evoformer. The structure module first turns this into a backbone of the 3D structure. It then finishes the modeling by placing the amino acid side chains and refining their positions.

If available, AlphaFold2 can feed supplied protein structures (e.g. structures derived from experiment) as templates into the Evoformer. However, AlphaFold2 tends to ignore such templates if there is enough information coming from the MSA. It is significant that AlphaFold2 doesn’t necessarily need templates as this differentiates it from homology modeling. Homology modeling is a computational method for predicting the three-dimensional structure of a protein based on its amino acid sequence and the known structure of a homologous protein (a protein with a shared evolutionary ancestry). It leverages the principle that proteins with similar sequences often share similar structures. AlphaFold2 doesn’t necessarily need these homologous proteins (templates) as data points for its structure prediction, meaning it can predict protein structures with a more limited evolutionary history — though AlphaFold2 still relies on evolutionary data through the MSA even if it doesn’t require homologous proteins.

AlphaFold2, and this step is incredibly important, then performs an iterative process called “recycling”. It feeds the MSA, the pair representations and the 3D structure back to the neural network, and generates a new 3D structure. This process is repeated numerous times, allowing AlphaFold2 to improve the accuracy of the final structure.

Along with the 3D structure, AlphaFold also provides confidence metrics, such as pLDDT, pTM and PAE enabling critical interpretation.

Essentially, AlphaFold2 analyzes 1-D data and based on that, predicts 2-D interactions. Both are refined and used to update each other, and then are finally used to predict the 3-D structure. The 1-,2-, and 3-D data is then all passed back to generate a new 3-D structure until no more improvements can be made.

RoseTTAFold, while not as accurate as AlphaFold2, uses significantly less computational energy and can thus predict the structures of large proteins on a gaming laptop. RoseTTAFold has a very similar architecture to AlphaFold and similarly relies on MSA. I won’t delve into how it specifically works but here is a great article if you want to learn more.

Gaseous protein folding

Professor Lars Konermann and I discussed another aspect of his lab and that is gaseous protein folding. In water, the driving force for protein folding is hiding the hydrophobic chains: it is the fear of water that drives things together.

In a vacuum, because there is no water, the opposite happens. The hydrophobic chains are very happy on the surface. The things that normally interact with water, the positively and negatively charged side-chains, would rather be packed together on the inside. Thus, in the absence of water, the hydrophobic side chains get squeezed to the outside and you get these super cool “inside out” structures. What is driving folding in a gaseous environment is the hydrophilic chains wanting to interact with something and so they pack together.

This is important because when mass spectrometry experiments happen, they occur in a vacuum or in a gaseous state, so its interesting and important to explore how the proteins behave in these areas. Since we bring a protein from a solution to the gas phase when we want to study its structure, the chemical reaction — the conversion of a properly folded protein into an inside-out protein — is incredibly slow because there is a high activation barrier. An activation barrier (or activation energy) is the minimum amount of energy that molecules need to overcome in order for a reaction or structural change to happen. To get the protein to rearrange itself into a new conformation (e.g., “inside-out”), you would need to supply a lot of extra energy — through heat, collisions, or some external force. If that energy isn’t supplied, the protein will basically stay as it is, because it’s energetically unfavorable to cross the barrier. This means that if you bring a properly folded protein into the gas phase, and if you do it under gentle conditions, you can actually preserve its structure very well because it requires an insane amount of energy to overcome the activation barrier again to unfold the protein.

You can also just start with the unfolded state of the protein and add heat to give it enough energy to overcome the activation barrier and properly fold. This is because when you add heat, you’re increasing the amount of energy each atom/molecule has in its vibrational, rotational, and translational motions.

Another application of gaseous protein folding is for drug development. Whenever you develop drugs, the first step in a drug action mechanism is always binding. If you have a small molecule and it does something to your body, that's always because it binds to something. In most cases, it will bind to a protein. For example, penicillin works by binding and blocking an enzyme’s active site.

To find a molecule that nicely binds to a protein is actually very difficult because its difficult to measure how effectively one molecule binds to another in a solution. However, if you bring a protein from solution into the gas phase, all you have to do is measure mass: if the mass increases it means that the molecule successfully bound to something.

You can do this in the pharmaceutical industry where you mix the protein of interest with thousands of compounds at a time and you can see exactly which one binded by measuring the weight of this new molecule. Mass is actually a very unique identifier in chemistry, so if you know the mass of each small molecule and the protein, the resulting mass of the compound would tell you if a molecule successfully binded and, if so, which one.

Thank you so much for taking the time to read this article! If you have any feedback, feel free to let me know!

Sources

1. Introduction to protein crystallization

2. What is protein crystallography

3. Cryo-EM explained

4. Analyzing Protein Structure and Function

5. Before and after AlphaFold2: An overview of protein structure prediction

6. A high level overview of AlphaFold

In genomics, we have something called the read, write, and edit (newly added due to CRISPR) model that has allowed us to examine, add, and even alter genomes.

The creation of this model has ultimately marked a massive transformation in the field of genomics. Only a decade ago, people with a PhD in genetics would often be sequencing a gene. Now, graduate students sequence genomes, and carry out massive functional studies trying to uncover how the order of bases in a DNA affects the traits of a human being.

To extrapolate from the genomics model, we can broadly categorize biotechnologies under what Elliot Hershberg, writer of Century of Biology and someone who I take massive inspiration from, describes as the 4-S model.

The four components of this model are Sequencing, Synthesis, Scale, Software.

Essentially, DNA sequencing and synthesis are being combined to produce large scale data that requires new software to wrangle and analyze it.

But what is the actual purpose of this model?

1. Helps identify commonalities across a range of projects and scientific papers.

   1. Instead of focusing on the minute details of every scientific paper, we can instead think about what part of the model each project fits in. This simplifies the process of reading otherwise complex papers.

2. Help us form a broad thesis about the foundational technologies that are powering the modern biotechnology revolution.

   1. We can begin to think about the role that each of these technologies are playing in reshaping our understanding of biology

Sequencing

In order to fully understand this model, let's dissect each part starting with sequencing.

This section refers to DNA sequencing which simply means to uncover the order of nucleotides, specifically the order of four bases: adenine, guanine, cytosine, and thymine. Because of the competition around developing efficient sequencing technologies, the cost to sequence a genome has become increasingly cheaper to the point where it costs less than 1,000 dollars.

The impact of cheap sequencing can not be overlooked. We are now able to explore far more complex relationships because of how affordable it is. For example, we can now begin to examine the relationship between genetic variation and health outcomes. This allows patients to be better with drugs based on their genetic and other personalized data. Furthermore, it allows us to create tools to better monitor our bodies and health.

Specifically, it has been said that the composition of one's gut microbes can be highly predictive of disease outcomes, and can even predict spikes in blood sugar after eating different types of food. Thus, examining our gut microbes, which is just simply the community of microorganisms in our digestive tract, could become essential for understanding how different diets impact our bodies on an individual basis.

With the drop in cost of genome sequencing, projects such as UK Biobank have emerged and have been able to generate whole-genome sequencing (WGS) data for 200,000 individuals, the largest amount of sequencing data in history at that time.

The All of Us program now aims to generate WGS for over a million Americans, really showing how ambitious we can now be. The most dominant force in genomics is Illumina with their development of next-generation sequencing (NGS).

This technology allows us to more quickly and efficiently sequence DNA in comparison to our previous technology in Sanger sequencing. This technology works by using flow cells, which are small devices that contain tiny channels or “lanes” where DNA samples can be placed for sequencing, each of the DNA sequences in the “lanes” can be sequenced simultaneously, greatly improving the speed of genomics. This technology is ultimately what makes modern genomics possible and cheap.

The development of new sequencing technologies does not stop there, companies such as Oxford Nanopore and Pacific Biosciences are both developing long-read sequencing technologies. This makes it possible to analyze complex sequences of the genome that were difficult or impossible to measure with technologies like Illumina.

This moves us closer to a future where an individual’s genome could be sequenced and assembled in an automated fashion.

Synthesis

Moving on, it is now time to discuss synthesis which just refers to DNA synthesis. DNA synthesis is the natural or artificial creation of deoxyribonucleic acid (DNA) molecules. The reason why it is important to efficiently create, artificially or naturally, DNA molecules is because the use cases of DNA are expanding.

DNA is now being constantly used as a substrate, meaning that it is being manipulated or modified to perform specific functions beyond its natural role in encoding genetic information. A central part of the synthesis category is oligonucleotides, commonly called oligos. These short sequences of nucleotides are essential for many of our foundational technologies such as PCR (Polymerase Chain Reaction), the aforementioned Illumina sequencing, and CRISPR for gene editing. Without oligos, these technologies would not work, preventing us from ultimately expanding the use of DNA in a lab environment.

As a result of the importance of oligonucleotides, oligos have become widely accessible, making it possible for them to be designed, ordered, and delivered all in a single day. The importance of oligonucleotides can not be understated.

Oligo synthesis, which is the process of generating multiple oligonucleotide sequences in parallel, is a new tool that has been developed by companies such as Integrated DNA Technologies (IDT) and Agilent. Thanks to large scale oligo synthesis, we are now able to use NGS technologies to measure properties of individual cells called single-cell sequencing.

Single-cell sequencing is extremely important because it allows us to study genetic information of individual cells. Traditional sequencing methods typically analyze a bulk sample of cells and provide an average representation of the genes present. This is slightly problematic because there could be the rare case where a couple of genes strongly affect the regulation of cells while others barely contribute; however, because it is an average you never get to know this and only see how as they group they were around average on their effect of cell regulation.

10X Genomics is a major provider of this single-cell sequencing technology, with an approach designed around using gel beads coated with oligos and containing barcode sequences that can be used to capture and identify individual cells for measurement. This is an example of a breakthrough technology built on top of both sequencing and synthesis.

Another example of the intersection between synthesis and sequencing is in the realm of storage.

Problem:

As a species, we are generating and collecting data at a rate that will soon make efficient physical data storage a hard problem

Solution:

DNA offers an enticing solution because it is the only highly stable nanoscale information storage technology that we know of. Conceptually, an exabyte (one million terabytes) of data could fit in the palm of your hand. Many labs and companies are working on combining large-scale synthesis and sequencing to both reliably store information (synthesis) and retrieve it (sequencing).

Some potential areas of improvement for the synthesis space are as follows:

1. Increasing sequence length

   1. Increasing the length and size of DNA will mean that we could potentially store more data in DNA

2. Decreasing cost

   1. Making oligos even cheaper means that it will become easier to continue to develop technologies. One project that is working on this issue is the GP-write project, a collaboration between academic labs and companies such as Twist Bioscience, Ansa Biotechnologies, and DNA Script, which aim to reduce the cost of engineering and testing large genomes in cell lines by more than 1,000-fold, all of this within ten years. Testing large genomes in cell lines refers to the process of evaluating the genetic makeup and functionality of a complete set of genes within cells.

Scale

While the sequencing and synthesis categories both refer to foundational technologies that independently change the space of biology, the scale portion augments the capabilities of sequencing and synthesis, serving as a complementary piece.

The goal of scale is to speed up the processes so that more data can be produced in a faster window of time. The method by which this speed is achieved is through the intersection of the semiconductor and biological revolution.

The semiconductor revolution refers to the development and mass production of computer processors, microcontrollers, and memory chips, enabling the computer revolution and the creation of extremely powerful machines. On the other hand, the biological revolution transitioned from macroscopic descriptions of living organisms to molecular descriptions of cells.

The intersection lies in the fact that with more powerful and eventually cheaper computers, we now have the machinery to run complex and powerful software tools that are being developed in the biotechnology industry. This ultimately allows for faster and better data collection and analysis. A clear example of scale that was already mentioned was sequencing DNA simultaneously in the Illumina sequencing model. Thanks to the semiconductor revolution, we are now able to focus on making an entire system faster.

A clear example of this already happening is from the transition from reporter assays to massively parallel reporter assays (MPRA).

Most experiments consist of perturbing a system and then measuring what occurs. In molecular biology and genetics, perturbing a system involves intentionally inducing changes or disturbances to genes in order to investigate their response and hopefully understand their function.

Going back to the reporter assays, these are perfect examples of this experimental model. Reporter assays are often used in molecular biology to figure out what regulatory sequences of DNA, which are just specific segments of DNA that control the activity of genes (on or off), are necessary to turn a gene on (meaning that it is actually functioning and producing proteins) and off.

The main premise is to construct a DNA strand that has both the regulatory element in question (the one responsible for turning the gene on or off) and a gene that can be easily detected when it is expressed (when the gene is on and producing proteins). The end goal of a reporter assay is to hopefully understand the function of a specific sequence of DNA. Historically, the core measurement tool would often be a reporter protein such as green fluorescent protein (GFP). Reporter proteins are just proteins that can be easily detected.

The major shift is that instead of measuring using a reporter protein we can now synthesize or mix them in a pool of oligos. This allows us to test a massive number of different sequences in parallel, hence the creation of the MPRA.

The results of an MPRA are then measured using sequencing. This is accomplished by taking the ratio RNA to DNA of a specific barcode in the oligo pool. If a sequence drives RNA expression, we should see far more RNA copies of the barcode than the original DNA barcode because the DNA activated the genes and told them to produce RNA.

This is a really important change because it means that we can explore more DNA strands quicker. This is especially huge for human geneticists who have been trying to understand how DNA differences lead to differences in traits.

One of the biggest challenges though is that most of your complex traits or traits that are influenced by multiple genetic factors (height, intelligence, or susceptibility to disease) are not actually influenced by the parts of your DNA that affect the DNA sequence in genes.

Genes are specific segments of DNA that provide instructions for the production of proteins or functional RNA molecules, the sequence in these genes ultimately influencing what proteins are produced which in turn influence your traits. However, many of our complex traits are influenced by genetic variants that do not affect the sequence of DNA inside genes, meaning that it becomes challenging to understand and explain how they influence these traits. It is not as straightforward as them affecting the sequence inside genes because they might actually be impacting gene regulation, protein interactions, or other complex molecular processes. Now that we have the technology to efficiently test countless strands of DNA, testing a huge variety of strands is not nearly as expensive.

To extrapolate from this one example, if you can first find a way to effectively use sequencing and synthesis approaches to tackle a problem, we can then begin to scale the experimental throughput. For any tool that can fit into this paradigm, you can go from a dozen to thousands of measurements.

Software

Our final section of the model is software. Similar to scale, this section also serves as a complementary piece to our core synthesis and sequencing technologies.

The goal of software is to compile and analyze the huge amount of data produced by both DNA sequencing and synthesis. While collecting data is important, the data means nothing without proper analysis and dissecting the key points.

Bioinformatics is literally an entire scientific discipline focused on developing software to better analyze data. Similar to scale, the better the software, the more efficient our DNA sequencing and synthesis technologies. Without proper software, sequencing a genome would not be possible.

The effectiveness of software in making data both readable and sensible has led to a concept called Software 2.0 as described by Andrej Karpathy, Director of AI at Tesla. The essential premise is that previously we used complex math and physics equations to help us understand the laws of our universe; however, we can now use an immense amount of data to understand the laws that define our universe.

In other words, instead of attempting to write down the rules of a program that would get us the solution, we work backwards by collecting large volumes of data that represent the solution, allowing us to understand the rules of the program.

A perfect example of Software 2.0 is machine learning. Recently, machine learning was used to effectively solve the protein structure prediction problem.

Similarly to translating languages, there is no simple equation to understand how amino acids fold into 3D machines; however, it is possible to learn how using massive amounts of data and enormous models.

Another use of software is as a platform for design and discovery.

Problem:

There is no dedicated platform for biological design and discovery, making it difficult to compare experiments in the space as everyone uses different methods.

Solution:

One approach has been to develop pure software platforms for scientists. In genomics, genome browsers have become the de facto representation of genetic information. A huge benefit of a dedicated genome browser is that they help eliminate waste and inefficiency. With common infrastructure, individual groups and companies don’t have to reinvent the wheel and build the same tools that everybody else is using, there is now a standard, making it easier to share and edit work.

This new biological revolution will constantly require new platforms spanning the entire spectrum from pure software all the way to model-driven labs. This makes for a fantastic opportunity for talented engineers to use their expertise to create an abundant future in the physical world. Who knows how the next software can affect how we approach biology and science as a whole.

Sources:

1. https://centuryofbio.com/p/4-s-model

2. https://centuryofbio.com/p/whats-different-part-one-sequencing

3. https://centuryofbio.com/p/whats-different-part-two-synthesis

4. https://centuryofbio.com/p/whats-different-part-three-scale

5. https://centuryofbio.com/p/whats-different-part-four-software

How drug delivery became essential

A couple decades ago, small molecules dominated the pharmaceutical landscape. These drugs, like aspirin or statins, relied on straightforward chemical and physical principles to enter the body and work their magic. Efforts at the time primarily focused on tweaking the molecules themselves: improving solubility, controlling release, or optimizing absorption.

But as treatments advanced, so did the challenges. Enter biologics: large, complex molecules like proteins, peptides, and nucleic acids. These groundbreaking therapies opened doors to previously untreatable conditions, but they came with significant hurdles. Proteins and peptides, for example, degrade easily in the body, while nucleic acids need to enter cells to work. A simple pill wouldn’t cut it anymore.

Suddenly, drug delivery systems—whether in the form of nanoparticles, liposomes, or hydrogels—weren’t just optional. They were indispensable. The modern era of medicine relies on these innovations to transform fragile molecules into effective therapies.

COVID-19 and the revolution in nanoparticles

The importance of drug delivery came into sharp focus during the COVID-19 pandemic. The much-lauded mRNA vaccines by Pfizer–BioNTech and Moderna were not just medical breakthroughs; they were also feats of engineering.

The star players? Lipid nanoparticles (LNPs).

These microscopic particles protected the mRNA from being destroyed by the body before it could reach its target. They also allowed the mRNA to slip inside cells, where it could instruct the body to produce its immune-boosting proteins. Without LNPs, the promise of mRNA-based therapies might still be a distant dream.

Why are nanoparticles so effective?

1. Protection from degradation. Molecules like mRNA are delicate and break down quickly in the body. Nanoparticles provide a shield, ensuring they stay intact long enough to work.

2. Controlled release. Injecting a drug directly can cause a sudden spike in concentration, potentially leading to side effects. Nanoparticles allow for a slow, steady release, reducing toxicity.

3. Cellular targeting. Nanoparticles are small enough to infiltrate cells, making them ideal for therapies that need to act inside the body’s building blocks.

Still, challenges remain. The mRNA vaccines needed ultra-cold storage—an obstacle in regions without cold-chain infrastructure. The vaccine was shipped at –70° Celsius and is usable from a refrigerator for about 10 weeks. Having to keep the vaccine in such cold temperatures was not convenient and meant that the vaccine could be inaccessible to areas where cold chain transport isn’t possible. Pfizer and BioNTech have already worked on and increased the stability of mRNA vaccines.

From rock star dreams to scientific stardom

This brings us to a fascinating conversation I recently had with Professor Jeremiah A. Johnson — a professor at the MIT Department of Chemistry. Long before his time of working at a lab bench, Professor Johnson dreamed of a very different kind of stardom.

Growing up, Johnson imagined himself as a rock star or a major league baseball player. When he enrolled at Washington University in St. Louis, it was largely because the baseball coach promised him a spot on the starting lineup. He even planned to major in music—chasing his dream of becoming a rock star while balancing baseball.

However, financial pressures led him to pursue something “practical.” Electrical engineering seemed like a safe bet, especially since it overlapped with his interest in building recording studios. Yet, after slogging through circuits and electromagnetism, Professor Johnson quickly realized this wasn’t his calling.

The accidental chemist

While aimlessly following the pre-med crowd in his dorm, Professor Johnson decided to take organic chemistry in his sophomore year—a class he initially saw as a checkbox for medical school. But what started as a whim turned into a passion.

“It’s like creating art. You’re sculpting molecules from scratch.” It was just as much of a science as an art; you are literally creating and designing molecules from scratch. Just how a sculptor makes a sculpture, he was creating a molecule and there is some kind of creativity and artistic beauty behind it.

At the same time, Professor Johnson picked up a job washing glassware in Professor Karen Wooley’s chemistry lab. He didn’t know it then, but Wooley was a pioneer in polymer chemistry and drug delivery. Over time, instead of just mindlessly washing glassware, he began asking the students at the lab about the experiments they were working on. One day, a graduate student encouraged him to join the lab as an undergraduate researcher.

By the end of his sophomore year, he declared a double major in Chemistry and Biomedical Engineering, with a minor in Music. His rock star dreams had shifted—now, he wanted to become a professor leading a groundbreaking lab, much like Wooley.

Why drug delivery?

Johnson’s journey into drug delivery stems from his love of chemical biology. But practicality also played a role. In a field driven by funding priorities, drug delivery offered an undeniably impactful application for his creative molecular designs.

If you can deliver drugs selectively to diseased cells, you can improve efficacy and reduce side effects: that combination saves lives.

His lab now focuses on creating polymer-based nanoparticles that can adapt to the unique needs of different diseases. For ovarian cancer, the challenge lies in overcoming drug resistance. For multiple myeloma, it’s about precision—ensuring that treatments don’t harm healthy tissue.

Lessons

I think Professor Johnson’s story offers a couple key takeaways:

1. You don’t have to have everything figured out. It’s okay to try different paths and let your interests evolve. A chance class or a part-time job could lead you to your passion. Professor Johnson often sees graduate students who don’t entirely know what they want.

2. Merge your passion with practicality. Professor Johnson saw chemistry through the lens of art and that made him enjoy doing it. He was able to retain the artistic process he loved but also have a translational impact.

3. Take initiative. Ask questions, seek opportunities, and immerse yourself in the things that excite you. Johnson’s curiosity turned a glassware-cleaning gig into a lifelong career.

Now lets get to Professor Johnson’s work.

Professor Johnson and I discussed two of his studies —published in Nature Nanotechnology (2022) and Journal of the American Chemical Society (2014)—that highlight his development of bottlebrush prodrugs, with each study adding modifications to the initial concept. Each study attacks a core problem in drug delivery and proposes a carefully engineered molecular solution.

Complexity in cancer drug delivery

Modern medicine increasingly relies on a single, deceptively simple concept: get the right drug to the right place at the right time. But this is far easier said than done — especially in cancer treatment which relies on combination chemotherapy, where multiple drugs are administered together to increase tumor cell death and prevent drug resistance. The problem is the treatment needs to be delivered in precise ratios, simultaneously, and only at the site of a tumor. Each drug behaves differently in the body: it may degrade too quickly, cause toxicity to healthy cells, or fail to reach the tumor entirely. The challenge thus lies in delivering all these drugs in the exact synergistic ratio (the ratio that works best when all drugs enhance each other’s effectiveness) directly to the tumor site.

Delivering chemotherapy drugs in nanoparticle form could help reduce side effects by allowing the drugs to release directly at the tumor site, reducing toxic side effects and improving efficacy. In recent years, scientists have developed nanoparticles that deliver one or two chemotherapy drugs, but it has been difficult to design particles that can carry any more than that in a precise ratio. Furthermore, only a handful of nanoparticle drug formulations have received FDA approval to treat cancer, and only one of these particles carries more than one drug.

Johnson’s team saw an opportunity: what if you could design a nanoparticle that is pre-loaded with multiple drugs—built from the ground up with the exact ratio and release profile you want? That’s where the bottlebrush prodrug concept came in. Instead of building the delivery vehicle and then attaching drug molecules, Professor Johnson created building blocks that already include the drug. These building blocks can be joined together in a very specific structure, and the researchers can precisely control how much of each drug is included.

The creation process

Professor Johnson’s bottlebrush prodrug concept came about after a series of reflections. He first examined traditional constructions of drug delivery vehicles:

1. The first popular concept is building a vehicle that will circulate throughout the body (such as nanoparticles)

   1. These vehicles are characteristically small - less than 100 nanometers

   2. The drug can be encapsulated, which is easier to make but you have less control of when the drug releases and it is less adaptable because you have to design the nanoparticle to accommodate the drug’s physical properties. As a result, one drug might work, but then if you try to use a different drug — with different physical properties — it might not be encapsulated within the delivery vehicle.

   3. The drug can also be chemically bonded to a polymer chain: a polymer-drug conjugate. This offers much more programmability: control over how the drug is bonded, the drug ratio, and how/when the drug gets released. On the other hand, the chemistry to create this vehicle is incredibly complex, especially if you are trying to bond multiple drugs. At the time, there were two templates for polymer-drug conjugates

      1. A linear polymer chain with drugs attached

      2. A star-shaped polymer (multiple arms attached to a core). Onto the ends of that star polymer you can attach your drug molecules

      3. In both of these approaches they have surface exposed payloads (visible to the “outside world” — our body — and readily available to be cleaved by an enzyme). This could cause an undesired release when the drugs come out too early (losing what is called selectivity: the degree to which a drug acts on a given site relative to other sites)

      4. Another issue, similar to the encapsulation method, is that the drug’s physical properties affect the whole vehicle. If a drug is very hydrophobic (”afraid” of water) and you attach it to your polymer, even if that polymer is hydrophilic (loves water), the polymer will become hydrophobic and unable to dissolve/bond with water. Instead, it will aggregate (stick together). If you change to a different drug, it might now all of a sudden become soluble again. Again, because these drugs are present on the surface, the drugs physical properties will dominate the properties of the polymer drug conjugate.

2. An implant: something that sticks to the body and then releases the drug

Clearly there needs to be a shift in drug delivery vehicle fabrication. Professor Johnson was inspired by a paper in 2006, outlining the design of a biodegradable bowtie dendrimer as a drug delivery vehicle of doxorubicin. A bowtie dendrimer is a highly branched polymer involving a base and then “branches” splitting off the base just like a tree, except this time it is from both ends — like in a bowtie.

In this paper, 16 of the arms had polyethylene glycol (PEG: a biocompatible polymer that shields the drug from immune detection and improves stability. PEG is found everywhere — in our shampoo, soap, and even the COVID vaccine was coated in PEG) chains and then the other 16 had drug molecules attached. The PEG chains are much longer than the core, which allow them to wrap around and shield the core and the chains with drugs. As opposed to traditional drug delivery vehicles where the drug’s physical properties dictate the systems’, the overall physical properties of this large structure are basically the same because the drugs are on the inside.

Yet, this structure had its limitations:

1. Spherical symmetry. If you want to conjugate (bind) this to an antibody or a specific cell, it is hard to do so because there is not one specific place.

• Synthesis. Making this structure is actually super difficult. In principle you could change the drug molecules to anything you want, but you'd have to go back and spend a year synthesizing this molecule every time.

The most critical point from this structure is that drugs and PEG chains can be combined in one single nanostructure and these PEG chains protect the drugs

This paper inspired Professor Johnson’s idea of a bivalent bottlebrush: the molecule has a rigid backbone with polymers on each repeat unit, extending the backbone. This molecule is similarly resistant against the drug’s physical properties, but the bonus of having drug molecules on every repeat unit is that his vehicle has cylindrical symmetry; the ends of the molecules provide a place to conjugate to an antibody.

This molecule is also far easier to synthesize, at least in principle, because rather than loading the drugs into a pre-made particle, the vehicle could be created using modular building blocks — each one a combination of:

• A drug molecule

• A linker (chemical bond that can be cleaved to release the drug)

• A PEG chain. PEG chains hide the drug from the body’s enzymes, delaying release until the nanoparticle reaches the target site.

Hundreds of these blocks were then linked together using a method Johnson’s lab pioneered called brush-first polymerization. Unlike older methods that try to attach drugs to a finished polymer, this process starts with monomers— the building blocks— and are then linked into a full bottlebrush polymer in a single, controlled step. The bottlebrush shape looks like a central linear backbone with densely packed side chains extending outward—like bristles sticking out from a brush. In Professor Johnson’s design, each of these bristles is a prodrug: a drug that’s chemically inactivated and reactivated only when it reaches the right biological conditions.

This synthesis strategy is revolutionary because:

• It avoids complicated post-synthetic modifications. The polymer can be built in a single step once the building blocks are constructed.

• It ensures consistency across batches.

• It allows researchers to mix and match drugs by simply mixing the corresponding monomers in the desired ratio.

Another significant advantage of the bottlebrush prodrug platform is instead of measuring the synergy (how effective the drugs are together) with drugs in a lab dish, Professor Johnson’s team tested drug combinations within the nanoparticle structure itself, giving a far more realistic view of how the drugs will behave in the body. Traditionally, researches test potential drug combinations by exposing cancer cells in a lab dish to different concentrations of multiple drugs, yet these findings often don’t translate to the human body. Testing the drugs while already embedded gives a more realistic view of how they’ll behave.

Theoretically, this system could be scaled for more drugs. If you wanted five drugs, you could make five distinct building blocks, each with a different drug in side, and assemble them into a particle. In principle, there’s no limitation on how many drugs you can add, and the ratio of drugs carried by the particles just depends on how they are mixed together in the beginning.

This new system essentially removes all the guesswork of wondering how different drugs will affect the overall nanoparticle or whether the physical properties of the drugs are compatible with the vehicle’s

Study 1: a new era of triple-drug delivery for ovarian cancer

In the 2014 study published in Journal of the American Chemical Society, Johnson’s lab tackled a long-standing issue in cancer nanomedicine: how to load multiple chemotherapeutic drugs into a single delivery vehicle while controlling their dosage and release.

Goal:

To create a new type of nanoparticle capable of carrying three common ovarian cancer drugs—cisplatin, doxorubicin, and camptothecin—in precise ratios, with each drug released through a different mechanism, and all protected from premature degradation.

Professor Johnson showcased his bottlebrush prodrug carrying three different cancer drugs each with a different release mechanism:

• Cisplatin: Released upon exposure to glutathione, an antioxidant present inside cells.

• Camptothecin: Released by esterases, enzymes also common inside cells.

• Doxorubicin: Doxorubicin was attached using a photo-cleaver linker, so that when ultraviolet (UV) light is emitted, it has enough intensity that the linker absorbs the light and releases the drug. This was mainly a proof of concept to show that external stimuli can also trigger a release. The biggest limitation is that light doesn’t go through your body so the tumor has to be near the surface and it would probably require creating a hole.

Once all three drugs are released, all that is left behind is PEG, which is easily biodegradable. This kind of control means that drugs can be timed to act in succession, a strategy that can enhance therapeutic effects and reduce side effects.

Impact:

Johnson’s triple-drug nanoparticle showed higher cancer cell kill rates than any single- or dual-drug particles. Moreover, the approach provided a scalable path to tailor nanoparticles for any combination of drugs. It laid the foundation for applying this technology to other types of cancer.

Study 2: tackling multiple myeloma with precision nanomedicine

In this more recent and expansive 2022 study, Johnson’s team pushed their bottlebrush system further—this time focusing on multiple myeloma, a cancer of plasma cells found in bone marrow. Healthy plasma cells help fight infections by making proteins called antibodies, which find and attack germs. In multiple myeloma, cancerous plasma cells build up in bone marrow — the soft matter inside bones where blood cells are made. In the bone marrow, the cancer cells crowd out healthy blood cells. Rather than make helpful antibodies, the cancer cells make proteins that don't work right. This leads to complications of multiple myeloma.

The challenge:

Multiple myeloma is usually treated with a three-drug combination: bortezomib (a proteasome inhibitor), pomalidomide (an immune modulator), and dexamethasone (an anti-inflammatory). But giving these drugs separately often leads to poor targeting, side effects, and non-synergistic absorption—meaning the drugs don’t reach the tumor in the effective ratio.

Using their bottlebrush prodrug platform, Johnson’s lab created nanoparticles that carry all three drugs at once. Using these particles, Professor Johnson was able to calculate and then deliver the optimal ratio of three cancer drugs used to treat multiple myeloma.

Experimental Design:

Mice with multiple myeloma tumors were treated with:

1. Free drugs in solution. The researchers showed that nanoparticles carrying three drugs in the synergistic ratio identified tumors much more than when the three drugs were given at the same ratio but untethered to a particle.

2. Mixtures of three single-drug nanoparticles

3. One three-drug bottlebrush nanoparticle

4. Interestingly, they also tested a bortezomib-only bottlebrush. Bortezomib is a proteasome inhibitor, a type of drug that prevents cancer cells from breaking down the excess proteins they produce. Accumulation of these proteins eventually causes the tumor cells to die. When bortezomib is given on its own, the drug tends accumulate in red blood cells, which have high proteasome concentrations (not ideal). However, when the researchers gave the bottlebrush prodrug version of the drug, they found that the particles accumulated primarily in plasma cells because the bottlebrush structure protects the drug from being released right away, allowing it to circulate long enough to reach its target. This shows that the structure improved efficacy with even just one drug by improving its bioavailability.

5. Result: The three-drug bottlebrush significantly outperformed all other groups in reducing tumor size.

The versatility of this nanoparticle platform, as demonstrated by these two studies, means it could potentially be deployed to deliver drug combinations against a variety of cancers, because you are able to apply such a variety of combinations of drugs.

Conclusion

Professor Johnson’s work embodies the best of scientific creativity and engineering pragmatism. His bottlebrush prodrug platform doesn’t just solve one problem—it offers a new way of thinking about drug design, delivery, and disease treatment. And this revolutionary work came from someone who didn’t even plan on being a scientist; life truly has an incredible way of surprising us.

If you want to read more of her work or read further into the works we mentioned, here are some interesting papers/projects:

1. https://www.scientia.global/wp-content/uploads/Sangya-Argawal.pdf

2. https://www.cell.com/cell/fulltext/S0092-8674(21)01331-3?uuid=uuid:b2aa2039-b939-413c-8325-b84e208f55b7

3. https://scholar.google.com/citations?view_op=view_citation&hl=en&user=c5RB6pcAAAAJ&citation_for_view=c5RB6pcAAAAJ:L8Ckcad2t8MC

4. https://scholar.google.com/citations?view_op=view_citation&hl=en&user=c5RB6pcAAAAJ&citation_for_view=c5RB6pcAAAAJ:kNdYIx-mwKoC

What is Obesity?

Obesity is a disease around abnormal or excessive fat accumulation that presents a risk to health. More than 1/3 of U.S. adults are affected by obesity and over 800 million people are worldwide. Almost 50% of U.S. adults are expected to be obese by 2030. Obesity significantly increases the risk of numerous comorbidities, including cardiovascular disease, type 2 diabetes, and certain cancers, contributing to approximately 4.7 million deaths annually, and accounting for 8% of global mortality.

Obesity was first recognized as a disease in 1948 by the World Health Organization (WHO), yet it took until 2013 for the American Medical Association (AMA) in the United states to recognize obesity as a disease. This recognition is incredibly important. Obesity for a while, and still, has been seen as an individual’s fault: eating too much and moving too little. And while these factors can certainly contribute to obesity, obesity is caused by an interplay of genetics and ones lifestyle choices.

Body mass index (BMI) has been the standard measure of obesity:

$$ weight(kilograms)/height^2(meters) $$

Someone with BMI of 30 or higher is considered to be obese, yet this marker has its limitations; it is not a definitive indicator of wether someone has obesity or not. Yet, when we talk about obesity, the conversation almost always begins and ends with the scale. The underlying assumption for decades was simple: if obesity is an excess of fat, then the solution must be to reduce overall weight. Doctors measure weight, calculate body mass index (BMI), and use those numbers to assess health risk. This logic gave rise to treatments focused narrowly on calorie balance, appetite suppression, or fat absorption. These tools are simple and standardized, but they overlook something critical: where fat is stored in the body matters more than how much of it there is.

Rethinking Obesity

Two people can have the same BMI and look similar on paper, yet carry very different health risks. The reason is that fat is not all alike—it exists in different compartments, and those compartments behave in very different ways.

1. Subcutaneous fat is the soft layer under the skin which can be pinched. It serves as an energy source, helps insulate the body, protects tissues, and supports nerves and blood vessels. It is often metabolically neutral or even protective.

• Visceral fat is the fat packed deep inside the abdomen, surrounding the liver, pancreas, and intestines. Unlike subcutaneous fat, it is biologically active. It constantly releases fatty acids and inflammatory molecules into the bloodstream, especially into the liver through what is called the portal vein. This process disrupts sugar regulation, raises cholesterol and triglyceride levels, and promotes high blood pressure.

It is visceral fat, not simply total body fat, that drives the cascade of complications we associate with obesity—type 2 diabetes, cardiovascular disease, fatty liver disease, and more.

Other markers for obesity include waist circumference, and waist-to-hip ratio (WHR), with WHR emerging as the most precise predictor of metabolic health and disease risk. Because visceral fat increases waist size while subcutaneous fat in the hips and thighs balances it out, WHR has emerged as the best simple measure of fat distribution. It is calculated by dividing waist circumference by hip circumference. A higher ratio means proportionally more abdominal fat.

Large-scale studies, including those involving hundreds of thousands of participants, consistently show that WHR is more predictive of long-term health outcomes and mortality than BMI. In other words, knowing someone’s waist-to-hip ratio tells us far more about their likelihood of developing diabetes, heart disease, or dying prematurely than knowing their BMI alone. This finding has been replicated across populations and ethnic groups, and it holds true even when BMI is within “normal” ranges.

The implications are profound: if our true goal is to reduce the diseases and deaths linked to obesity, treatments must do more than lower weight or BMI—they must also improve fat distribution. A lower WHR, reflecting less visceral fat and proportionally more subcutaneous fat, is what truly protects against long-term metabolic harm.

GLP-1 receptor agonists - a class of type 2 diabetes drugs that not only improve blood sugar control but may also lead to weight loss — have long been the standard for obesity medication. Patients on GLP-1 therapies often see dramatic reductions in total body weight, and for many, this is life-changing. GLP-1 agonists, like Ozempic, work by mimicking a natural hormone called glucagon-like peptide-1 (GLP-1) in the body. This hormone helps regulate blood sugar levels and promotes feelings of fullness. By activating GLP-1 receptors in the brain and gut, these medications suppress appetite, increase satiety, and slow down gastric emptying, leading to reduced food intake and weight loss.

But here is the limitation: despite their success on the scale, GLP-1 agonists do not reliably change WHR. A 2023 meta-analysis of ten randomized controlled trials found that while patients on GLP-1 drugs lost substantial weight—including some visceral fat—their WHR remained largely unchanged. That means the most dangerous pattern of fat distribution, centered in the abdomen, persisted.

This matters because it helps explain why losing weight does not always translate into the expected reductions in cardiovascular risk. If the distribution of fat remains unfavorable, the metabolic risk remains high.

In short, today’s most effective weight-loss drugs shrink fat mass, but they do not solve the fat distribution problem. What is needed is a therapy that doesn’t just make people lighter, but makes them metabolically healthier by reshaping where fat is stored. There is currently an unmet need for obesity treatments targeting WHR.

Waist-to-hip ratio is not only a predictor of metabolic health—it is also a long-recognized marker of reproductive fitness. Anthropologists and evolutionary biologists have studied WHR for decades, and across cultures, it is strongly correlated with fertility, reproductive success, menstrual cycle length, and the timing of menopause.

How is reproduction related to fat distribution?

Reproductive processes are tightly linked to hormones, many of which also influence metabolism. Estrogen, for example, promotes fat storage in the hips and thighs, which lowers WHR. During menopause, as estrogen levels decline, fat distribution shifts toward the abdomen, raising WHR and increasing visceral fat. This transition is one reason post-menopausal women often experience a surge in metabolic risk, even without major changes in weight.

Beyond menopause, conditions such as polycystic ovary syndrome (PCOS) provide additional clues. PCOS is characterized by abnormal secretion of reproductive hormones, particularly luteinizing hormone (LH), and women with PCOS frequently show higher waist-to-hip ratios and central obesity. Similarly, altered patterns of LH secretion during the menopausal transition are linked to changes in both BMI and WHR. These associations suggest that reproductive hormones—and especially LH—play a meaningful role in determining how fat is distributed in the body.

A Hormone at the Center: Luteinizing Hormone (LH)

Among the reproductive hormones, LH stands out. LH is produced in the pituitary gland, a small but powerful organ at the base of the brain that coordinates many of the body’s hormone systems. In women, LH triggers ovulation and stimulates the ovaries to produce estrogen and progesterone. In men, LH signals the testes to produce testosterone.

But there is reason to believe LH does more than control reproduction. Some large population studies have shown correlations between LH levels and body composition, suggesting that LH may also act as what researchers call a “pro-lean hormone”—a hormone that reduces fat storage and increases energy expenditure. The idea that a reproductive hormone could double as a regulator of fat metabolism opens a fascinating possibility: could the receptor for LH be a new target for obesity treatment?

For LH to have an effect, it needs to bind to its protein receptor. That receptor is called LHCGR (luteinizing hormone/choriogonadotropin receptor). LHCGR is part of a family of proteins known as G protein-coupled receptors (GPCRs). GPCRs sit on the surface of cells and act like molecular switches: when a hormone or chemical binds to them, they trigger a cascade of signals inside the cell that change how the cell behaves.

Traditionally, LHCGR is thought to exist only in reproductive tissues—the ovaries and testes. There, it drives the production of sex hormones and controls fertility. But if LHCGR were also found in fat tissue, it would mean that LH could directly influence fat metabolism, bridging the gap between reproduction and obesity.

Finding this out would be incredibly significant because we currently can’t change the genetic factors that influence waist-to-hip ratio (WHR), and lifestyle factors like exercise and diet often fail in the long run because people struggle to stay consistent. Reproductive factors are our last hope. Specifically, targeting the expression of a receptor could provide a way to modify WHR, and this approach could realistically be used in a pharmacological setting.

In summary, the history of obesity and its treatments shows two things clearly:

1. Reducing weight is not enough. Treatments must also improve the quality and distribution of fat.

2. Reproductive hormones may hold the key. Observations from menopause, PCOS, and fertility research all point to hormones like LH influencing fat distribution.

This hypothesis—that LHCGR is present in human fat, helps regulate fat distribution, and play a role as a reproductive determinant of WHR—is what motivated our research. To test it, we needed to proceed step by step:

1. Confirm whether LHCGR is actually present in human fat cells.

2. Investigate whether genetic variation in LHCGR is linked to differences in waist-to-hip ratio.

3. Test whether activating LHCGR changes fat distribution in experimental models.

Each step required a different set of tools, from genetics to microscopy to animal studies. Together, our study built a case for reimagining LHCGR not just as a fertility receptor, but as a novel therapeutic target for obesity.

Step 1 — Is the LH receptor (LHCGR) actually in human fat?

If LHCGR isn’t present in adipose tissue, there’s no plausible way for LH signaling to directly reshape fat distribution. So the first thing we had to establish—beyond any doubt—was cellular presence: does LHCGR show up in human fat, and in which cells?

To answer that convincingly, we used orthogonal (independent and complementary) approaches:

1. Immunohistochemistry: a protein-level method to see the receptor in place inside tissue

2. Bulk RNA-seq and Single-cell RNA-seq: RNA-level methods to confirm the gene is actively expressed

Immunohistochemistry

Immunohistochemistry is a molecular “staining” technique. You slice the tissue very thin, place it on a slide, and apply an antibody that binds only to your protein of interest—in this case, LHCGR. A detection system makes the antibody’s binding visible (a colored or brown signal under the microscope). If you see a clean signal in the right cell types and locations, the protein is there.

Antibodies can sometimes bind to the wrong thing — “off-target” binding. To prevent false positives, we included positive and negative controls:

• Positive control: a tissue where LHCGR is unquestionably present. You used testis, where LHCGR is well known to be expressed in Leydig cells. We observed strong staining in Leydig cells—as expected—confirming the antibody works.

• Internal negative control: in the same testis section, Sertoli cells do not express LHCGR. We saw no staining there, which is exactly what we wanted; it reassures specificity.

With encouraging results from the positive and negative control, we moved on to our tissue of interest: human visceral adipose tissue (VAT) and subcutaneous adipose tissue (SAT). Here, the question was open—does LHCGR show up in fat depots at all?

Using our antibody and a high-sensitivity detection kit, we observed abundant LHCGR staining in adipocytes (fat cells) from both VAT and SAT sections.

IHC tells us that the receptor protein itself is present and where it sits in the tissue. The fact that the signal is in adipocytes (not just surrounding stromal or immune cells) is essential, because adipocytes are the cells that store and release fat. If the receptor is on adipocytes, LH binding to LHCGR can, in principle, directly influence fat cell behavior. Our findings here moves the LHCGR hypothesis from “speculative” to “biologically plausible.”

In short: Immunohistochemistry established protein-level presence of LHCGR in the exact cells that matter for WHR.

Bulk-RNA sequencing

Our next step was to perform bulk-RNA sequencing. Bulk-RNA sequencing essentially can tell which genes are “on” through the amount of mRNA present (since mRNA is produced through the process of transcription by a specific gene. It captures the total mRNA from a collection of cells and then sequences — figures out the exact order of the nucleotides (A, U, C, and G) — that RNA in order to determine which genes were active, or expressed, in those cells. Using high-throughput sequencing machines, a single experiment can capture the expression levels of thousands of genes at once with high accuracy.

RNA-seq experiments generate huge amounts of data, often tens of millions of short sequences. The number of these sequences that match a gene reflects how active that gene is. With the right experimental setup, RNA-seq can identify which genes are expressed when comparing two conditions, such as a treatment group versus normal controls.

If the LHCGR gene is transcribed into RNA in adipose tissue, we will detect it—even across many different donors. This is a powerful cross-check against immunohistochemistry: it asks the same question — is the receptor present? — except at the gene expression level and across large cohorts.

The general steps in RNA sequencing is as follows. I am going to gloss over them, and mention and link some tools that are used in the analysis process. If you are interested just check out their websites to learn more and here is the paper that I followed whilst performing this.

1. Short pieces of RNA sequences are captured during sequencing (reads) and are then analyzed to produce several useful outputs, including lists of genes, transcripts and expression levels for each sample

2. Alignment of the reads to the genome

   1. HISAT2 aligns RNA-seq reads to a genome and discovers transcript splice sites —points at which the introns are removed. Introns are a segment of a DNA or RNA that don’t code for proteins, essentially not useful for this experiment.

3. Assembly of the alignments into full-length transcripts

   1. StringTie assembles/merges the reads as best it can to give you a transcript. StringTie merge compares all of the transcripts to the reference and then aligns all the transcripts to the reference in the best possible way.

   2. The genes and isoforms present in one sample are rarely identical to those present in all other samples, but they need to be assembled in a consistent manner so that they can be compared. We address this problem by merging all assemblies together using StringTie's merge function, which merges all the genes found in any of the samples. Thus, a transcript that was missing an exon in one sample because of a lack of coverage might be restored to its full length.

4. After assembling with StringTie, the full set of assemblies is passed to StringTie's merge function, which merges together all the gene structures found in any of the samples.

   1. This step is required because transcripts in some of the samples might be only partially covered by reads, and as a consequence only partial versions of them will be assembled in the initial StringTie run. The merge step creates a set of transcripts that is consistent across all samples, so that the transcripts can be compared in subsequent steps.

   2. The merged transcripts are then fed back to StringTie one more time so that it can re-estimate the transcript abundances using the merged structures.

5. Quantification of the expression levels of each gene and transcript

6. Calculation of the differences in expression for all genes among the different experimental conditions.

   1. Ballgown takes all the transcripts and abundances from StringTie, groups them by experimental condition and determines which genes and transcripts are differentially expressed between conditions.

We used GTEx (v8)—a canonical public resource with RNA-seq across many tissues and donors. We pulled counts for:

• Subcutaneous adipose (both males and females)

• Visceral adipose (both males and females)

• Positive-control gonadal tissues. Gonadal tissue refers to the gonads which are our primary reproductive organs

• Negative-control tissue (e.g., skeletal muscle)

We normalized the counts — which just mean organizing and structuring data in a dataset to reduce redundancy and make it more consistent and accurate — and ran a differential expression framework (DESeq2) to compare across tissues.

We found three main things:

1. Adipose tissues (VAT and SAT) showed clear LHCGR expression in both sexes.

2. **Gonads (as expected) expressed LHCGR strongly (**our positive control at the RNA level).

3. Skeletal muscle showed absent or negligible expression (our negative control at the RNA level).

Bulk RNA-sequencing validates, at scale, that LHCGR isn’t a one-off artifact in a single biopsy—it’s consistently expressed in adipose depots across many individuals. It is also importantly present in both VAT and SAT across both sexes: male and female. Together with immunohistochemistry, this confirms LHCGR is really there in human fat.

In short: bulk RNA-seq established gene-level expression of LHCGR in human adipose across large cohorts, reinforcing the immunohistochemistry result.

Single-cell RNA sequencing

Bulk RNA-seq is an average across all cells in a tissue chunk. Adipose is a mosaic—mature adipocytes, preadipocytes, endothelial cells, immune cells, fibroblasts. If the goal is to modify specifically adipocytes, we need to know which exact cell types express LHCGR. That’s what single-cell (or single-nucleus) RNA-sequencing answers. Single-cell RNA-sequencing can describe RNA molecules in individual cells with high resolution and on a genomic scale.

• High resolution = You know which cell type each RNA came from. Thus you know gene expression of every single cell because each mRNA is barcoded.

• Genomic scale = Because you are pooling all different cell types that use their genome differently

In plain terms, single-cell RNA sequencing measures gene expression cell-by-cell. After sequencing, cells are grouped into clusters based on their expression profiles (think of clusters as “cell neighborhoods” that share a molecular identity). Known marker genes label each cluster; for adipocytes, a canonical marker is ADIPOQ (adiponectin gene). If LHCGR and ADIPOQ are located in the same cluster, it means mature adipocytes are expressing LHCGR.

We leveraged the Human Adipose Atlas, a high-quality, publicly available single-cell map of human and mouse white adipose tissue. We examined:

• The adipocyte cluster (identified by ADIPOQ enrichment).

• LHCGR expression within that cluster and across sub-clusters

• Whether expression patterns differ by depot (VAT vs SAT) or sex.

We we found:

• LHCGR is enriched in the adipocyte cluster that is marked by ADIPOQ—that is, in fat cells rather than just stromal or immune compartments.

• This holds across depots and sex

Single-cell data nails down cell-type specificity: adipocytes themselves carry the receptor. The sub-cluster enrichment provides a mechanistic clue: LHCGR is not just anywhere in adipocytes—it’s enriched in subtypes that have been associated with leaner or healthier adipose phenotypes. That aligns with our overall therapeutic idea: turning up LHCGR signaling might shift adipose toward these more favorable cell states.

With these three layers together—Immunohistochemistry (protein in adipocytes), bulk RNA-sequencing (robust expression across many donors and both depots/sexes), and single-cell RNA-sequencing (adipocyte and sub-cluster specificity)—we’ve cleared the most basic but essential hurdle: LHCGR is present in the right cells, in the right tissues, across the right populations.

This matters because target validation is the cornerstone of any translational effort. Before you ask, “Does activating this receptor improve WHR?”, you must answer, “Is the receptor even there, and in the cells that control WHR?” Step 1 answers that with a confident yes.

Step 2 — Do human genetics link LHCGR to fat distribution (WHR)?

Presence isn’t proof of importance. To show that LHCGR matters for WHR, we turn to human genetics. The goal of this step is to move from “the receptor is there” to “changes in this receptor’s expression causally track with WHR.” In other words, does this receptor matter for WHR in real people? This is the kind of evidence we need before testing a drug.

To answer this, I turned to human genetics. The strategy was threefold:

1. Genome-Wide Association Studies (GWAS): find common genetic variants near LHCGR that are associated with WHR.

2. Expression Quantitative Trait Loci (eQTL): check whether those same variants change how much LHCGR is expressed in fat tissue.

3. Colocalization analysis: test whether the same variant is responsible for both the WHR association and the expression change.

If all three lined up, it would provide strong evidence that LHCGR isn’t just present in fat but is actually causally involved in shaping fat distribution.

GWAS: scanning millions of people for clues

A Genome-Wide Association Study (GWAS) is a powerful way to connect DNA variation to traits. The idea is simple: look across the genomes of millions of people, identify small differences in their DNA (called single nucleotide polymorphisms, or SNPs), and ask whether any of these differences consistently show up in people with a particular trait—in this case, a higher or lower WHR.

Most SNPs are harmless “spelling differences” in DNA, but some influence how genes are turned on or off. Importantly, GWAS doesn’t assume we know which genes are relevant. It just scans the entire genome, flagging any regions where genetic variation correlates with the trait.

For this project, I used data from the Type 2 Diabetes Knowledge Portal, which aggregates GWAS results from 31 large datasets, covering nearly 4 million individuals. This is the kind of statistical power you need, because the effect of a single SNP on a trait like WHR is usually very small.

Within the LHCGR region, GWAS identified multiple intronic variants (variations in the non-coding regions inside the gene) that were strongly associated with WHR. Intronic variants don’t change the protein structure, but they often act as regulatory switches, influencing how much of the gene is expressed.

One variant in particular—rs62135431—stood out. It was associated with WHR at a genome-wide level of significance (p = 2.06 × 10⁻¹¹). In practical terms, this means the statistical evidence that this SNP is connected to WHR is extraordinarily strong.

eQTL: connecting DNA variants to gene expression

Finding SNPs linked to WHR is suggestive, but the next step was to ask: do these SNPs actually affect how much LHCGR is expressed in fat?

That’s where expression quantitative trait loci (eQTL) analysis comes in. An eQTL is a genetic variant that changes how strongly a gene is expressed (i.e., how much RNA is made from that gene). Think of it as a dimmer switch: some SNPs turn gene expression up, others turn it down.

For this analysis, I again used GTEx, which provides large-scale RNA sequencing data across human tissues, including both SAT and VAT. By combining genotype data (which SNPs someone carries) with expression data (how much RNA their genes make), it’s possible to see whether a SNP like rs62135431 influences LHCGR levels.

In visceral adipose tissue, rs62135431 behaved as a significant eQTL for LHCGR. Individuals with a genetic variant associated with higher WHR had lower expression of LHCGR in their visceral fat. This relationship held even after adjusting for confounders like age, sex, and ancestry.

In simpler terms: the same SNP that predicts a WHR also predicts lower levels of LHCGR in abdominal fat.

Colocalization: are these the same signal?

Here’s the challenge: just because a SNP is associated with WHR and also affects LHCGR expression doesn’t necessarily mean it’s the same causal variant driving both effects. They could be two nearby signals that just happen to overlap.

To disentangle this, I used colocalization analysis. This is a statistical framework that asks: given the GWAS data (SNPs associated with WHR) and the eQTL data (SNPs associated with LHCGR expression), what’s the probability that the same variant explains both?

Colocalization produces posterior probabilities (PP) for different hypotheses. The one we care about is PP.H4, which represents the probability that there is a single shared causal variant. Values above 0.8 are considered strong evidence.

For rs62135431 in visceral adipose tissue, colocalization analysis showed a 99.99% probability (PP.H4 = 0.999) that the WHR association and the LHCGR expression signal are driven by the same causal variant. This is about as strong as the evidence gets: the statistical fingerprint says that the genetic difference which changes WHR also changes LHCGR expression in fat.

Together, these results move the story forward in a big way:

1. GWAS told me that common genetic variation near LHCGR is associated with WHR.

2. eQTL analysis showed that these same variants influence how much LHCGR is expressed in fat.

3. Colocalization confirmed that it’s the same causal variant doing both.

The biological interpretation is straightforward: lower expression of LHCGR in visceral fat causally contributes to a WHR. This is exactly the kind of human genetic evidence we look for when considering a new therapeutic target. It means that if we could find a way to boost LHCGR signaling in adipose tissue, we might be able to shift fat distribution toward a healthier profile.

Step 3 — Does Turning On LHCGR Actually Change Fat?

By this point, I had shown two key things:

1. LHCGR is present in human fat cells, especially in adipocyte subtypes tied to healthier profiles (Step 1).

2. Genetic variation that reduces LHCGR expression increases WHR, suggesting the receptor causally influences fat distribution (Step 2).

The natural next question was: what happens if I flip the receptor “on”? To test this, I turned to a small molecule called Org 43553.

Org 43553 is a synthetic molecule originally developed in reproductive medicine. It binds to LHCGR and activates it—in the same way the body’s own luteinizing hormone would. Unlike hormones like LH, which circulate systemically and can raise sex steroid levels (estrogen, testosterone), Org 43553 is a selective agonist: it activates LHCGR directly without massively altering sex hormone profiles. That makes it ideal for asking the clean question: if I activate LHCGR in fat cells, what happens to fat metabolism itself?

In vitro: testing Org 43553 in fat cells

Before leaping into animals, I started with cell-based systems. These allow precise control over conditions and let me look at fat biology up close.

3T3-L1 cells are a standard mouse cell line used to model fat biology. They start as preadipocytes (immature precursors). With the right cocktail of hormones (insulin, IBMX, rosiglitazone), they turn into mature adipocytes that accumulate lipid droplets, mimicking fat cells in the body. This system is widely used because the cells behave very much like “real” fat cells, and they allow me to test interventions like Org 43553 directly.

I set up this experiment in both a 2D and 3D context to ensure conclusive evidence:

• In 2D monolayers, cells grow flat on a dish, where I can measure fat droplet formation, gene expression, and metabolic changes.

• In 3D spheroid cultures, cells cluster into rounded structures that more closely resemble mini-fat depots. These spheroids allow me to study how adipocytes organize, accumulate fat, and respond to drugs in a 3D context.

There were three specific outputs we were looking at:

• Fat accumulation: measured by staining for perilipin-1 (PLIN1), a protein that coats lipid droplets.

• Cell signaling: checked for rapid activation of ERK1/2 phosphorylation, a known signaling cascade triggered by LHCGR.

• Energy expenditure: measured oxygen consumption rates (OCR) using a metabolic analyzer, which tells me how much energy cells are burning.

When treated with Org 43553, 3T3-L1 adipocytes showed reduced lipid droplet accumulation. In other words, they stored less fat. ERK1/2 signaling was rapidly activated after Org 43553 exposure, confirming the receptor was being engaged and downstream pathways switched on. Org 43553 also increased oxygen consumption rates, especially after oligomycin challenge (which forces cells to reveal their “spare” energy-burning capacity). This indicated enhanced thermogenesis—the process of burning energy as heat rather than storing it as fat.

These cell-level results showed that activating LHCGR can:

• Directly alter adipocyte metabolism, making them less likely to store fat.

• Switch on energy-burning pathways, hinting at a mechanism for increasing whole-body energy expenditure.

Importantly, these are direct effects in fat cells, not indirect effects through sex hormones. That distinction is critical if we’re thinking about a therapeutic angle.

In vivo testing in mice

Of course, cell cultures can only tell part of the story. The real question is whether activating LHCGR changes fat distribution in a living organism. For this, I moved to mice.

Male C57BL/6J mice, a standard laboratory strain, were fed a high-fat diet to induce obesity. At 14 weeks old, these mice were randomized to receive either Org 43553 injections or a vehicle control, for nine weeks. This setup mimics the metabolic stress of obesity and lets me see whether Org 43553 can alter fat accumulation under these conditions.

Here were the specific outputs we were looking for:

• Total fat mass: measured with quantitative nuclear magnetic resonance (qNMR), which provides a precise readout of lean vs. fat mass.

• Adipose tissue weights: dissected and weighed VAT and SAT after treatment.

• Histology: VAT sections were fixed, stained (H&E), and analyzed with digital pathology software to quantify adipocyte size distributions.

Here are the results:

• Total body fat mass was significantly reduced in Org 43553–treated mice compared to controls.

• Visceral adipose tissue (VAT), in particular, was diminished, while subcutaneous fat was relatively preserved. This suggests a depot-specific effect, consistent with shifting fat away from the waist.

• Vehicle-treated mice had large, hypertrophic adipocytes—classic hallmarks of metabolically unhealthy fat. Org 43553–treated mice had smaller adipocytes, more closely resembling the profile seen in leaner animals and in humans with healthier WHRs.

These in vivo results provide the first functional evidence that activating LHCGR can remodel fat distribution in a whole organism. The reduction in visceral fat, preservation of subcutaneous fat, and shift toward smaller adipocyte sizes all align with a healthier WHR profile. In essence, Org 43553 treatment made obese mice “carry” their fat more like metabolically healthier humans do—fewer large abdominal fat cells, more balanced fat distribution, and enhanced energy burning.

With these experiments, I moved from association (Steps 1 & 2) to causation: activating LHCGR directly changes fat metabolism and fat distribution.

• In cells, Org 43553 reduced fat storage and increased energy expenditure.

• In mice, Org 43553 reduced visceral fat, shifted adipocyte size distributions, and produced histological profiles matching healthier WHR patterns in humans.

This is exactly the kind of evidence needed to nominate LHCGR as a therapeutic target for obesity—not just for weight loss, but specifically for improving waist-to-hip ratio, the most predictive marker of long-term health risk.

Summary

When I began this work, the guiding question was simple but ambitious: could LHCGR, traditionally viewed as a reproductive receptor, also be a regulator of fat distribution? Answering that required moving step by step, layering evidence from different angles until a clear picture emerged.

Step 1 — The receptor is present in fat

First, I needed to prove LHCGR actually exists in human adipose tissue. Using immunohistochemistry, bulk RNA-seq, and single-cell RNA-seq, I established that LHCGR is expressed in mature adipocytes, not just accessory stromal cells. Even more compelling, LHCGR expression was enriched in adipocyte subtypes associated with lower BMI and healthier metabolic states. This suggested that the receptor is not just present but potentially functionally relevant.

Step 2 — Human genetics link LHCGR to WHR

Next, I turned to genetics. Through a meta-analysis of nearly 4 million individuals, GWAS revealed that common variants near LHCGR are strongly associated with WHR. The lead SNP, rs62135431, was then shown by eQTL analysis to reduce LHCGR expression in visceral fat. Colocalization analysis sealed the case: the same variant that lowers receptor expression is the one that raises WHR, with a posterior probability of 99.99%.

Step 3 — Activating LHCGR changes fat biology

Finally, I asked whether LHCGR could be functionally manipulated. By using Org 43553, a selective agonist, I showed that turning on the receptor changes fat biology. In adipocytes, Org 43553 reduced lipid storage and boosted energy expenditure. In mice, it selectively reduced visceral fat, preserved subcutaneous fat, and remodeled adipocyte size toward a profile resembling humans with lower WHR. These changes aligned perfectly with the genetic evidence: higher receptor activity favors healthier fat distribution.

Taken together, these steps form a cohesive argument.

1. The receptor is present in the right cells.

2. Genetic variation in its expression alters WHR.

3. Pharmacological activation shifts fat distribution in the predicted direction.

Thank you for reading and I really hope you enjoyed this article. If you want to look at the paper, which is admittedly very jargon-heavy, it is here. Finally, here is also a link to my poster that synthesizes my work in a different format, though it is also a bit confusing.

Another thing that is in common with all of these things, they all make contact with our body and thus interact with our immune system. A super important question for all of these products is how can they evade (in the case of contact lenses, heart valves, and breast implants) or activate (in the case of vaccines) the immune system. These questions fall under the realm of immunology, a branch of biology and medicine that covers the study of immune systems in all organisms.

These two topics, immunology, and biomaterials, were the subject matter of my conversation with Jessica Stelzel, a Ph.D. student at Johns Hopkins University (JHU).

Background

Jessica attended Georgia Tech for her undergraduate studies with a passion for engineering, although she was unsure of which major to pursue. This is a common theme, as most people are usually uncertain about their exact path in college, and even those who think they know often change their minds. Jessica chose materials science because she became interested in how we can understand the properties of materials based on their structures.

One thing that Jessica was sure about was her desire to pursue a Ph.D. and remain in academia. This ambition brought her to Johns Hopkins, where she found the PIs to be particularly amazing. PI, or principal investigator, is a faculty member or research scientist appointed by the university to conduct research. At JHU, Jessica aimed to investigate how the immune system responds to biomaterials.

Jessica’s reasoning for pursuing academia was the autonomy it provides, allowing her to feel in control of her daily activities. Unlike industry research, which must be justified by profit or commercial potential, academia offers the freedom to explore any scientific question, provided you can secure funding through grants. This freedom enables more early-stage research, which may not be profitable now but could be crucial in the future.

More often than not, preclinical work happens in academia. To get a drug to market, several steps must be completed in both pre-clinical research and clinical trials:

1. Cell cultures: The first step, called in vitro research, involves testing your drug on cells in a petri dish, test tube, or labware.

2. Mice: If cell cultures prove successful, the next step is to test the drug in mice. Mice are used because they are genetically and biologically similar to humans, are small, easy to keep, and can be bred in large numbers.

3. Monkeys: Monkeys are also sometimes used because they are even more genetically similar to humans.

4. Humans: If everything else is successful, clinical trials on humans can begin, which are broken into four phases:

   1. Phase I trials: Involves 20 to 100 healthy volunteers with the disease/condition, lasting several months. Approximately 70% of drugs move to the next phase.

   2. Phase II trials: Includes up to several hundred people with the disease/condition, lasting several months to 2 years. Approximately 33% of drugs move to the next phase.

   3. Phase III trials: Involves 300 to 3,000 people, lasting up to 4 years. Around 25%-30% of drugs move to the next phase.

   4. Phase IV trials: Includes several thousand people, lasting several years.

Here is a great overview of all the steps to bring a drug to market (Source).

In the end, only one in ten drugs make it to patients, and the cost of bringing a drug to market is a ridiculous $1.3 billion. Clearly, there is a need for a more streamlined approach to identify failing drugs early, preventing wasted time and money. Moreover, there are significant differences between models, and optimizing a drug for one model doesn't guarantee it will work well in others, requiring repeated adjustments. What if we could replicate a human’s organs without actually using a human?

Introducing organ-on-a-chip, a multi-channel 3-D microfluidic chip that mimics the functions and responses of a whole organ or organ system. It’s not a functional organ but a combination of cell culture with microfluidics to emulate them. These models can simulate the human environment without the ethical concerns of using animals or humans, potentially allowing us to skip steps like mice and monkeys in the pre-clinical phase.

The FDA has already approved organ-on-a-chip technology. The Wyss Institute, a research and development institute at Harvard, already uses this technology to conduct research on a wide range of human diseases, including COVID-19, influenza, malnutrition, radiation exposure, and cystic fibrosis. However, organ-on-a-chip models are still new, and only recently have companies emerged in this space; Emulate is one of the leaders, actually providing organ-on-a-chip models for the Wyss Institute.

In addition to her research, Jessica co-taught a four-week undergraduate course at JHU focused on immunology and biomaterials. She taught the first half on immunology, while her labmate taught the part on biomaterials. The class explored questions such as the materials used for different medical applications, the meaning of biocompatibility, and the definition of the foreign body response. Jessica chose to develop and teach this course to see if she enjoyed teaching, as it is an option to pursue after her Ph.D; she in fact did enjoy this experience and it will definitely inform her future career decisions.

Jessica’s research - silicone breast implants

The last part of our conversation involved discussing Jessica’s research projects, both the ones she helped with and led herself. The first project focused on silicone breast implants and how the roughness of the implant’s surface influences the immune system’s foreign body response. This project was led by Jessica’s professor, Dr. Joshua Doloff. 

Silicon, a material used in devices approved by the FDA, is generally considered biocompatible. Yet, the team found that even a material touted for its safety can elicit an immune response if the implant has a rough texture. Highly textured implants triggered a significant cytotoxic T cell response and even have been observed to lead to T cell-associated cancer (Breast Implant-Associated Anaplastic Large Cell Lymphoma (BIA-ALCL)). Cytotoxic T cells are a type of white blood cell that help the immune system fight germs and protect the body from disease.

Comparing textured implants with those having little texture (4 μm roughness being ideal), the team noticed more regulatory T cells (which calm cytotoxic T cells) around less textured implants. Conversely, highly textured implants lacked these calming regulatory T cells, resulting in a stronger immune response. Just by changing the texture, the immune response can be dramatically altered.

The lab studied the responses in mice and rabbits to both smooth and textured implants. Roughness was evaluated through techniques such as scanning electron microscopy and profilometry. Jessica specifically helped by using flow cytometry to confirm the presence or absence of fibroblasts. A major issue with these implants are processes called fibrosis and capsular contracture. In fibrosis, fibroblasts deposit collagen around the implants to wall them off. This is essentially the body’s way of saying that it doesn’t like whatever entered. Then, in capsular contracture, those collagen fibers shrink and squeeze the implant, which can cause medical complications. The team found that there were more activated fibroblasts in samples that had more surface texturing.

This research is incredibly significant beyond just breast implants, showing that even generally biocompatible materials can become unsafe and even carcinogenic (causing cancer) if certain properties, like surface texture, are altered. We need to be more cautious of generalizations as even materials that are generally considered biocompatible can be harmful in certain circumstances, as this study perfectly demonstrated. 

Jessica’s research - nanofiber hydrogel composite

Nanofiber hydrogel composite is quite a mouthful so let’s break it down. The core component is the hydrogel, a jello-like polymer that can hold a lot of water and is chemically or physically crosslinked into a mesh. Hydrogels hold a lot of promise in the field of drug delivery, tissue engineering, and regenerative medicine because they are highly customizable and can be very easily injected with a needle. Hydrogels are customizable because of that crosslinking process, which we have only gotten better at doing as we become more familiar with the chemistry involved.

A nanofiber is a fiber, which is a material resembling a hair-like strand, on the nanoscale (a nanometer is a billionth of a meter). The composite means that you are combining two different types of materials, in this case the nanofiber with the hyaluronic acid, to create something new. Jessica’s composite was one that her lab already made, which involved crosslinking hyaluronic acid with a polycaprolactone (PCL) nanofiber.

Now I know I used the term crosslinking a lot without properly defining it. Crosslinking is the process of chemically joining two or more molecules by a covalent or ionic bond. Polymer chains on their own are sort of loose, but crosslinking intertwines those chains, making them more stable because they can’t move independently of each other. Think of a bunch of spaghetti, when it’s not intertwined, it is very loose and malleable; however, if you start attaching one piece of spaghetti to another, all of a sudden it can’t flow so easily and it is far more stiff and solid.

In the nanofiber hydrogel composite materials Jessica studies, the crosslinked hyaluronic acid hydrogel has additional covalent bonds with the PCL nanofibers because the nanofibers give the hydrogel a lot more stiffness while still maintaining large pores in the hydrogel. The size of the pores is an important parameter as pores are the holes in the crosslinked mesh that cells can pass through. To preface, Jessica is studying the hydrogel composite in the field of regenerative medicine, which involves using materials to replace or "regenerate" human cells, tissues, or organs to restore or establish normal function. Regenerative medicine involves creating materials that could act as a scaffold and cause immune infiltration into the material and then cause the material to be remodeled into something that looks like human tissue. Going back to why pores are so important, large pores are necessary for even cell distribution and connection throughout the engineered tissues and helps in the diffusion of nutrients and oxygen. These pores function somewhat similar to veins for a material that lacks a vascular system. Pore size also influences cell growth and tissue thickness, with there being optimal sizes for different types of tissue regeneration. For instance, skin regeneration requires pores of 20-125 μm, while bone regeneration needs 100-350 μm. Here is a good paper on this if you want to read more.

On a brief side tangent, a really exciting application of hydrogels is their use as a depot for drug delivery. A depot means that you can put the hydrogel under your skin and it can remain there for a while, slowly releasing the drug in the process. You can engineer the hydrogel through crosslinking to degrade easily or you can engineer them to be more solid and last longer. Often the drug is mixed in with the hydrogel, so as the hydrogel degrades so does the drug. However, the drug can also be directly attached to the hydrogel . Either way, the hydrogel degrades when enzymes travel through holes in the mesh and break down the crosslinks that connect the chains. Depending on how small you make the holes in the mesh, you can control where the cells and enzymes pass through and how quickly they do so. For example, you can make the holes on the inside of the hydrogel bigger so that it is easier for stuff to pass through the middle, meaning that the hydrogel will degrade from the inside. This amount of programmability with the hydrogel has made it so people have placed one drug on the inside and one in the outer layer, allowing for combinatorial therapy.

Jessica is tackling two essential questions surrounding this material: what immune cells are most pivotal towards causing tissue remodeling and what is the necessary amount of inflammation to produce wound healing properties? Jessica studied the microenvironment that the immune system created around it and initially noticed that there were inflammatory signals after the injection of the composite. This makes sense because the PCL fibers are known to be very inflammatory, which is super important because you can’t get tissue regeneration or remodeling without inflammation. While inflammation is often seen as a bad thing and something that we try to avoid, we have to remember that inflammation just means an immune response. If you don’t have inflammation, you won’t have any immune response and thus none of that wound healing process. Thus, inflammation should be seen as a positive thing in this context.

In order to find which immune cells are most pivotal towards tissue remodeling, Jessica knocked out, or in other words, eliminated immune cell  types one at a time, waiting to see which immune cell when removed would cause the biggest decrease in remodeling. 

The second question that Jessica tackled was inspired by the lab’s findings that when this composite was injected into animal tissue, it caused a bunch of blood vessel development, and the inside of the material got remodeled into soft adipose (fat) tissue. As a result, Jessica is trying to figure out the necessary amount of inflammation to get those wound healing properties: blood vessel development and the presence of soft adipose tissue. She is studying the cells that the material recruited and their function in this context. If she finds which cells are most vital for those wound healing properties, she can try to target the recruitment of only those cells, skewing the immune response towards those wound healing effects.

What is interesting about the hydrogel’s structure if you look back to that image is how it looks disconnected. This is because it was purposefully fragmented. The hydrogel was passed through a mesh screen so it can then be easily injected with a needle. Once it enters the body, the material forms back together into a cohesive structure due to the pressure in our body. This material and delivery method was developed with a Johns Hopkins plastic surgeon. He provides valuable input on whether their research would be adopted in the clinic and is a great example of the communication between the lab and the clinic. A big focus of the lab is how they can get their work translated, which is why they have a lot of clinical collaborators.

Conclusion

Wrapping up our conversation, Jessica mentioned her plans to become a professor after her amazing experience teaching a class. It really goes to show that you never know how much an experience can influence your trajectory. If you are presented with an opportunity to do something, always seize it. The worst that can happen is that you do the experience and find out that it was a waste of time; however, if you pass on an opportunity, who knows if that opportunity could have been a life-changing moment for you.

Jessica’s experience also definitely portrays the amazing parts of science research in academia. You are given far more freedom in the questions that you can explore and you are motivated by your passion for science rather than profit. It would be silly however to neglect the advantages of industry in that your research is far more likely to make contact with people. Academic research is largely focused on advancing the field of knowledge and doing cutting-edge research while industry research is often applying science to solve problems at the current moment. Regardless of where you go, having a strong understanding of the fundamentals and having those intangibles like problem-solving and resilience is absolutely essential. With those talents, you will succeed anywhere you go - industry, academia, or something totally separate from science.

Here are some additional sources besides that one that I already linked

• https://www.fda.gov/patients/drug-development-process/step-3-clinical-research

• https://en.wikipedia.org/wiki/Cost_of_drug_development#:~:text=A new study in 2020,drug development as %242.8 billion.

• https://my.clevelandclinic.org/health/body/24630-t-cells

In my previous article, I discussed the problem of protein structure prediction and the impact of AlphaFold2 in solving said problems. However, there are still two others aspects of the protein field.

1. The question of what balance of interatomic forces or bonds dictate the structure of the protein

2. The kinetics question of what routes or pathways some proteins use to fold

This article will cover my project, where I combined the technologies of FoldX with Yasara to study how proteins interacts with other molecules and how mutations affect a proteins structure, thus tackling mainly the second problem. Through this project, I gained a better understanding of what causes a protein to misfold, and how improper folding results in a protein not achieving its native structure.

Furthermore, while it is not at the core of the protein folding problem, I also got a better understanding of how a protein interacts with peptides and DNA. This is key because proteins have emerged as promising drug delivery vehicles and are often the targets for many therapeutics. Thus, understanding how they interact with molecules is necessary to ensure the efficacy and safety of your therapeutic.

Lets take a step back though. Why is it important to study these problems? To understand this, lets understand the history of the protein folding field.

The History of Protein Folding

Protein folding is the process by which a protein assumes its characteristic structure, known as the native state. The fundamental question in this field is how an amino acid sequence both specifies a native structure and a pathway to attain said structure. To boil it down even further, the protein folding problem is the question of how order (a native structure) arose from disorder (an on-rigid structure) in proteins.

From this overarching question, there are three main subquestions that have largely been answered from theory and statistical mechanics combined with experiments:

1. How did the various native structures of proteins arise from interatomic driving forces encoded within their amino acid sequences?

   1. The answer to this is that chain randomness is overcome by solvation-based codes. In other words, although the linear sequence of amino acids in a protein chain may initially lack specific structure (chain randomness), solvation-based codes, which involve complex interactions between the protein chain and surrounding solvent molecules, play a crucial role in overcoming this randomness and guiding the protein towards its correct folded structure.

2. How does protein folding occur so fast given that a protein has to search through countless possible confirmations (structures) before finding its native structure?

   1. To answer this, we use the needle-in-a-haystack metaphor, where native states are found efficiently because protein haystacks (conformational ensembles) are funnel-shaped. In other words, as the protein gradually folds, the amount of potential structures exponential decreases to the point that the protein only has a few possible native structure and can easily find the needle (native structure) in the large amount of possible protein structures (haystack).

3. Is there a folding mechanism, a particular route or pathway that is both specific enough to explain how a given amino-acid sequence reaches its native state, and general enough to say why most proteins have such routes?

   1. In other words, is there a mechanism or narrative of a protein's folding pathway that is general enough to apply to all proteins, yet still granular to account for the differences across proteins. This problem is super tricky and still unsolved because it can't be an atom-by-atom nanosecond-by-nanosecond sequence of events, for that would surely apply only to one protein. However, the funnel concept alone is also not a sufficient descriptor of the mechanism because it is too generic to tell us differences in folding routes of different proteins.

A "Brief" Chronology

The notion of a folding "problem" first emerged in 1957, with the first experimental determinations of globular protein structures by Perutz et al. Here, John Kendrew and Max Perutz created the first atomic structure of the protein myoglobin, showing that proteins had atomistically detailed structures that were different for different proteins, and that those differences matter for function. The last part is super important and is at the core of the idea that a proteins structure determines its function.

A key milestone was the work of Dr. Christian Anfinsen on ribonuclease (RNase). In 1961, Dr Anfinsen's hypothesis was that the native structure of a protein is the thermodynamically stable structure; it depends only on the amino acid sequence and on the conditions of solution, and not on the kinetic folding route or sequence of steps by which a protein folds. In other words, the information required to fold a protein into its three-dimensional shape, or the folding code, comes from its sequence of amino acids.

This work was absolutely key for the field, and he was later recognized for the Noble Prize in Chemistry in 1972 for his work. Two powerful conclusions followed from Dr. Anfinsen's work:

1. Now that we know that it doesn't matter where the protein folds, the research enterprise of in vitro protein folding began, meaning that we can understand how proteins fold and their native structures through experiments inside test tubes rather than inside cells.

2. The Anfinsen principle suggests that while evolution can modify the sequence of amino acids in a protein, the actual folding of that protein into its proper shape and the speed at which it happens are determined by the laws of physics and chemistry.

Following Anfinsen's work, in vitro studies showed that the folding process typically occurs on a milliseconds-to-seconds time scale, much faster than the rate estimated assuming that folding proceeds by a random search of all possible conformations.

Based upon this observation, in 1968, Cyrus Levinthal proposed that a random conformation search does not occur in folding and that proteins fold by specific 'folding pathways.' On these pathways, the protein molecule passes through well-defined partially-structured intermediate states. In other words, the protein may start to form secondary structural elements such as alpha helices or beta sheets, but the overall structure is not fully formed and these states only exist temporarily during the folding process. Each intermediate state is a step closer to the final folded structure.

Levinthal thought this because given the knowledge that proteins fold super fast, it would take forever for a protein to search through the estimated 10^300 possible structures to find the native state. Out of a near infinite possible ways to fold, a protein thus picks one in just tens of microseconds; this same task would take 30 years of compute time.

The implications of this study was that even though folding is stochastic, it must still involve some physical basis for not exploring the vastness of entire conformational spaces. It was later understood that the chain randomly formed increasingly many hydrophobic-hydrophobic (HH) contacts, lowering the free energy and at the same time reducing the remaining space of searchable conformations. Folding is not a random search, but rather pruned by exponentially diminishing the number of conformations available to search as the protein becomes increasingly more compact.

Following these key experiments, one of the earliest proposed mechanisms for protein folding, the nucleation-condensation model, was formulated. This model was based on nucleation theory, which suggests that when a protein folds, the process begins with the formation of a stable nucleus, which serves as the starting point for the folding of the protein molecule into its native conformation. This initial formation of a nucleus could be then considered the rate-limiting step in the folding process. Based on this idea, the model required a small number of residues (folding nucleus) to form their native contacts in order for the folding reaction to proceed fast into the native state.

Furthermore, the cooperative nature of the protein folding is similar to the behavior exhibited in first-order phase transitions, which happen via formation of nuclei (nucleation) followed by their growth. Because of these similarities, terminology used in studies of phase transitions, such as energy landscapes and nucleation, was introduced into the discussion of protein folding.

The concept of energy landscape was key for the protein folding field. A key conclusion now is that proteins have funnel-shaped energy landscapes: many high-energy states and few low-energy states.

A simple chemical reaction goes from reactants A to product B. A protein cannot just go from its unfolded state straight to its folded state because its reactant, the denatured or unfolded state, is not a single microscopic structure. It's a state of disorder where the protein's amino acids are not organized. This goes back to the idea that folding is a transition from disorder to order, not from one structure to another. Simple one-dimensional reaction path diagrams do not capture the reduction in conformational degeneracy. A funnel, on the other hand, accurately portrays a proteins conformational heterogeneity, or just the existence of multiple possible structures that a protein can adopt. As a protein folds, the amount of possible structures decreases and the funnel becomes narrower to represent that.

Now I know I stated before that protein folding can be done in vitro because the folding code lies in the amino acid sequence, there are a few key differences between in vitro and in vivo studies that still makes it important to study protein folding in vivo.

The first difference is that protein folding is usually assisted by molecular machinery, such as chaperones and molecular cofactors. Molecular chaperones, such as the heat shock protein Hsp70, often help in the protein folding process. One way that they do this is by binding to the exposed hydrophobic regions on the protein surface, shielding them from other proteins and allowing the protein to fold without interference from other molecules. They essentially create a specialized microenvironment that promotes efficient and accurate protein folding.

Furthermore, nearly one third of all proteins in living cells interact with small molecule cofactors. The work of Wittung-Stafshede et al. was key to revealing the role of cofactors, showing that bound metals stabilize the native fold which means that cofactor binding to unfolded proteins dramatically accelerates folding time.

The second big difference is that the concentrations of macromolecular solutes in cells can be hundreds of grams per liter in cells, but most in vitro studies are performed in buffered solution with less than 1% of the concentration. The crowded environment in vivo means that their is less space for the amino acids chains to move and fold without the bumping into other molecules. Furthermore, a more crowded environment means that proteins have the risk of interacting with other molecules, altering the folding pathway and potentially leading to different conformations.

So… Why Does This Matter?

Now that we have a general idea of the field. Let's circle back to the initial question: why is studying protein folding and proteins important in the first place?

Proteins are vital in the drug development and drug discovery space. They have not only shown promise as delivery vehicles, but even as therapeutics. Protein can be genetically engineered to bind to specific targets with high affinity and selectivity, meaning that there is a low chance that they bind to the wrong molecule and cause unwanted side-effects. Moreover, proteins are have lower immunogenicity, which is important if you want to do numerous dosages. A specific application that is of interest to me is their ability to revert misfolded proteins. Proteins could be potentially tailor made to bind to misfolded proteins and fufill the missing part of their structure, restoring their function. If this were to become a reality, neurodegenerative diseases such as Alzheimer's and Parkinson's would be a thing of the past.

Just thinking generally however, among the classes of molecules, proteins occupy a niche of special importance: Not only are they components of all living things and involved in every major cellular processes, and not only do they matter for health and disease as discussed, but their material powers are extraordinary: they are working components of being that walk (function/interact), talk (send signals and communicate), and think (decide how they want to fold and which structure they will adopt). Proteins are truly a special molecule and better understanding them will unlock the secrets to life itself.

What is FoldX and Yasara?

Now that we understand why this problem is important, lets discuss what I worked on.

The first tool that I used was Yasara, which is a computer program for molecular visualizing, modeling, and dynamics. It allows me to view protein structures on my computer and interact with them. I used the Yasara model, which came free using an academic license.

The features in this included everything from the Yasara View package along with some other cool features:

1. Analyze contacts, hydrogen bonds, hydrophobic/pi-pi/cation-pi interactions and protein secondary structure.

2. Align small molecules like ligands automatically, by superposing them on the largest common fragment.

3. Measure distances, angles and dihedrals between groups of atoms like helices or planar side-chains.

4. Show and calculate partial Van der Waals, molecular and solvent accessible surfaces, distinguish between outer and inner surfaces, calculate volumes. Analyze and show cavities formed by these surfaces.

5. Identify cis-peptide bonds and wrong stereoisomers. Stereoisomers is where molecules have the same molecular formula and sequence of bonded atoms, but differ in the three-dimensional orientations of their atoms in space.

While Yasara is cool, FoldX is the really cool part of the project. Centered around the laboratory of Luis Serrano at the European Molecular Biology Laboratory in Heidelberg and at Center for Genomic Regulation in Barcelona, FoldX has constructed the FoldX suite, a package of all of FoldX's biggest technologies.

Released in 2019, FoldX's newest FoldX5 suite contains all of the latest technology:

FoldX

FoldX provides fast estimations of how important a certain interaction (disulphide bonds, hydrophobic, electrostatic, polar and hydrogen bonding) is to the stability of the protein. FoldX uses a detailed atomic structure description of proteins and incorporates different energy factors, which are adjusted based on real-world protein engineering data. Its energy calculation is computationally inexpensive, making it suitable for protein design and predicting protein structure and folding pathways quickly and accurately.

For its energy function, FoldX uses the following equation to calculate the free energy of unfolding (ΔG) of a target protein:

Here, "a - l" are relative weights of the different energy terms in calculating free energy. The proteins interaction with a solvent is treated in two steps. First, FoldX considers the overall liquid environment as a desolvation factor, which changes as atoms become more buried or shielded from the surrounding environment. This desolvation factor is divided into two parts: one related to non-polar groups (ΔGsolvH) and and another to polar groups (ΔGsolvP). Those water molecules that make more than two hydrogen bonds with the protein are calculated explicitly in the ΔGwb term. Van der Waals terms (ΔGvdw) are calculated similarly to desolvation, but experimental energy transfers from water to vapor are also included.

Hydrogen bonds are figured out using basic geometric rules, and their energy (ΔGhbond) is determined from protein engineering experiments involving double mutant cycles. Double mutant cycles are super interesting and are a type of technique where two amino acid residues within a protein are first individually mutated to different amino acids, and the effects of these mutations on the protein's stability are measured. Then, a double mutant is created where both residues are mutated simultaneously. By comparing the effects of the single mutations to the double mutation, we can infer the energetic contribution of the interaction between the two residues to the overall stability of the protein.

The electrostatic contribution to the free energy (ΔGel) is determined using Coulomb's law: the force between two charged objects is directly proportional to the product of their charges and inversely proportional to the square of the distance between them.

For protein complexes, or two or more protein molecules that come together, there's an extra electrostatic contribution calculated between atoms from different polypeptide chains, denoted as ΔGkon.

What differentiates FoldX from other force fields -  which are just sets of mathematical equations to describe the interactions between atoms and molecules - lies in how they estimate entropy. While traditional methods involve simulations to capture the flexibility of protein structures, FoldX calculates the entropy pentaly for fixing the protein backbone in a specific conformation (ΔSmc) by analyzing the distribution of amino acids seen in crystal structures. This penalty is then adjusted based on factors like accessibility (how easily other atoms can interact with a specific part of the protein structure) of backbone atoms and energetics (energy changes) of hydrogen bonds. The entropy cost (ΔSsc) of locking a side chain into a specific shape is determined by adjusting a predefined set of entropy parameters, calculated by Abagayan and colleagues, according to how buried the side chain is.

Additionally, FoldX considers steric clashes between atoms in the structure to calculate a term called ΔGclash. Steric clashes are just when atoms in a molecular structure come too close to each other and their electron clouds repel each other, resulting in strain or distortion of the molecule's structure as atoms cannot occupy the same space simultaneously.

Finally, FoldX can mutate the 20 natural aminocids. When mutating DNA, the code finds the matching base pair and changes both at once to maintain pairing.

LoopX

LoopX allows for quick and precise predictions of the structure of protein loops  - regions which connect two secondary structures. It uses a library, LoopXDB of irregular structure elements grouped based on the distance between the regular (well-defined recurring) parts that surround the loop. This tool also helps researchers swap loops with others from the library, aiding in protein engineering by finding possible variations.

PepX

PepX allows for fast and accurate prediction of peptide docking by means of BackXDB, a library of protein building blocks, and PepXDB, a library of protein-peptide binding motifs.

The PepX algorithm takes a peptide sequence and a domain structure and finds the possible shapes the peptide can take when it binds to other molecules. It does this by using interaction rules with parts of the protein known to be important for its binding. To speed things up, it uses a shortcut algorithm (CSP) to narrow down the search among the many possible shapes stored in the PepXDB database. This database holds over 7 million interactions from various protein structures, giving a wide range of peptide-binding information.

BackX

While this module is currently under development, BackX models backbone moves. This is important because protein backbones are flexible and move, allowing them to adjust to fit nicely with other proteins.

DnaX

Over the last 25 years, there has been noticeable progress in terms of increasing the number of protein-DNA crystal structures stored in the PDB (Protein Data Bank). DnaX is designed to predict how DNA binds to proteins by analyzing these structures. The team has analyzed over 2500 protein-DNA crystal structures and organized the findings into a database based on their structural characteristics.

After that rather long explanation, you can see the real power of these tools and all that they can do. They really are amazing and the fact that they are available for anyone to use is even better.

Showcase

Now let's get to the fun part, here I will my results using these tools.

Ligand Docking

One of the things that I did was analyze ligand docking. Here is one of the proteins that I analyzed, I got this protein by searching through the Protein Data Bank and this was a COX-1 structure found in humans. I then easily imported it onto Yasara and selected one of the many types of ways that I could view this molecule. What is really cool about Yasara is that you can see the molecule in many different ways. For example, this is the ribbon structure of with side chains. A-helices are shown as coiled ribbons or thick tubes, β-sheets as arrows, and non-repetitive coils or loops as lines or thin tubes.

Here is an example of another way that I can model this protein. This model is called the ball and stick model because all it shows are the balls (atoms) and sticks (connections between them). While this model doesn't give you a very good idea of where the helices and sheets are, or the full shape of the protein, it is simple and is an example of how customizable this tool is.

The first thing that I did was I studied ligand binding onto this protein. I first "cleaned" up the molecule by removing extra cofactors and ligands (those large red and blue balls that appear above), and also not showing any side chains, as all of these things are not needed when performing docking studies. Now, my protein looks like this:

Now I want to find my ligand. I went on pubchem and I wanted to see how aspirin would bind to COX1, which is my receptor.

Now, running the simulation, because I did not specify a specific place where the ligand should bind, I got a bunch of possible places where aspirin (all of the large red, blue, and grey circles) could bind to COX1.

While there are many places where aspirin could bind, we need to find the best place. The best one is the one with the highest binding energy in kcal/mol. Once you run the docking analysis, you will get a file that shows an analysis of all of the results and you can look there to find the best one. Here is an image of my best structure, you can see the aspirin ligand in the middle of the protein:

Predicting the Effect of Mutations

Mutations in proteins can have various origins. Some possibilities include errors during DNA replication, exposure to radiation or certain chemicals, or as a result of genetic recombination processes. Mutations can have varying effects on a proteins structure and/or function. For example, mutations can have no effect at all, be stabilizing or destabilizing; in the last two cases, these types of mutations can lead to diseases.

Traditionally, mutations are made in the wet lab to study the effect of a single residue position on protein stability, interaction with peptide ligand, and much more. However, such procedures in the wet lab are laborious and costly. With the introduction of bioinformatic tools such as FoldX, this process is now a lot easier. You can first use these tools to predict the effects of mutations and then test the really interesting ones in the lab.

In terms of how FoldX interprets mutations, the main focus is on the prediction of free energy changes (what happens to the free energy of the protein when we mutate an Asp to a Tyr for example). FoldX will then calculate the free energy of the wild type (WT) and the mutant (MT) and calculate the difference:

ddG (change) = dG(MT) - dG (WT)

As a rule of thumb:

1. ddG(change) > 0 : the mutation is destabilizing (weaken interactions within the protein)

2. ddG(change) < 0 : the mutation is stabilizing (strengthen interactions within the protein)

The protein that I am working with now is called 2AC0, which is a part of a tetrameric complex of the transcription factor P53 bound to DNA. After repairing or minimizing the structure, here is what my protein looks like:

I then used FoldX to mutate the ALA 159 residue to Trp. I also selected the option to show the Van der Waals (VdW) clashes in WT and mutant.

Now here is a view on the mutated Ala159Trp region, notice how there are a lot more red Van der Waals clashes. Remember how, anything above 0kcal/mol is assumed to be destabilizing. When I analyzed it I saw an energy change of +29 kcal/mol!

Conclusion

While I could show a lot more, I recommend that you just download the tools for yourself and try them out. This project was amazing and working with all of these tools was truly a gift. Hopefully, you also see the amazing power of these tools and all that they are capable of. This is honestly a huge testament to the scientific community and what they are able to produce and willing to make available.

Sources

As always, the information that I presented in this article did not come from thin air, here are the key resources that I used to inform my learning:

1. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2173875/

2. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2443096/

3. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8547331/

4. https://foldxsuite.crg.eu/products

5. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1160148/

If I know the sequence of amino acids, how do I figure out the three-dimensional structure and as a result the function of a protein? The concept of a protein folding problem has been around since 1960, with the appearance of the first atomic-resolution protein structures.

The release of AlphaFold 2 three years ago marked a massive step in solving the protein folding problem, determining the structures of around 200 million proteins. We’re going to examine AlphaFold 2, how it was able to achieve such spectacular results, and a particular implementation of AlphaFold 2 — ColabFold — that I used to predict the structures of proteins. I used ColabFold mainly because it was less memory intensive as compared to AlphaFold 2, which required over 3 TB of storage to run. Besides being less memory intensive, ColabFold is virtually the same thing as AlphaFold 2, except that it combines AlphaFold 2 with the fast homology search of MMseqs2 (Many-against-Many sequence searching). MMseq2 is a software that performs fast and deep clustering and searching of large datasets. Similarity clustering reduces the redundancy of sequence databases, improving speed and sensitivity of iterative searches. MMseqs can cluster large databases down to around 30% sequence identity at hundreds of times the speed of BLASTclust and much deeper than CD-HIT and USEARCH. In combining AlphaFold 2 and MMseq2, researchers can leverage both sequence similarity information and structural predictions to gain insights into the functions and properties of proteins.

Why Predict Protein Structure?

Why does this problem even matter in the first place? Well, here are three big reasons why this problem is worth solving.

Structure-Based Drug Discovery

One of the biggest implications of predicting the structure of proteins is for structure-based drug discovery, where using the structure of the protein, candidate drugs are picked based on if they can bind to the target with high affinity and selectivity. Affinity refers to the strength of attraction between the drug and its receptor, with a high affinity indicating a lower dose requirement. Drugs that are highly selective are also favorable because it means that they mainly affect a single organ or system, reducing unwanted side-effects that usually occur when a drug affects other parts of the body. This is especially crucial for drugs treating chronic conditions where long-term use is necessary.

Knowing a proteins structure is especially important for binding site identification — the first step in structure-based drug discovery. Here, scientists look for concave surfaces (surfaces that are curved inwards) on a protein that can fit drug-sized molecules. These areas need to have special surfaces (hydrophobic, hydrophilic, etc) that make them attractive for molecules to bind to, ensuring effective that the drug will stick. If we do not know the structure of a protein, how do we expect scientists to find these areas that a drug can potentially bind to?

Engineering Proteins

Related also to drug discovery, but if we can predict the structure of a protein in a short timeframe — days, rather than months or years — we could work backwards and first design a desired structure for a protein, specifically an active site, and then try to fine-tune a protein sequence to adopt this structure. This means that instead of adopting our drugs to fit the structure of a protein, we could design proteins to fit the structure of our therapies.

Understanding Protein Function

Protein function is dependent on its structure. If we learn more about a proteins structure, we can gain a better understanding of its function, and ultimately help us answer the question: “What is life?”

AlphaFold 2 (AF2)

A few hours before the CASP14 (14th Critical Assessment of Structure Prediction) meeting, the latest biannual structure prediction experiment where participants build models of proteins given their amino acid sequences, this image went viral on twitter.

As you can see, one group, 427, absolutely destroyed the competition: that group was AlphaFold 2. Clearly AlphaFold 2 was ahead of the game, truly being more advanced in terms of how it predicted protein structure.

Just How Good was AlphaFold 2?

To provide a little more context on how good AlphaFold 2 was, here are some other statistics that clearly showcase AlphaFold 2’s incredible performance.

Using a value called Root-mean-square deviation (RMSD) of atomic positions, which is a way to measure the the average distance between the atoms of superimposed molecules. RMSD allows us to know how different or similar two molecules are, giving us a picture of if the predicted protein structure is a indeed an accurate prediction. In proteins, we focus on certain atoms (Cα atoms) and measure the RMSD of their positions after aligning the molecules in the best possible way. For reference, a lower RMSD represents a better predicted structure, and most experimental structures have a resolution around 2.5 Å

For AlphaFold 2, about a third (36%) their submitted targets were predicted with RMSD under 2 Å, and 86% were under 5 Å, with a total mean of 3.8 Å.

Using a value called GDT_TS, AlphaFold 2 was also compared to the results of all previous CASP competitions. For reference, a GDT_TS of 100 represents perfect results and 0 is a meaningless prediction. A GDT_TS around 60% represents a “correct fold”, meaning that we have an idea of how the protein folds globally; and over 80% we start seeing side chains that closely resemble the model. As you can see, AlphaFold 2 achieves this objective for all but a small percentage of the tasks.

In case you aren’t already convinced, I wanted to show a few examples of some of the predictions made by AlphaFold 2.

Here is the prediction of the ORF8 protein, a viral protein involved in the interaction between SARS-CoV-2 and the immune response (labeled as T1064).

What impressive is how AlphaFold 2 predicted the loops. Loops are types of secondary protein structures that connect α-helices and β-strands, the two most common types of secondary structures in a protein. Both structures are held in shape by hydrogen bonds, which form between the carbonyl O of one amino acid and the amino H of another.

Loops usually cause a change in direction of the polypeptide chain, allowing the protein to fold back on itself to create a more compact structure. Loop regions are characterized by a lack of secondary structure, meaning that there is not a scaffold of hydrogen bonds that maintains the structure together, as opposed to α-helices and β-sheets. Thus, the structure of loops is far more flexible and generally considered harder to predict, showing how impressive AlphaFold 2’s performance was.

Another example is the target T1046s1. In this case, AlphaFold 2 is virtually indistinguishable from the crystal structure with an impressive total RMSD of 0.48 Å!

After three decades of competitions, the assessors declared that AlphaFold 2 had succeeded in solving a challenge open for 50 years: to develop a method that can accurately predict a protein structure from its sequence. While there are caveats and edge cases, the magnitude of the breakthrough as well as its potential impact are undeniable.

How it Works

So how did AlphaFold 2 achieve these incredible results?

The first step is that AlphaFold 2 system uses the input amino acid sequence to query several databases of protein sequences, and constructs a multiple sequence alignment (MSA). AlphaFold 2 comes equipped with a “preprocessing pipeline”, which is just a bash script that calls some other code. The pipeline runs a number of programs for querying databases and, using the input sequence, generates a MSA and a list of templates.

Templates are proteins that may have a similar structure to the input. These templates are used to construct an initial representation of the protein structure, which it calls the pair representation. This is, in essence, a model of which amino acids are likely to be in contact with each other.

A MSA is the the result of aligning three or more biological sequences, generally either a protein, DNA, or RNA. In AlphaFold’s case, it finds multiple similar proteins and aligns their sequence together with the target protein’s sequence.

To find similar sequences, the sequence of the protein whose structure we intend to predict is compared across a large database (normally something such as UniRef, although in later years it has been common to enrich these alignments with sequences derived from metagenomics). Then, JackHMMER, HHBLITs, or PSI BLAST are used within the MSA to maximize scores and correctness of the alignments.

What the MSA is ultimately looking for are conserved regions and co-evolutionary patterns. Conserved regions in a protein sequence refer to segments of the amino acid sequence that remain relatively unchanged across different species or variants over evolutionary time. Proteins mutate and evolve, but their structures tend to remain similar despite the changes. Conserved regions usually occur on a small scale, where pieces of the protein (for example, the active center of an enzyme) remain mostly unchanged while their surroundings evolve. By aligning the sequences of related proteins, researchers can pinpoint positions where the amino acids remain unchanged, indicating conservation. These conserved regions indicate that certain positions in the amino acid sequence are crucial for the structural and functional integrity of the protein, serving as a guide to construct the structure of the target protein.

To illustrate conservation, let us take the structures of four different myoglobin proteins, corresponding to different organisms. You can see that they all look pretty much the same, but if you were to look at the sequences, you would find enormous differences. The protein on the bottom right, for example, only has around 25% amino acids in common with the protein on the top left.

The next step is to take the MSA and the templates, and pass them through a transformer. AlphaFold’s transformed is called the Evoformer, which has the task of essentially getting all of the information out of the MSA and the templates. In the Evoformer, information flows back and forth throughout the network.

Before AlphaFold 2, most deep learning models would take a MSA and output some inference about geometric proximity of elements in a sequence. Geometric information was therefore a product of the network. In the Evoformer, the pair representations is a both a product and an intermediate layer.

At every cycle, the model leverages the current structural hypothesis to improve the assessment of the MSA, which in turns leads to a new structural hypothesis, and so on, and so on. Both representations, sequence and structure, exchange information until the network reaches a solid inference and can not be tangibly improved anymore.

For example, suppose that you look at the MSA and notice a correlation between a pair of amino acids: A and B. You hypothesize that A and B are close, and translate this assumption into your model of the structure. Subsequently, you examine said model and hypothesize that, since A and B are close, there is a good chance that C and D should be close. This hypothesize can then be confirmed by looking back at the MSA and searching for correlations between C and D. By repeating this several times, you can build a pretty good understanding of the structure.

Lets dive into a more technical explanation of the Evoformer because I think that it is so interesting.

The first step in the Evoformer is to define embeddings for the MSA and templates. Essentially, an embedding is a technique that allows the transformation of a discrete variable (MSA) to a continuous/embedded space so that the network can be properly trained and the Evoformer can work its magic. MSA’s are a discrete variable because they are ultimately sequences of symbols (A, G, C, and T) from a finite alphabet. On the other hand, neural networks are intrinsically continuous devices that rely on differentiation to learn from their training set.

The way that this embedding is done is by defining a layer of neurons that receives the discrete input and outputs some continuous vector. In AlphaFold 2, the embeddings are vanilla dense neural networks.

The next step of the Evoformer is to identify which parts of the input are the most important to pay attention to. Transformers all have an attention mechanism, which identifies which parts of the input are more important for the objective of the neural network.

Lets take an example, imagine that you are trying to train a neural network to produce image captions. One possible approach is to train the network to process the whole image, which could be around 250k pixels in a 512×512 picture. This may work with smaller images, but there are some reasons why it is not the best idea for larger pictures. First of all, because this is not what we do: when you look into a picture, you do not see it “as a whole”. Instead, even if you do not realize it, we focus on key elements of the image: a child, a dog, a frisbee. Luckily, we can train a cleverly-designed neural network layer to perform this exact task. And, empirically, it seems to improve the performance by a lot.

Bringing this back to proteins attention reveals which parts of the sequence are important for the current part of the translation. The only reason why transformers are not widely adopted is because the construction of the attention matrix leads to a quadratic memory cost, which may not be economical for a lot of researchers.

The Evoformer has two transformers with one clear communication between the two. One transformer is focused on the MSA while the other is focused on pair representations. They also regularly exchange information in order to refine the data that they should focus on.

For the one focused on the MSA. The network first computes attention in the horizontal direction, allowing the network to identify which pairs of amino acids are more related; and then in the vertical direction, determining which sequences are more informative. This reduces what would be otherwise an impossible computational cost.

For the one focused on pair representations, attention is arranged in terms of triangles of residues.

The Evoformer process is organized in blocks that are repeated iteratively until a specified number of cycles (48 blocks in the published model).

The final step is to actually generate the model. Here, AlphaFold 2 takes the refined MSA and pair representation and leverages them to construct a three-dimensional model of the structure. Unlike the previous state-of-the-art models, AlphaFold 2 does not use any optimization algorithm. Instead, it generates a static, final structure, in a single step. The end result is a long list of Cartesian coordinates representing the position of each atom of the protein, including side chains.

The way AlphaFold 2 does this is by modeling every amino acid as a triangle. These triangles float around in space, and are moved by the network to form the structure. These transformations are parameterized as “affine matrices,” which are a mathematical way to represent translations and rotations in a single 4×4 matrix. The key is a system called Invariant Point Attention (IPA), devised specifically for working with three-dimensional structures.

After generating a final structure, the MSA, pair representation, and predicted structure will all be passed back to the beginning of the Evoformer blocks in order to refine the prediction.

Summary

I realize I just threw a lot of information at you so let me take a step back and quickly summarize. The main steps to AlphaFold 2 are:

1. Finds similar sequences to the target protein and construct a MSA

2. Use templates to construct a pair representation

3. Pass MSA and pair representation into the Evoformer

4. Generate structure using refined MSA and pair representation

5. Pass MSA, pair representation, and generated strucutre back to Evoformer and repeat steps 3–5

Limitations

Although it is incredibly advanced, AlphaFold 2 is not a perfect model, here are some existing limitations that must be acknowledged:

1. AlphaFold depends on a multiple sequence alignment as input, and it remains to be seen if it can tackle problems where the inputs are shallow or not very informative, as happens with designed proteins or antibody sequences. The DeepMind team suggests that “accuracy drops substantially when the mean alignment depth is less than around 30 sequences.”

2. Hasn’t solved membrane proteins.

3. While AlphaFold 2 has solved one facet of the problem — predicting structure from sequence, they didn’t solve the more fundamental question of **how the protein folds. If they understand this, they can as a result understand how proteins misfold, which is the biggest reason Alzheimer’s and Parkinson’s exist. Only until we understand how proteins fold can we being to dream about curing those diseases.

Results

I tried predicting some protein structures myself using ColabFold. I used UniProt to find my protein sequences and played around with all the different options that I could customize. I definitely recommend making sure to show the side and main chains but beyond that I left most of the parameters untouched in the end.

The results were overall very good and I got a lot of accurate protein structures. If you want to check it out and try it yourself, here is the Github and Google Colab. Here is an example:

Query sequence: PIAQIHILEGRSDEQKETLIREVSEAISRSLDAPLTSVRVIITEMAKGHFGIGGELASK

Sources

I wanted to again highlight some of the the best articles that I read to help inform my research:

1. https://www.nature.com/articles/s41586-021-03819-2

2. https://www.nature.com/articles/s41592-022-01488-1#:~:text=ColabFold enables researchers to upload,from the HH-suite8.

3. https://www.blopig.com/blog/2021/07/alphafold-2-is-here-whats-behind-the-structure-prediction-miracle/

4. https://www.blopig.com/blog/2020/12/casp14-what-google-deepminds-alphafold-2-really-achieved-and-what-it-means-for-protein-folding-biology-and-bioinformatics/

Here are the two papers that we referenced if you want to read them for some extra context:

1. https://www.biorxiv.org/content/10.1101/2023.08.18.553925v1

2. https://pubmed.ncbi.nlm.nih.gov/37696656/

For nearly every cancer patient, there are a set of standard treatments that anyone can receive: chemotherapy, radiation therapy, hormone therapy, and surgery. But with there being over 200 strains of cancer, why is there so little variety in treatment types? Should we not cater to the patient’s specific cancer strain to ensure that we can minimize the side-effects of our treatments?

The answer is that we need an efficient way to classify different types of therapeutics and a safer way of introducing these treatments into our bodies. If we can effectively find nuances between different treatments, we can be better suited to prescribe treatments that fit the exact profile of an individual’s cancer. Furthermore, our body contains many complex biological systems, and the broader a therapy is introduced, the greater the likelihood of inducing adverse and harmful side effects.

Luckily, with the advent of AI, we can use a specific machine learning model in SVMs (Support Vector Machines) with Gene Therapy to create safer and more personalized cancer treatments.

Introducing: Gene Therapy

Currently used in clinical trials with successful outcomes, Gene Therapy involves introducing DNA into a patient to treat a genetic disease. The new DNA usually contains a functioning gene to correct the effects of a disease-causing mutation. This DNA is packed into a within a vector, with the job of carrying the DNA into the cells of a patient. Vectors usually come in the form of a virus, bacterium, or plasmid, mainly because these vectors are naturally good at entering our bodies undetected and infecting our cells. Once inside the cells of the patient, the DNA is expressed by the cell’s normal machinery leading to production of the therapeutic protein and treatment of the patient’s disease.

There are three main Gene Therapy techniques:

1. Killing of specific cells

   1. The aim is to insert DNA into a diseased cell that causes that cell to die. This technique is suitable for diseases such as cancer that can be treated by destroying certain groups of cells.

      

2. Gene Inhibition Therapy

   1. This method aims to eliminate the activity of a gene that encourages the growth of disease-related cells. This method is also suitable for cancer and other diseases caused by inappropriate gene activity.

      

3. Gene Augmentation Therapy

   1. This treatment targets diseases caused by a mutation that stops a gene from producing a functioning product, such as proteins. Gene augmentation therapy aims to introduce DNA containing a functional version of the lost gene back into the cell. This method is only successful if the effects of the disease are reversible or have not resulted in lasting damage to the body.

      

So Why Are We Not Already Using Gene Therapy?

While Gene Therapy is highly flexible, as there are multiple treatment approaches to solve a specific problem, how do we know which gene to use? Furthermore, how can we ensure that our gene will not produce any unnecessary immune responses?

Luckily for us, this is where SVMs come in.

Gene Therapy’s Right-Hand Man: SVMs

SVM is a supervised learning model that creates a decision boundary between data groups. Being a supervised learning model, the data that is fed into SVMs has already been identified (image, text file, video, etc) and given some meaningful and informative labels to provide context for the model. This information could be as general as tagging whether an image has a bird to as granular as identifying the specific pixels in the image associated with the bird.

The essential objective of training an SVM model is to create these decision boundaries or hyperplanes that are oriented as far as possible from the closest data points (support vectors) from each of the classes or groups. The way that SVMs create these hyperplanes is by finding the optimal weight and biases that separates the support vectors and maximizes the margin or distance between those support vectors.

Let’s get technical here. The exact way in which the hyperplanes are calculated is using the equation:

w^t(x) + b = 0

Here, b is our intercept or bias term, and w is our weight. Notice how this equation is very similar to slope-intercept form: y = mx + b. The dimension of the hyperplane is simply one less than the dimension of how the data points are represented. For example, if we defined the data points in a two-dimensional space, our hyperplane would be one-dimensional or a line.

Now, to calculate the distance between the hyperplanes and the support vectors, we use the standard form of an equation for a line: Ax + By + C = 0 and the Euclidean norm. The first equation gives us the distance between a line and a point where x and y are the respective coordinates of that point. The Euclidean norm, on the other hand, tells how far a vector extends from its starting point to its endpoint, taking into account all the components of the vector. The Euclidean norm is calculated using the Pythagorean theorem (a²+b² = c²), where we are solving for the hypotenuse or the length of the line.

What is special about the Euclidean norm, however, is that it accounts for dimensions. For two dimensions, the equation is √(x² + y²) and for three dimensions, it’s √(x² + y² + z²). Generally, for an n-dimensional vector, the Euclidean norm is the square root of the sum of the squares of all n components. Thus, the distance of a hyperplane from the support vector on a two-dimensional plane can be written as:

d= |Ax + By + C|/ |w|

Here, |w| represents our Euclidean norm.

The reason why accounting for dimensions is useful is because SVMs can classify data in multiple dimensions. What I mean by this is that sometimes data may not be linearly separable, which means that some data may not be separated by one thing but instead by a combination of multiple factors. Adding higher dimensions allows us to compare various data features, better allowing us to cluster them into groups and form our hyperplanes. SVMs accomplish this through kernel functions, which are just a class of algorithms that can take some data points and make infinitely many dimensions to accurately separate them.

Drug discovery

Now that we know a bit too much about how SVMs work, let’s actually delve into the issue: how do we find the appropriate treatment for a patient that is as safe and effective as possible?

Currently, drugs for a variety of deadly cancers remain limited, and we are now at the stage of using general-purpose treatments that generally work for all types of cancers. The problem with these treatments is that they produce side effects (chemotherapy results in hair loss for example) and do not account for the individual and their personal bodily reactions to the treatment. Moreover, the current drug discovery process involves continuously testing if a compound works against a specific biological target, which is extremely time-consuming and costly when going through a large dataset of compounds.

SVMs can aid this process by using maximum-margin hyperplanes to separate the active compounds from the inactive ones. These specific hyperplanes make sure to create the largest possible distance between any labeled compound, significantly reducing the number of compounds that researchers need to test and conduct further research on.

Once researchers narrow down their list of potential anticancer drugs, SVMs could be further useful in drug classification, using a drug’s genomic features to separate which one would be most helpful given a certain cancer cell line. This breakthrough could usher in a new era of personalized medicine where patients can have treatments catered to not only their branch of cancer (breast, skin, brain) but also to their exact cancer cell line (including any potential mutations), reducing a potential side effects or risk of the treatment not working.

SVMs use in drug classification has already been tested by a group of researchers led by Sudheer Gupta. In their work, they investigated the drug profile of 24 anticancer drugs that were tested against a large number of cell lines with the goal of understanding how changes in the genotype of a cancer cell line might be connected to the cell’s resistance to these drugs. The team used an SVM regression model to help them predict and prioritize how these drugs were against these cell lines. For reference, a cell line was only sensitive to a drug if its growth inhibition IC50 value is less than 0.5 μM.

In their study, they concluded that only a few drugs — Panobinostat, Paclitaxel, Irinotecan, 17AAG, Topotecan — were effective against more than 50% of the cell lines with Panobinostat and Paclitaxel being the most impressive. Panobinostat was observed to be effective against more than 99% cell lines while Paclitaxel was effective against 83% cell lines while also being effective against 100% of the cell lines belonging to Autonomic Ganglia tissue.

In examining the effect of variation and mutation in drug resistance, the team found that even altering one gene can completely change the efficacy of a drug. For example, the team found that the PDE4DIP gene mutated in 241 cell lines and 99% of these cell lines were also resistant to the anticancer drug PF2341066. While this does not obviously indicate a direct causation, it is interesting to note how gene mutation can make a drug completely useless against that cell line.

So Why Do These Results Matter?

Taking both of these results into account, it becomes increasingly clear that a one size fits all therapeutic is nearly impossible. With so few drugs initially proving effective against various cancer cell lines, it is challenging to envision creating a therapy that can also maintain its effectiveness against mutated cell lines. Thus, personalized medicine quickly becomes the most promising option for effective cancer treatment. Instead of giving everyone the same general treatment, how about we study everyone’s cell line, accounting for any potential mutations and variation, and provide them with the most effective drug based on their specific disease. As demonstrated in Sudheer Gupta’s work, this is exactly where SVMs come in as they are the ideal model for pinpointing what is the most effective therapeutic treatment. Going back to Gene Therapy, if SVMs can find the best gene to use for a patient, we can use Gene Therapy techniques and viral vectors to carry this treatment directly into the cells of the patient, greatly reducing the risk of extreme side-effects. Moreover, SVMs could also be used to test the potential therapeutic effect on other genes or biological processes within our body to ensure that it will neither produce unintended side effects nor trigger an immune response. Thus, in combining these two technologies, we can use SVMs to find the best therapeutic and then use Gene Therapy techniques to safely deliver this medication to the patient.

Challenges

So why is this not already a reality? Well, SVMs are not an entirely perfect model:

1. SVMs are extremely computational intensive, especially as we get to higher degrees of dimensions which will happen if we want to test extremely large and complicated datasets.

   

2. Finding the best model requires a lot of time in order to test the various combinations of kernels and model parameters.

3. SVMs can be slow to train, particularly if the input dataset has a large number of features.

Another issue in general is finding the genes to classify in the first place. While SVMs are good at finding nuanced differences in genes that will help researchers identify the best gene for the specific issue, researchers need to find promising genes in the first place.

However, I believe as computational power becomes even better and with the potential introduction of quantum computers to allow for even higher dimensions, SVMs will be the best method for discovering new target genes for various cancers. The reason why quantum computers can elevate the potential of SVMs is that they are extremely good at processing and storing data, meaning that they can classify more potential therapeutics in parallel than our current classical computers. Furthermore, since quantum computers have more processing power, they can also handle higher dimensions, allowing SVMs to separate data at a more granular level and ultimately give researchers a better picture of the most effective therapeutic for a particular scenario.

Now, with the added power of Gene Therapy, this duo will potentially solve an almost 5,000 year old problem (that’s right, we have known about cancer for 5,000 years as it was first documented in Ancient Egypt) and bring us one step closer to a cancer-free future.

Sources

Thanks for reading until the end. If you want to see where I got all of my information from, check out these links!

1. https://pubmed.ncbi.nlm.nih.gov/34161210/

2. https://aws.amazon.com/sagemaker/data-labeling/what-is-data-labeling/

3. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5822181/#:~:text=Machine https://ai.plainenglish.io/support-vector-machines-svm-clearly-explained-65f79471d4df

4. https://ourworldindata.org/cancer#:~:text=Almost

5. https://www.cancer.org/cancer/managing-cancer/treatment-types.html

6. https://www.analyticsvidhya.com/blog/2020/10/the-mathematics-behind-svm/

7. https://www.nature.com/articles/srep23857#Tab1

And here are examples of two similar patients with glioblastoma. Notice how those two blue lines (representing the cerebral ventricles), which should normally be symmetrical along the middle of the brain, have been either pushed to the side or compressed by the red outline, representative of a glioblastoma tumor.

Imagine that you are a patient and you just received an MRI scan showing similar results as above.

You’ve just received a death sentence: a glioblastoma tumor diagnosis.

Current Treatment Methods

Glioblastoma is the most common malign cancer that affects the brain and central nervous system; it accounts for around one in six brain tumors. Glioblastoma is considered a grade 4 glioma brain tumor, arising from glial cells, specifically a sub-category of glial cells called astrocyte cells (a brain tumor’s grade refers to how likely the tumor is to grow and spread, grade 4 is the most aggressive and serious type).

The tumor’s cells are abnormal, and the tumor creates new blood vessels as it grows. Given your results, you are now lined up with a three-step treatment plan: surgery (craniotomy), chemotherapy, and radiation therapy.

Currently, the first treatment step — surgery — is determined by the results of the MRI scan. Most patients with glioblastoma first have a CT scan of the brain when they go to the doctor. If a tumor is found and there is no bleeding, an MRI with standard T2-weighted (T2w), T2-fluid-attenuated inversion recovery (T2-FLAIR), gradient echo, T1-weighted (T1w), and T1-weighted contrast-enhanced (T1CE) sequences. Sequences, in this context, refers to the different types of images produced during an MRI. Each sequence uses specific MRI settings to highlight various features and properties of the brain tissue. Neurosurgeons use utilize high-resolution MRI (0.5–1.2 mm slice thickness) to help plan and guide the surgery and decide how much of the tumor can be safely removed without harming important areas of the brain.

Advanced MRI techniques like functional MRI (fMRI) are crucial before surgery, especially when the tumor is near critical brain areas. This helps plan a safer and more effective surgery. Other techniques like diffusion tensor imaging (DTI) also produce detailed images of brain pathways, aiding in surgical planning, and distinguishing between effects of surgery and remaining tumor tissue.

After your surgery, it’s common to use an MRI scan within 24–48 hours to better assess how much of the tumor was removed. Usually around 97% of the cancer cells will be removed. While 97% is a lot, you need to remove every single cancer cell to kill the glioblastoma tumor. If any cancer cell remains, the tumor will grow back. Furthermore, research has shown that removing a brain tumor causes remaining cancer cells to grow 75% faster than the original tumor, meaning that the cancer will multiply quickly and spread to other areas in spite of the surgery. Furthermore, potential side and long-term effects after brain surgery include swelling (oedema) in the brain, difficulty walking, weakness in an arm or leg, difficulty concentrating or remembering things, problems with speech, fatigue, and epilepsy.

After surgery, the next part of your treatment plan will involve chemotherapy and radiation therapy to remove all remaining cancer cells. This treatment usually starts 3 to 6 weeks after surgery, giving the patient time to recover. When planning the radiotherapy, doctors use the MRI taken after surgery along with a CT scan to carefully plan the treatment area, focusing on any remaining tumor and affected brain tissue. In radiation therapy, powerful energy beams from X-rays and protons kill cancer cells. Radiation is often combined with chemotherapy, where strong drugs (typically temozolomide, or TMZ) are used to kill cancer cells.

Chemotherapy

Chemotherapy works by killing any cell that is replicating fast. In cancer, the cells divide super quickly until there is a huge mass of cells, which become a lump, or tumor. Because cancer cells divide much faster than most regular cells, chemotherapy is much more likely to kill them than other cells.

In each living cell is a nucleus, the cell’s control center. The nucleus contains chromosomes, which consist of genes and are the target of chemotherapy drugs. Some drugs may damage the cells at the point of splitting while others damage the cells before they split — while the cells are replicating their genes. You might also have a blend of different chemotherapy drugs which damage cells at different stages of cell division, increasing the chance of killing cancer cells.

Your chemotherapy can be administered in a variety of ways:

1) Intravenous chemotherapy

In this method, you can have treatment through a thin short tube (a cannula) that enters a vein in your arm each treatment session, or you might have treatment through a tunnelled central line. Long plastic tubes inject the drug into a large vein in your chest. A huge portion of the line remains in your body for months to be used to either extract blood or inject more treatment.

Sometimes problems may occur with tunneled central lines:

1. You may get an infection.

2. The line may get blocked.

3. A blood clot can develop.

4. The line may split, although rarely.

The main problem with injecting chemotherapy drugs into the blood is that they can affect healthy body tissues where the cells are constantly growing and dividing, such as:

1. Your hair, which is always growing

2. Your bone marrow, which is constantly producing blood cells

3. Your skin and the lining of your digestive system, which are constantly renewing themselves

2) Intrathecal chemotherapy

This treatment involves injecting chemotherapy drugs into the fluid-filled space between the thin layers of tissue that cover the spinal cord and brain.

There are two different ways to do this. One way is to inject the drugs into an Ommaya reservoir (a dome-shaped container that is placed under the scalp during surgery; it holds the drugs as they flow through a small tube into the brain). The other way is to inject the drugs directly into the cerebrospinal fluid (CSF) in the lower part of the spine.

The main issue with the Ommaya reservoir is that surgery is required to both implant and remove the reservoir itself, raising the likelihood of getting an infection or other side effects. Furthermore, you are limited in your activities as you can’t do anything that would risk damaging the reservoir (e.g. contact sports) and you must be constantly wary about hitting your head. Here is a breakdown of all of the risks involved with this procedure:

1. There is a small risk that you could bleed into your brain.

2. There is a small risk that you could have some loss of function.

3. There is a small risk that you could get an infection in your brain.

4. The Ommaya reservoir may need to be adjusted. To make sure it’s in the right place, you will get a computed tomography (CT) scan the day after your surgery. If your reservoir isn’t in the right place, you may need to undergo another surgery to correct it.

5. The Ommaya reservoir may fail. To make sure your Ommaya reservoir is working, a CSF flow study may be done after your surgery.

3) Chemotherapy wafers

In this method, a surgeon will put chemotherapy drugs into the brain tissue as a wafer (e.g. Gliadel wafer). The chemotherapy drugs are inside the gel wafer and as the wafer slowly dissolves over 2 to 3 weeks the chemotherapy is also slowly released into the brain tissue.

However, the rigid structure of gliadel wafers limits drug loading capacity, and they might be dislodged from the original site of implantation, missing the target tumor. Furthermore, gliadel wafers require surgery to be implanted. Wafers have also led to the occurrence of seizures, intracranial hypertension, meningitis, cerebral edema, and impaired wound healing in neurosurgical patients

4) Oral chemotherapy

Oral chemotherapy allows you to swallow a tablet that will dissolve under the tongue. This removes the need for needles, IV lines, and constant trips to the hospital.

However, this process is strict and the patient must follow set rules to ensure effective treatment. It is critical to take oral chemotherapy drugs according to the exact schedule that the doctor recommends. The medication may be less effective if a person misses a dose, takes two doses too close together, or takes other drugs alongside the chemo.

Strict rules on dosing mean that you must remember to order new prescriptions on time, which sometimes requires ordering prescriptions days or weeks in advance. People who do not keep up with the ordering process may miss doses, which may ruin their treatment.

In addition, you must take special precautions (such as wearing gloves) when handling the chemotherapy tablets. The medications also require specific storage:

1. The tablets must be kept in their original container

2. The tablets must be away from other medications

3. In a cool, dry place away from heat, sunlight, or moisture

While oral chemotherapy does provide convenience, all of these strict rules mean that the patient has to be constantly on top of their treatment, which can be annoying and tiresome.

Oral chemotherapy also has the same effects as intravenous chemotherapy: it will also kill healthy cells that are dividing. In addition, this oral chemotherapy is expensive, costing around $180–2,600 per month. Given that oral chemotherapy is usually for around 3–6 months, these costs can add up.

Given these options, chemotherapy treatment seems to be a mixed bag of poor solutions. All have their own complexities and complications that are detrimental.

The Problems with Glioblastoma Treatment

Radiation therapy is usually done for 5–8 weeks, 5 days a week. Chemotherapy is usually done alongside radiation therapy, but chemotherapy is done every day. Four weeks after the end of radiation therapy, you may enter the monotherapy phase of treatment. This involves taking TMZ for the first five days of a 28-day cycle.

You can also face side effects from radiation and chemotherapy, including fatigue, hair loss, easy bruising and bleeding, infection, anaemia (low red blood cell counts), nausea, and kidney problems.

After all of this, your treatment plan comes at a steep cost. The average cost of glioblastoma treatment, depending on how long your treatment, comes to $160,974 for 6 months post-diagnosis, $201,749 for 12 months post-diagnosis, and $268,031 for up to 5 years post-diagnosis.

Alright, but maybe this price is worth it if it ensures you survive. After all, your life is priceless. Unfortunately, the most devastating part about glioblastoma is its survival rate. The average glioblastoma survival time with treatment is 14–16 months from diagnosis. Only 25% of patients survive more than one year, and only 5–7% survive more than five years.

Every year, around 250,000 people are diagnosed with glioblastoma and are forced to go through this process. On top of that, about 200,000 people die each year from glioblastoma as the tumor grows and destroys essential tissue in your brain.

So why is glioblastoma so deadly, and why can’t we treat it properly? As with all cancers, glioblastoma is caused by DNA mutations that result in uncontrolled cell growth. Glioblastoma tumors are especially hard to treat because they aren’t contained in a defined mass with clear borders. The tumor includes thread-like tendrils that extend into nearby areas of the brain. Furthermore, glioblastoma cells have more genetic abnormalities than cells of other types of astrocytoma brain cancer. This genetic diversity within tumor cells often leads to more aggressive behavior, as various mutations enable the cancer to grow rapidly, evade the immune system, resist apoptosis (programmed cell death), and enhance invasiveness. On top of that, the extensive genetic abnormalities make glioblastomas highly resistant to standard treatments like chemotherapy and radiation therapy. Each genetic mutation can potentially alter the tumor’s response to different therapies, making it difficult to eradicate all cancerous cells effectively.

So herein lies the problem: we currently have an extremely expensive treatment that confers an extra 8–11 months of life (you are projected to survive 4 months without treatment). Even worse, approximately 90% of glioblastoma patients will face recurrence within two years of the initial diagnosis. This happens because a few stem-like cells often evade these therapies and reinitiate tumor development. You literally have to eliminate every single cancer cell to prevent tumor recurrence and treat glioblastoma, something that is impossible with current treatments. This means that a patient likely needs checkups every 2–3 weeks to scan the brain and ensure there is no tumor recurrence. These costs can add up as the average cost of a brain MRI is on average around $5,000 (range of $1,600-$8,400). These constant checkups and time being sick lead to an additional indirect cost. One study at a hospital reported that the total indirect cost was $120,131.

To summarize,

1. Current treatments are expensive

2. Current treatments do not significantly improve the survival rate

3. Most glioblastomas recur, even multiple times. This means that patients need to go through the treatment cycle several times and spend much time getting scanned to track tumor recurrence. This can put people in an endless cycle of ineffective treatments.

However, it is important to note that surgery will always play a part in glioblastoma treatment as it is the most effective way to remove a large part of the tumor. The real improvements lie in finding an alternative to chemotherapy and radiation therapy and a treatment that can work when the glioblastoma has gone to healthy tissue and it is hard to preform surgery. Chemotherapy and radiation therapy are incapable of removing all the remaining cancer cells because most of the drug or the delivery vehicle itself doesn’t ever reach the brain. Remember, to treat glioblastoma, you have to remove every single cancer cell, even one remaining one can lead to tumor recurrence. Thus, our treatment solves the last small but critical precent that is the key to truly solving glioblastoma.

So why haven’t we had a new solution? If we can clearly see that our status quo is problematic, what is preventing us from developing a new one?

The Blood Brain Barrier

The blood-brain barrier (BBB) is a network of blood vessels and tissue that helps keep harmful substances from reaching the brain. The BBB is essential for protecting normal brain function by impeding most normal molecules from getting into the brain. No matter what, if you want to get a drug to the brain you must somehow go through the BBB. Although a nasal-to-brain pathway has been explored that can potentially bypass the BBB, several disadvantages make this solution less optimal: low bioavailability (fraction of an administered drug that reaches its target), irreversible damage of nasal mucosa, and a far more limited possible dosage. Pretty much all macromolecules (diameter ranging from about 100 to 10,000 angstroms) cannot penetrate the brain endothelium (tissues that line blood vessels). Furthermore, 98% of small molecules also can’t be transported across the BBB. Only positively charged small lipid-soluble molecules with a molecular weight < 400 Da can naturally cross the BBB. These restrictions make it challenging to design effective therapies that can enter the brain. The BBB consists of tightly packed cells and structures including endothelial cells, pericytes, astrocytes, neurons, and membranes. These components form a strong barrier in brain capillaries. The capillary endothelial cells are tightly joined without gaps, significantly limiting the passage of small molecules and proteins. Additionally, connections between these endothelial cells create a continuous barrier that restricts water-soluble substances from passing through. The permeability of the BBB is largely governed by these connections, including tight junctions, adherent junctions, and gap junctions.

More specifically, adherens junctions mainly control how substances pass through endothelial barriers. Tight junctions are crucial for maintaining the barrier that regulates the passage of materials in epithelial cells and endothelial cells, which helps keep tissues stable. Pericytes, astrocytes, and basal membrane surround the ECs and form the impermeable BBB. Additionally, efflux transporters are located in brain capillary ECs, which are further obstacles against substances entering the brain.

Furthermore, glioblastoma initially mimics the BBB and then eventually damages it as the glioblastoma grows and spreads. Once the glioblastoma is self-sustaining and can survive on its own, it constructs a new barrier called the blood-brain tumor barrier (BBTB). The BBTB has new blood vessels that supply the tumor with nutrients and oxygen and help cancer cells spread to other brain areas. This new barrier is different from the BBB in that it is more permeable (able to be penetrated or passed through). Although the BBTB allows more substances to pass through compared to the BBB, its inconsistent ability to let small and large molecules through and its varied blood flow still cause poor drug delivery to brain tumors. Thus, there is a new challenge to be overcome to create an effective drug delivery system.

Despite these many challenges, our current treatments do not have to remain as the status quo. Glioblastoma does not have to be a death sentence and we can develop a better, cheaper, and less painful solution.

Introducing Sanavo Part 1: Hyaluronic Acid Nanogels

A nanogel is a polymer-based, crosslinked hydrogel particle on the sub-micron scale. It essentially combines the advantages of a hydrogel with nanoparticles. To understand nanogels we must first understand hydrogels. A hydrogel is a three-dimensional network formed by hydrophilic polymers through chemical (covalent or ionic bonds) or physical (hydrogen bonds, van der Waals force, physical entanglement) cross-linking. Hydrogels can swell (enlarge) and retain a significant fraction of water within their structure, but they will not dissolve in water. It is produced by the simple reaction of one or more monomers: a molecule that can be bonded to other identical molecules to form a polymer. Think of hydrogels as essentially a gel: they are stable yet still possess a fluid and free-flowing quality.

The ability of hydrogels to absorb water arises from hydrophilic (tending to interact with water) functional groups attached to its backbone, while their resistance to dissolution arises from cross-links between network chains. Hydrogels are also stable in conditions of sharp and strong fluctuations in temperature.

Along with those advantages, natural polymers are ideal skeletons for hydrogels because of their diversified properties such as their biocompatibility, biodegradability, and environmental friendliness. Natural polymers can be defined as polymers that are formed from photosynthesis, biochemical reactions in the natural world, or extracted from natural products. Hydrogels based on natural polymers (eg., alginate, starch, cellulose, chitosan, gelatin, collagen, hyaluronic acid) show good degradability, biocompatibility, nontoxic degradation products, good flexibility similar to natural tissue, and are in natural abundance.

Biocompatibility is one of the most significant characteristics of hydrogels for biomedical applications, referring to the ability of a material to connect with bodily organs with minimum damage to the surrounding tissues and without triggering undesirable immune responses. Hydrogels based on natural polymers usually have excellent biocompatibility due to the intrinsic properties of natural polymers.

Biodegradability, the capacity of a substance to break down after interactions with biological elements, is another essential property of hydrogel materials for biomedical applications. Most hydrogels made from natural polymers can be broken down by enzymes. The speed at which these hydrogels decompose depends on factors such as the polymer’s molecular weight and whether the polymer is more amorphous (lacking a clearly defined shape) or crystalline (having the structure/form of a crystal), and if it is hydrophilic or hydrophobic.

Now that we understand the importance of hydrogels, lets return to nanogels. A nanogels characteristics, such as size, charge, porosity, amphiphilicity, softness, and degradability, can be fine-tuned by varying their chemical composition. They can be designed in a spherical shape or in a more porous structure, with holes in their shape. From a drug delivery perspective, this essentially allows us to design the system based on how we want the drug to be released. Based on the structure, drugs can be released in different orders. For example, in a spherical shape, the outermost contained drug would be released first; however, in a porous structure, you could theoretically have the innermost structure released first. The porous structure is sort of like Swiss cheese in that there are ways to get to the middle of the structure, allowing you to theoretically release drugs from the inside out.

Here is a full list of the advantages of nanogels:

1. Being mostly hydrophilic in nature, nanogels are highly biocompatible with a high loading capacity for guest molecules. This is important because we want the drug delivery system to be able to hold a high amount of cargo so that it can properly attack the entire tumor and not run the risk of running out of cargo. As previously mentioned biocompatibility is important to ensure that we do not illicit an immune response.

2. Nanogels not only protect the cargo from degradation and elimination but also participate actively in the delivery process due to their characteristic properties, like stimuli-responsive behaviour (responding to changes in the body) softness and swelling to help achieve a controlled, triggered response at the target site. Protecting the cargo is important to ensure the drug isn’t accidentally released in the wrong place. Softness is also critical to ensure the drug delivery system does not scratch or harm internal organs. Finally, as will be discussed later, this stimuli-responsive behavior is key as the hydrogel can essentially be programmed to release the drug based on certain conditions in the body.

3. The versatility of their architecture allows for the incorporation of a range of molecules, from inorganic nanoparticles to biomacromolecules like proteins and DNA. Both hydrophilic and hydrophobic drugs can be formulated in nanogels. This is extremely important because as we develop newer and more effective therapies, we will want this system to remain useful and still applicable.

4. Nanogels prevent biomolecules like enzymes and genetic material from degradation while their macromolecular properties help increase the circulation half-lives of small molecules.

5. Nanogels serve as a highly convenient platform for combination delivery of therapeutic molecules. Combination drug delivery refers to the approach of using two or more therapeutic agents delivered together using a single delivery system. Combining drugs can often lead to enhanced therapeutic effects as the drugs can work synergistically to treat disease more effectively than either drug could alone. It also requires less dosage as each drug can be used at a lower dose than if used alone, potentially reducing the side effects associated with higher doses of single drugs.

6. They can be targeted specifically to the site of interest by conjugation (bonding) with a targeting ligand or due to the passive targeting that is a characteristic feature of their nanoscale size.

7. Their very rapid response to a change in environmental conditions. They can adapt to different pH, temperature, and other characteristics in the body, which is important given how varying the climate is and we can’t control what happens in vivo. The responsiveness of nanogels to the external physical or chemical signals can also be tightly regulated by controlling the structure of the materials used for the preparation of the nanogels.

8. Nanogels are highly swollen and can incorporate 30% of their weight in biological molecules and drugs. These loading capacities are unusually high and exceed those of liposomes and polymeric micelles.

9. Nanogels are quite small (30–50 nm) and flexible, meaning that they have a great chance of crossing the BBB.

Hyaluronic Acid (HA)

Hyaluronic acid is a natural polysaccharide found in the extracellular matrix prominently throughout the body. It is comprised of N-acetyl-glucosamine and D-glucuronic acid residues. The extracellular matrix (ECM) helps cells attach to, and communicate with, nearby cells, and plays an important role in cell growth, cell movement, and other cell functions. As a main component of the ECM, hyaluronic acid plays a significant role in lubrication, water absorption, and retention for tissue and the ECM, structural and space-filling functions, and interacts with various cell receptors to coordinate cell communication and behavior.

So why did we choose our nanogel to be made out of hyaluronic acid? We choose hyaluronic acid because it is well known for its bioactivity, which means that it can produce a certain therapeutic effect when interacting with biological systems. Hyaluronic acid also has other advantageous properties such as its nontoxicity, nonallergy, biocompatibility, and biodegradability.

The chemical modification of hyaluronic acid is mainly focused on three distinct functional groups: the glucuronic carboxylic acids, the primary and secondary hydroxyl groups, and the N-acetyl groups. Altering any of these functional groups can result in changes in the properties (mechanical or chemical) and biological activity of hyaluronic acid. This is important because we want our system to be customizable to fit the needs of a specific patient based on where the tumor is located or what specific drug we are using.

Hyaluronic acid hydrogels can be formed by auto-cross-linking within a single molecule and between different molecules. This happens through a reaction between the carboxyl and hydroxyl groups, leading to the formation of ester bonds, which increases the hydrogels’ viscoelasticity (how it stretches and recovers) Finally, by adjusting the conditions under which this bonding reaction occurs, we can control the stiffness and stretchiness of the hydrogels.

Hyaluronic acid contains hydroxyl and carboxyl functional groups in the main chain, which can be chemically modified to obtain hyaluronic acid derivatives with unique biological and physicochemical properties. The carboxyl group in hyaluronic acid can be changed to create derivatives. Ester derivatives can be made using alkyl halides, diazomethane, or tosylate activation and amide derivatives can be formed using carbodiimides or carbonyldiimidazole. Furthermore, hyaluronidase, an enzyme that breaks down hyaluronic acid, is found at higher levels in many cancer cells. Because of this, hyaluronic acid-based hydrogels, which can be broken down by the body, are often used to deliver local therapies. While I am talking about hydrogels here, it is important to note that these characteristics can still be applied to nanogels, which are just a smaller version of hydrogels.

Hyaluronic acid is primarily produced either by extracting it from animal tissues, such as human umbilical cords, the vitreous humor of cattle, bovine synovial fluid, and rooster combs, or through microbial fermentation using both pathogenic (causing disease) and nonpathogenic bacteria. It is significant that hyaluronic acid can be found in our bodies or animals as it drastically reduces costs as opposed to chemically producing this polymer. Hyaluronic acid is often used in labs already and the average retail price of it is typically around $59.69 per 3, 30 capsules bottle.

A full list of the most important advantages of hyaluronic acid:

1. Viscoelasticity (viscous = thick and sticky, elastic = stretchy)

2. Biocompatibility

3. Biodegradability

4. High water absorption and retention capacity, make it a versatile lubricant, filler, and structural support in the human body.

5. Hyaluronic acid’s molecular weight is a crucial parameter that determines its physicochemical and degradable properties. For example, increasing HA’s molecular weight can increase its viscoelasticity.

6. Hyaluronic acid has a high affinity to the CD44 receptor, which is commonly over-expressed in cancer cells. This means that hyaluronic acid helps the drug delivery system enter and affect the tumor cells more directly. Hyaluronic acid can bind to these CD44 receptors using six monosaccharide units — HA6.

Despite all of these advantages, nanogels still have yet to be applied to clinical use, meaning that this technology is still nascent and of course, some issues need to be fixed. We will talk about those later. Now that we understand this technologies benefits, and there certainly are a lot of them, let’s talk about how we make this.

Manufacturing

How will this product actually be made? Our nanogel will be manufactured through chemical cross-linking. Nanogels that are formed through chemical crosslinking, which creates strong covalent bonds, are stable in the body. This stability is crucial as it prevents the premature release of drugs by keeping the gel network from falling apart unexpectedly. We will use the most common method: reverse microemulsion crosslinking. In this technique, tiny water-in-oil particles called micelles serve as miniature reactors where hyaluronic acid molecules are crosslinked, allowing for precise control over the size of the resulting nanoparticles. A typical formula for these microemulsions includes water for the hyaluronic acid solution, isooctane as the oil, and sodium bis (ethylhexyl) sulfosuccinate as a surfactant. Finally, 1-heptanol will be added as a co-surfactant to increase microemulsion stability.

For the crosslinking agent, we will use 1,4-butanediol diglycidyl ether (BDDE), which has gained approval from the Food and Drug Administration (FDA). We chose BDDE because it has shown the most significant swelling response to pH changes, making it promising for various biomedical applications where biodegradability and biocompatibility are crucial. Furthermore, the swelling of nanogels in an aqueous environment allows these cargos to be released more efficiently into the surrounding environment.

We will also perform the process of PEGylation, where polyethylene glycol (PEG) chains are conjugated (bonded) to a molecule. PEGylation of the nanogel surface imparts them with ‘stealth’ properties by making the surface more hydrophilic, shielding a charge that the core might carry, and establishing a physical barrier that reduces interactions with blood proteins. However, this is highly dependent on the size of the nanogel, its shape, molecular weight, and surface density of the PEG used. PEGylation is a process in which polyethylene glycol (PEG), a non-toxic and biocompatible polymer, is chemically attached to another molecule, such as a drug or a therapeutic protein. The main purpose of PEGylation is to improve the properties of the molecule it is attached to. This can include increasing the molecule’s stability, solubility, and duration in the body while also reducing immunogenicity (the likelihood of triggering an immune response). By doing so, PEGylation can enhance the efficacy and safety of medications, making them more effective for longer periods and reducing the frequency of dosing.

Injection and Delivery of Our Nanogel

To cross the BBB, our nanogel will first be attached to red blood cells (RBCs) ex vivo and will then use red blood cells to get to the brain in a process known as red blood cell hitchhiking. We choose to attach the nanogel as opposed to loading it inside the RBCs as the nanogel can be released more easily and precisely, making this approach effective and efficient for drug delivery.

RBCs are abundant, readily available, and relatively expendable. They are simpler and more uniform than other cells, with each cell being about 6–7 micrometers wide and 2 micrometers thick. RBCs are shaped like biconcave disks, which allows them to have a larger surface area and flexibility, allowing them to carry more drugs and have a long-lasting presence in the bloodstream without being detected and eliminated by the immune system. Their incredible durability and flexibility come from their strong, elastic membrane and cytoskeleton, and their lack of nuclei and other internal structures. Not having a nucleus is important, as it provides RBCs with extra room to carry drugs and makes them suitable for genetic modifications. Furthermore, RBCs’ high flexibility allows them to move through very narrow blood vessels, such as those in the spleen and potentially the BBB. Furthermore, since RBCs travel within blood vessels, they minimize the interaction of the encapsulated drugs with other substances, lowering the chance of metabolic breakdown and reducing unintended side effects. Finally, RBCs have proteins on their surface that prevent them from being easily swallowed up by the immune system, allowing them to circulate for long periods and release drugs gradually.

In order to bind the nanogel to the RBC, we will follow a successful process that was outlined in this study. First, we will draw blood from the patient using tubes and then spin the collected blood at a force of 1000 times gravity for 10 minutes at a cold temperature (around 4 degrees Celsius). This process separates the blood into layers, allowing us to remove the top layers containing plasma and the buffy coat, allowing us to reduce the volume of the final unit. We will then wash the remaining RBCs in a saltwater solution, spinning the blood at 500 times gravity for 15 minutes at around 4 degrees Celsius. This washing step allows us to thoroughly clean the RBCs to allow for better binding.

Then we will mix our nanogels with RBCs in a ratio of 200:1. This mixture will be incubated for around an hour under constant rotation again at around 4 degrees Celsius in a phosphate-buffered saline (PBS) solution. After incubation, the nanogel RBC mixture will be washed three times using PBS; the wash involves spinning the mixture at 100 times gravity for 8 minutes to remove any nanogels that had not attached to the RBCs.

While RBCs normally travel to the lungs, using intra-arterial (IA) catheters, we can direct RBCs loaded with the nanogel to specific organs. Using IA catheters allows the targeting of other organs by injecting the RBC nanogel complex directly into arteries that feed those organs. The arteries that we will target are the vertebral arteries. This is because spinal injections are generally safe procedures. If complications occur, they are usually mild and self-limited. Since blood normally travels to the brain, the RBC will have no problem getting through the BBB and getting to the brain.

Once inside, the nanogel will break apart from the red blood cell and travel to the tumor site. Because hyaluronic acid has a high affinity to cancer cells expressing the CD44 receptor, they have a great desire to find these cells and bind to them. Once it reaches the tumor, these nanogels will be engineered to be sensitive to hypoxic (low oxygen) and highly acidic conditions, which are commonly found within and on the surface of glioblastoma tumors. Thus, when the nanogel encounters this environment it will slowly degrade and slowly release the drug payload. Sustained drug release, slow release of a drug at a programmed rate to deliver the drug over a prolonged period of time, is extremely advantageous, mainly because it reduces the number of doses, lowering expenses and improving patient compliance. Furthermore, sustained drug release also decreases side effects and provides the ability to maintain a constant level of medication within the body. Releasing drugs directly in the tumor (otherwise known as targeted drug delivery) leads to greatly enhanced therapeutic efficacy because the drug is already at the site of action, making it easier to eliminate cancer cells while also not destroying healthy cells. Furthermore, targeted drug delivery can expand the pool of drugs that can be safely and effectively administered, enabling the development of novel therapies for previously untreatable conditions.

We took into account numerous safety considerations when designing this method. For example, that same study showed that attaching nanogels to RBCs would not cause the RBC to clump together. Clumping is dangerous because it can lead to blockages in blood vessels, which could cause a heart attack or stroke. The paper’s results (linked above) showed that nanogels did not cause significant clumping of the RBCs. Furthermore, it was also found that nanogels attached to RBCs did not influence oxygen levels in the blood. Comparing blood oxygen levels between mice injected with nanogels attached to RBCs and plain RBCs, there was no difference in blood oxygen levels.

We have confidence in this approach because it has already been tested on mice and produced extremely promising results. When injecting free nanoparticles (ones that did not bind to RBCs) and nanoparticles bound to RBCs directly into the right internal carotid artery of mice, a major artery that supplies blood to the brain, the study found that the brain-to-blood ratio was 27 times higher with the nanoparticles bound to RBCs than with free nanoparticles, indicating effective localization within the brain rather than remaining in the bloodstream. In assessing any potential damages caused by the red blood cell hitchhiking method, the results showed no morphological differences in the brain, suggesting that the method did not cause acute toxicity in the brain tissue.

Therapeutic

Now you may be wondering what our drug of choice is. We believe that the main issue is getting the drug to the tumor and that targeted drug delivery will by itself greatly enhance therapeutic efficacy. That is why we also emphasized having a flexible system that can adapt based on the newest and most efficient therapies. This drug delivery system can be specifically designed to cater to the specific treatment used and hold multiple treatments at the same time. There is tons of research being done into designing new and more effective therapies for glioblastoma, but not enough about delivering the drug and overcoming the main problem: the BBB. For reference, when you go on Google Scholar, there are over 445 papers on “therapies for glioblastoma” but only around 14 for “drug delivery for glioblastoma.” In fact, here is a study published just last month on a new potential therapy for glioblastoma. At the present moment however, the idea is to do combinatorial therapy, where we would combine a drug such temozolomide, a type of chemotherapy that is already largely used to treat glioblastoma and is considered the “gold standard,” with another promising drug to more effectively treat glioblastoma. For example, we could add treatments such as bevacizumab, which is typically used as a second-line treatment for recurrent glioblastomas.

Challenges

Of course, this solution is not perfect and some hurdles must be overcome. After all, if it was perfect it would have already been done.

On the surface level, an obvious challenge is the FDA and passing clinical trials. As with any implantable system, FDA regulations will be strict and testing will be even more imperative. Furthermore, making sure that this solution can pass all phases of clinical trials (not just the first ones) will be important as if we can’t get our solution to the patient, all of the money spent on research will be pointless.

Furthermore, one of the main challenges in making nanogels through chemical crosslinking is controlling their shape and size. Hyaluronic acid can turn into either a large gel network or tiny particles depending on the technique used. Our chosen method of mixing the hyaluronic acid in a water-in-oil mixture before adding the chemicals is an effective method to control the shape and size of nanogels. In this method, factors such as the ratio of water to oil, the types of surfactants (substances that reduce surface tension) used, and how vigorously the mixture is stirred can control the size of the hyaluronic particles.

Another challenge is the presence of surface charge on a nanogel. Nanogels with nearly neutral surface charge circulate longer in the blood. On the other hand, a key property of nanogels is their ability to respond to environmental changes, which is often enabled by charged groups in their structure. These charged groups also help attach drugs to the nanogel, making it challenging to balance responsive behavior with minimizing unwanted interactions with the body.

Being precise with the amount of crosslinking agent is also critical. If the amount of crosslinking agent used is too small, then the physical interaction between polymer bonds breaks easily, causing the nanogel to become water soluble and could result in it dissolving too early. However, if too much crosslinking agent is used, then a high crosslinking degree causes a low swelling degree of hydrogel, which, as discussed, makes the drug release process a lot harder.

In a study testing chemical crosslinking using BDDE, high crosslinking yields of over 90% were confirmed for all samples except when a 1:10 hyaluronic acid to BDDE molecular weight ratio was used. In these cases, the excess BDDE caused problems due to a large amount of unreacted BDDE, which limits the effectiveness (reduced solubility, improved elasticity, and enhanced mechanical strength and stability) of using high concentrations of BDDE in making hyaluronic acid nanogels. It was also found that using a low amount of BDDE significantly impacted the size of the nanogels, regardless of the HA’s molecular weight. Nanogels made with just 0.2 equivalents of BDDE had much larger particle sizes compared to those made with higher ratios of hyaluronic acid to BDDE (1:1 and 1:10). These low-crosslinked nanogels (1:0.2 hyaluronic acid:BDDE) also displayed better swelling properties, meaning they could absorb more water and expand more when the pH changed or when transitioning from dry to wet states. Thus, we now encounter a situation where we need to evaluate the trade-offs of using a high amount of BDDE versus a low amount.

Consistently reproducing these nanogel-biomolecule combinations is hard due to the unpredictable nature of the chemical reactions used. Additionally, adding targeting molecules to nanogels can change their surface properties, leading to issues like faster clearance from the body.

Furthermore, RBC hitchhiking is a newer method for delivering drugs. As this approach develops and moves towards clinical use, it will encounter various challenges, some of which might be unexpected at this early stage. For instance, in attaching drug-containing nanogels to red blood cells, the nanogel might negatively modify both the drug and the cells. Loading of the drug cargo inevitably alters the RBC, the question is whether changes in RBC biocompatibility are tolerable. For instance, encapsulating the drugs directly into the RBC leads to the loss of hemoglobin, reduced flexibility, and alterations to the cell membrane that could make the RBC more vulnerable to immune system attack. These changes resemble those seen in diseased or aging RBCs, such as those from patients with malaria or sickle cell disease. Moreover, this altered state of RBCs can lead to increased clearance by the immune system’s phagocytes, which poses additional risks including triggering inflammatory responses in the body or even blocking small blood vessels. While these are all issues for encapsulating the drug directly into the RBC, it raises questions of whether similar issues could occur when binding the nanogel to the RBC.

While RBCs are very flexible, they have their limits. If an RBC’s surface area increases by just 3–4%, the cell can burst. This means that careful handling is necessary to maintain the stability of their surface area.

Part II: Poly(lactic-co-glycolic acid) Biosensors

While the hyaluronic acid nanogel is certainly promising (despite the aforementioned challenges), it only works in tandem with the second part of our solution. Ultimately, we still need to solve the tumor recurrence problem and improve cancer imaging as a whole. Thus, we developed an implantable biosensor to provide better imaging of tumors before initial treatment and to spot tumor recurrence early.

As previously discussed, surgery has been crucial for treating solid tumors, but hasn’t changed much over the last century. Surgeons typically use their judgment to feel and see the difference between cancerous and healthy tissues during surgery, which can lead to incomplete removal of the tumor or unnecessary removal of healthy tissue. Furthermore, it is impossible to remove every cancer cell from surgery and surgeons usually can’t risk operating on healthy tissue. There needs be a more precise method to determine how much surgeons can take out and what specific tissues they can operate on. On top of that, after surgery, it’s normal for the brain tissue to shift as it fills the space left by the removed tumor. This can lead to mismatches between the MRI and CT scans used in planning treatment, sometimes with shifts as big as several centimeters. Thus, beyond imaging for tumor recurrence, we need more precise imaging as a whole that provides the doctor with a real-time picture of the tumor and what he is dealing with.

Furthermore, catching tumor recurrence early is essential as it gives us the best shot to stop it before it grows out of control and into healthy brain tissue. Recurrence of glioblastoma is nearly universal, and patients with recurrent glioblastoma fare even worse than the first time, with a median survival of only five to seven months with optimal therapy. Finally, current diagnostic imaging techniques, which include ultrasound, X-ray, CT, MRI, and PET scans, often struggle with low specificity and sensitivity in detecting early-stage cancer. They sometimes do not provide clear differentiation between cancerous, benign, and normal tissues, highlighting the need for more effective imaging methods.

Core Material

We designed a biocompatible and biodegradable system made out of Poly(lactic-co-glycolic acid) (PLGA). PLGA is a copolymer made from two types of acids: poly lactic acid (PLA) and poly glycolic acid (PGA). We choose this material due to its established safety profile and tunable degradation characteristics. Research by Makadia & Siegel demonstrates the biocompatibility and controlled release properties of PLGA nanoparticles, making them ideal for long-term brain applications.

Delivery

Similar to how we delivered our nanogel, we will also use red blood cell hitchhiking to deliver our biosensor across the BBB. Studies by Qie et al., demonstrate the effectiveness of CD47 antibody-conjugated nanoparticles for enhanced RBC adhesions and subsequent BBB penetrations. These antibodies target the CD47 “don’t eat me” signal on RBCs, preventing macrophages from removing them from circulation. Essentially, we will add antibodies that can bind to the CD47 protein on red blood cells.

Sensing the Tumor

In order to detect tumor recurrence, we will focus on the elevated levels of glucose in tumors. Tumors exhibit high glucose levels because cancer cells require a lot of glucose for energy and to produce substances that support their survival, growth, and metastasis. Glucose oxidase (GOx) will be encapsulated within the PLGA core. When GOX encounters glucose, it generates hydrogen peroxide (H202). The presence of H202 then triggers the Cy7 molecule, which can be measured using Fluorescence Resonance Energy Transfer (FRET). Research by Li et al., describes a FRET-based sensor where GOx oxidation of glucose reduces the fluorescence of a reporter molecule, indicating high glucose levels.

Generating a Detectable Signal

Our biosensor will also include a reporter molecule that generates a detectable signal upon interaction with the tumor. Selecting the ideal fluorophore for brain tumor monitoring via red blood cell-hitchhiking biosensors requires careful consideration of several factors:

1. Deep tissue penetration. Brain tissue significantly absorbs light, particularly at lower wavelengths. NIR fluorophores are crucial for deeper tissue penetration, allowing for excitation and signal detection through the scalp and skull.

2. Excitation and emission wavelengths. The chosen fluorophore should have excitation and emission wavelengths that are distinct from background tissue autofluorescence. Autofluorescence refers to the inherent fluorescence of biological tissues, which can interfere with the biosensor signal.

3. Brightness and photostability. High quantum yield (brightness) and photostability (resistance to photobleaching) are essential for strong and sustained signal detection.

For brain tumor monitoring via transcranial fluorescence imaging, near-infrared (NIR) fluorophores are the ideal choice. Here’s why:

1. Tissue penetration. Compared to visible light, NIR light experiences less scattering and absorption by biological tissues, allowing for deeper penetration into the brain. This is crucial for reaching brain tumors located deep within the skull.

2. Reduced background autofluorescence. Biological tissues exhibit inherent autofluorescence, particularly in the visible spectrum. NIR fluorophores emit at longer wavelengths where background autofluorescence is minimized, offering a clearer signal-to-noise ratio.

In order to select the optimal NIR flurophore, we had to consider various factors:

1. Excitation and emission wavelengths. These wavelengths should be chosen to maximize tissue penetration and minimize interference from autofluorescence.

2. Quantum Yield (QY). This parameter reflects the efficiency of converting absorbed light into emitted fluorescence. High QY is desirable for a strong signal.

3. Photostability. The fluorophore should be resistant to photobleaching, which is the loss of fluorescence intensity upon prolonged exposure to light.

4. Biocompatibility. The fluorophore needs to be non-toxic and well-tolerated by the body.

We choose Cy7 for our flurophore, mainly because of its excitation peak at 756 nm and an emission peak at 779 nm, which are both excellent. Similar to our nanogel, Cy7 will undergo the process of PEGylation. This is because PEGylated fluorophores have demonstrated enhanced imaging capabilities, such as better spatial resolution in the vascular and lymphatic systems and significantly higher tumor-to-background signal ratios when tagged with a targeting ligand, compared to similar NIR fluorophores. Moreover, LUM015, a PEGylated activatable fluorophore, is currently in clinical trials for detecting soft-tissue sarcoma and breast cancer, showcasing how PEGylation can effectively improve the usability and efficacy of these imaging agents in clinical settings. This technique not only addresses the solubility and clearance issues of the fluorophores but also helps maintain their safety profile for use in humans.

Sending the Signal

Cy7 will be incorporated within the PLGA core of the biosensor. Once injected, the biosensor reaches the brain and interacts with the target analyte (e.g., tumor marker). This interaction triggers a change within the biosensor, which can be detected by the fluorophore in several ways:

1. Fluorescence intensity change. In some cases, the interaction with the target molecule may directly alter the fluorophore’s brightness. For instance, some fluorophores exhibit aggregation-induced quenching (ACQ), where aggregation reduces fluorescence. The biosensor design could leverage ACQ, where upon target binding, fluorophore aggregation leads to a decrease in overall fluorescence intensity.

2. Fluorescence Resonance Energy Transfer (FRET). This technique utilizes a donor and acceptor fluorophore pair. The donor fluorophore, upon excitation, transfers its energy to the acceptor fluorophore, resulting in emission at the acceptor’s wavelength. The biosensor can be designed with the target analyte affecting the distance or interaction between the donor and acceptor fluorophores. Changes in this distance will alter FRET efficiency, impacting the detected fluorescence intensity or emission spectrum.

Cost

No matter how innovative or effective our solution is, cost is a key consideration that we must keep in mind. After all, if no one can afford our therapy then all of this work is useless. Our goal was to ensure that this solution cost less than the current treatment regime. Thankfully, this very well could be a reality

Currently, for our two core materials, as mentioned before, the average retail price of hyaluronic acid is typically around $59.69 per 3, 30 capsules bottle. The PLGA copolymer costs $65 for 500mg and $110 for 1g.

However, where the real price considerations come into play is with the fabrication, for both the nanogel and biosensor. While it is currently hard to estimate how much reverse microemulsion crosslinking will cost. We can at least estimate how much it would cost to obtain all the materials necessary. Isooctane, which is our oil, costs $91 dollars for 500mL and $140 for 1L. Sodium bis (ethylhexyl) sulfosuccinate, which functions as our surfactant, costs $415 for 10mg and $1,660 for 100mg Finally, 1-heptanol, our co-surfactant, costs $40 for 500mL and $108 for 2500mL. While these materials are certainly not cheap, the total price to obtain all of these materials and the hyaluronic acid is around $2,000 (assuming we took the larger amount that I noted).

Now for the biosensors, the cost of Cy7 is $180 for 5mg solid and $270 for 1mL aqueous. The two other important costs that I simply can’t find data for would be the cost of the system (FRET or Fluorescence intensity change) that can receive the Cy7 signal and the cost of PEGylation of Cy7.

Finally, we must also consider the costs of synthesizing all of these materials, as the material cost alone is not nearly indicative of the final cost. Some equipment that will definitely be used for the fabrication of both of these systems, especially the nanogels, will be an electrospinning machine, a plasma treatment machine, an incubator, and a centrifuge. Furthermore, everything will be prepared in a sterile cell culture hood. While I don’t know the exact prices, these machines are definitely expensive. Although it is irrational to accept a huge price decrease as the years pass, it is reasonable to assume that the general price to produce nanogels and biosensors will decrease as our technology improves, making their creation even more economical.

Conclusion

Now, with this technology, imagine a world where a glioblastoma diagnosis is not the end of the world. One trip to the doctor and a couple of hours is all it takes to rid you of this disease. Imagine a world where tumor recurrence is not an issue, where there is no need for for constant checkups or expensive routine MRI scans. You have one system, a biosensor, that can alert you immediately when tumor recurrence is spotted. If tumor recurrence occurs, it will be spotted early and taken care of swiftly.

So what are you waiting for! Join our team at Sanavo and lets work to make this imagined future a reality — where having glioblastoma is no longer a death sentence.

Sources

This article took an insane amount of research and I don’t want to play my ideas off as purely my own. Here is where I got my information from:

1. https://www.tandfonline.com/doi/full/10.1080/13696998.2017.1364258

2. https://www.americanlifefund.com/cancer/treatment/costs/brain/#:~:text=Brain Surgery Phase Costs Estimates&text=The costs associated with this,anywhere from %2430%2C000 to %24100%2C000.

3. https://www.mayoclinic.org/diseases-conditions/glioblastoma/cdc-20350148#dialogId50748516

4. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10610124/#:~:text=Hydrogels based on natural polymers,which endow them with widespread

5. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4348459/

6. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10610124/#:~:text=Hydrogels based on natural polymers,which endow them with widespread

7. https://pubs.rsc.org/en/content/articlelanding/2015/cs/c4cs00341a

8. https://pharmacy.unc.edu/2016/06/removing-brain-tumor-makes-remaining-cancer-aggressive-unc-study-finds/

9. https://www.cancerresearchuk.org/about-cancer/brain-tumours/treatment/surgery/recovering

10. https://www.thebraintumourcharity.org/brain-tumour-diagnosis-treatment/treating-brain-tumours/adult-treatments/chemotherapy/temozolomide/

11. https://www.aans.org/en/Patients/Neurosurgical-Conditions-and-Treatments/Glioblastoma-Multiforme

12. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8246629/#:~:text=The BBB is disrupted during,in brain tumours3–6.

13. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9144811/

14. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8277719/

15. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5424548/

16. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8162335/

17. https://news.mit.edu/2019/how-tumors-behave-acid-0320#:~:text=Scientists have long known that,become more invasive and metastatic.

18. https://www.mskcc.org/cancer-care/patient-education/about-your-ommaya-reservoir-placement

19. https://www.medicalnewstoday.com/articles/323482#cost

20. https://www.cancerresearchuk.org/about-cancer/brain-tumours/treatment/chemotherapy-treatment#:~:text=Chemotherapy uses anti cancer

21. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8835860/#:~:text=One such biodegradable polymer delivery,for treating GBM [11].

22. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9498307/#:~:text=If the amount of crosslinking,low swelling degree of hydrogel.

23. https://www.nature.com/articles/s41467-018-05079-7#Sec11

24. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5683405/

25. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4422082/#:~:text=CD44%20Binds%20to%20Soluble%20Extracellular%20HA%20Molecules%20and%20ECM&text=CD44%20is%20endogenously%20expressed%20at,six%20monosaccharide%20units%20(HA6).

26. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8156977/

Brandon Le is largely interested in genetics and genomics, specifically statistical genetics and genetic epidemiology. Brandon decided to work with Dr. Allison Ashley-Koch mainly due to the strong community and exciting project. He felt that the lab was full of passionate researchers who were eager to contribute. Furthermore, Brandon was intrigued by the data for the project because there were so many potential avenues that he could explore.

Background

In the first segment of our discussion, Brandon provided some important background on sickle cell disease, which was the subject of his current project. Sickle cell disease is what we call a mendelian disorder, which means that a single genetic mutation will cause sickle cell disease. The genetic mutation that causes this is a gene called beta-globin, which is part of the hemoglobin molecule. As previously described, when this genetic mutation occurs, your red blood cells will turn into a sickle shape, which in turn causes sickle cell disease and the earlier mentioned effects.

Sickle cell disease is usually triggered under hypoxic, low oxygen conditions. Hypoxic conditions occur more often than you think. While you may normally think of low oxygen situations as times where you are at high altitudes, these situations may occur during cold weather or intense physical labor such as running or going to the gym. Essentially any time that you are breathing really hard is because your body is trying to increase your breathing rate so your cells and tissues can get more oxygen. For healthy people, we don't have to worry about when we go to the gym or for a run because we don't have sickle cell disease; however, for people with that disease, even going exercising can cause sickle cell disease problems such as:

1. Pain crises

   1. This is characterized by a very acute and sudden rapid onset occurrences of pain throughout the entire body.

2. Lightheadedness and chance of fainting

   1. Essentially, hypoxic conditions are very dangerous for people with sickle cell disease because if they fail to cool down or stop to make sure that their bodies return to a normoxic condition, they can easily pass out and endure further and more serious sickle cell disease complications.

One very interesting thing that the lab found was that people that only receive one copy of this beta-globin mutation actually exhibit resistance to malaria and what we call sickle cell trait. This point links nicely to the fact that this genetic mutation primarily affects people who have African or Indian genetic ancestry, the two regions that are known to be endemic to malaria. The lab used this fact to hypothesize that the only reason that this mutated gene has evolved and persisted in the first place is to provide malaria resistance. Essentially, people who had this trait survived in malaria endemic places such as India and Sub Saharan Africa and continued to pass this trait on.

People with sickle cell trait face similar problems to people with sickle cell disease but at a reduced intensity. They still produce enough normal hemoglobin that they do not need to worry about being under hypoxic conditions, but there are still certain complications that can come up. For example, these people would still have sickled red blood cells that can affect the kidneys and degrade kidney function. However, most of these problems occur later in their lives in comparison to sickle cell disease patients who might experience kidney failure around the age of 30-40.

Project

While we know that a specific mutation in the beta-globin gene causes sickle cell disease, Brandon’s project aimed to find the specific mutations that would most impact the development of sickle cell disease. Using this, the lab would be able to engineer a vaccine that can fix this mutation and potentially cure sickle cell disease.

Brandon’s team looked at two specific metric throughout their studies:

1. Estimated Glomerular Filtration Rate(eGFR):

   1. How quickly your blood is filtered by your kidneys.

2. Proteinuria:

   1. The protein content of your urine. In healthy people, you expect that there's essentially no protein content in your urine and any kind of deviations from this is a sign that you may have some kind of kidney damage.

The reasoning behind choosing these two measurements is that the lab found that these two metrics are related to either early or mortality in sickle cell disease patients or a lower quality of life. Essentially, these two renal outcomes are good predictors of how well or how poorly a person with sickle cell disease might progress throughout their life.

The lab used data provided by NHLBI (National Heart Lung and Blood Institute), specifically from the TOPMed program. TOPMed tries to sequence the genomes of as many people as possible so that it can be publicly available to analyze for researches such as Brandon. Brandon is specifically looking at a bunch of sickle cell disease cohorts which just has data from patients with sickle cell disease.

Brandon also used an analysis tool called BioData Catalyst, which is another initiative by NHLBI. This is a cloud computing platform that allows Brandon to analyze huge datasets.

With these datasets and analysis tools, Brandon was able to make numerous comparisons such as do lower eGFR scores correlate to a lower quality of life or a shorter lifespan. The advantage of having so many cohorts is the variety of data that Brandon was presented with. Each cohort represented a specific group with the OMG SCD cohort mainly consisting of just adults and the PUSH cohort being mainly patients under 18. Using these two specific groups, Brandon could compare how sickle cell disease affects patients of different ages. Essentially, the possibilities are endless.

For his project, Brandon performed a genome wide association study (gwas). This process involved taking the genetic data after sequencing all of your patients and then asking for every single base pair in your genome or SNP (single nucleotide polymorphism), what would a mutation in that base pair cause. For example, does a mutation that causes a change from A to T result in a change in the eGFR score. The lab then does this repeatedly across all of the different mutations (every possible base pair change) within each chromosome.

From this analysis, Brandon was able to create numerous graphs that tell them how strongly a specific mutation is related to a particular outcome. While this graph may look confusing, let’s dissect it piece by piece:

• Each point on the graph represents a single mutation or SNIP.

• The x-axis represents physically where each mutations is located.

• The y-axis measures how strongly associated the mutation is with a particular outcome.

• The lab established cutoffs (each of the horizontal lines) that determines how statistically significant a mutation is. The cutoffs essentially differentiate mutations that by chance cause a disease or disorder from genes that are guaranteed to affect you, getting rid of any false positives. The blue line represents normal significance or 5 * 10^-5. These mutations are significant but there is still the possibility that they by chance caused something. Maybe you didn't have enough people in your study or your data quality isn't that high. However, on the middle the graph, the lab included an extra red cutoff value (the other ones don’t have it because no point reached that high, which represents points that are 5 * 10^-8. The red line is what we call genome-wide significance because it is the accepted threshold for statistical genetics or marks the place where a mutation should be further studied.

What the lab was essentially looking for in these graphs were really strong signals, for example, all the people with said mutation have a really low eGFR score and all the people without it have a really high eGFR score. In identifying this trend, you can assume the probability of that happening by chance is essentially zero. This creates an obvious signal that would warrant further analysis. In summary, these cutoffs essentially allow the lab to focus on a specific few mutations, narrowing down what they need to further analyze.

Challenges

1. BioData Catalyst

   1. Since this was a relatively new platform, developed only two or three years ago, Brandon was one of the first users on the platform so he spent a lot of time debugging all of the tools.

2. Data

   1. A big problem was ensuring that the data was accurate and that there were no errors during the sequencing process. This is a really important issue to tackle, in biology, there is currently a thing called a reproducibility crisis in biology because it is very easy to make a hypothesis, obtain some data that supports your claim and then make your conclusion/association while in reality your results could have been based on chance or some error in your data. If your end goal is to cure a disease, you need to confirm your results in other way, maybe by seeing if fixing the mutation actually causes the disease to go away (which inversely proves that the mutation does cause the disease).

   2. How can we filter our data to only include the most meaningful figures?

   3. How do we deal with faulty data? One of the patients had an eGFR score of 8,000, which is physically impossible. Does the lab go back to the sequencing facility to ask for a new dataset or do they just eliminate that and continue with a smaller sample size.

3. Data Management

   1. How do they store the large amount of sequencing data. This issue was actually solved by NHLBI and all the other NIH institutes where they developed a few different tools for researchers like Brandon to actually store this data on the aforementioned TOPMed program. This service is really convenient because the NIH can just manage these datasets and it allows for easier sharing. Honestly, it is really cool how all of these NIH institutes are coming together to fix pressing issues in the scientific space.

Other Work

Besides his main project, Brandon has been working on a few other projects.

One extremely interesting project is with mitochondrial data. This is a really new type of study, only within the paste five years have people actually started to look into this. Some background on mitochondria, the mitochondria is of course the powerhouse of the cell. There are actually between 20-200 mitochondria per cell, it really depends on how much energy a cell needs. For example, something like your heart, arm, or leg muscles will have more mitochondria (100-200) because they required a lot of energy; however, fat or liver cells have way less because they expend less energy.

Mitochondria have their own genomes, cell walls, and even internal cell membranes. Brandon was wondering if there is an association with sickle cell disease outcome within the mitochondrial genomes. In other words, could a mutation in the mitochondrial genome be associated with sickle cell disease in any way. Brandon was also looking to see if there is a difference between people in different haplogroups. A halpogroup is just a fancy word for a group or class of mitochondria. People of different ancestry have different haplogroups. What Brandon can do is split a population based on their haplogroups (L, A, and B) and see if there is a difference in mitochondrial mutations or sickle cell disease outcomes.

With these questions in mind, Brandon used the mitochondrial data from the same four cohorts that were shown before. This happened because in the process of sequencing the entire genomes of the people you're looking at, you tend to pick up on other other genomes such as mitochondrial genomes. What happens is when you take a cell sample or tissue sample from new people, you blend it up, and you purify the DNA inside of it. Well, there's not just human DNA, there's other DNA such as mitochondrial DNA. They then preceded to plot on a graph how many mitochondria per cell (copy number) do people have on average. The majority people had somewhere between 100-120 copies of mitochondria per cell. Brandon and his team were interested in why some people had an absurd amount such as 500 mitochondria per cell on average.

From this project, the team concluded that people with a higher average copy number per cell happen to be associated with certain disease. Brandon is making the hypothesis that maybe one of the associated disease is sickle cell disease.

The last project that Brandon mentioned that he was involved in was a heteroplasmy study. When mitochondria divide, they divide just like bacteria do. They do binary fission, meaning that they grow until they eventually split in half. For mitochondria, they do not undergo the process of recombination; instead, they always split in half and directly copy their entire genome from start to finish, preferable no mutations. Thus, there should be direct copies from one generation to the next, meaning that our mitochondria should be conserved. Heteroplasmy is what happens when one copy is different from the other for some reason. Those kinds of mutations are what Brandon was trying to look at. Specifically, Brandon is trying to see if these mutations are related to genetic human diseases such as cancer and potentially sickle cell disease.

For this project, Brandon is asking the question: within a single person, how many different copies of how many different mitochondrial genomes do we see? He then plotted his data on a graph. For people who have who are in the zero column, that means that all of their mitochondria have the same genomes, no mutations. For people in column one, they have at least one outlying mitochondria and it has at least one mutation and so on. As mentioned before, some mutations do not actually have any effect on us; however, with mitochondria, if you see a mutation, it has been shown that it usually suggests that a problem has occurred, or that it will cause problem. If you try to change the mitochondrial genome, scientists have found that the mitochondria is really sensitive to changes and that it just doesn't work as well. In contrast, if you give a random mutation to a human genome, it is probably going to still keep working. Brandon found that in the OMG-SCD sickle cell disease cohort, there are lots of people that have at least one heteroplasmy, which raises questions if there actually is an association between mutation in the mitochondria and sickle cell disease.

Overall, my interview with Brandon showcased the amount of things that you can study with a dataset. Each dataset provides you with a ton of possible avenues and it entirely matters on your enthusiasm to come up with a project idea. Furthermore, our discussion really highlighted the immense potential of the genomic and genetics field. With the completion of the human genome project in 2003, we have the ability to ask a lot more complex scientific questions. Hopefully, with the continued advancement of both of these fields, we can continue to cure genetic disease such as cancer and sickle cell disease, moving us closer to a future where genetic diseases are a thing of the past.

This paper presents the Optimized Hydrogen Liquefaction Machine (OHLM), a cooling system designed for converting hydrogen gas into a liquid state, with a focus on its application in automotive technology. The proposed system is both economical and compact, making it an ideal solution for integrating hydrogen fuel in cars. Comprising three key components—the pre-cooling system, hydrogen cooling system, and HTS (high-temperature superconducting) coil cooling system—the design optimally addresses the challenges associated with hydrogen liquefaction. The pre-cooling system sets the foundation for efficient cooling, followed by the hydrogen cooling system that utilizes O-P (Ortho-Para) conversion to achieve effective liquefaction. The synergy of these components results in a highly economical and space-saving solution — our design is the first hydrogen-powered storage system to cost less than $10,000 for production (our final cost being $9651.76), paving the way for the practical industry-wide implementation of economical hydrogen-powered vehicles with improved sustainability and performance.

Present Technology

Currently, around four-fifths of the global primary energy comes from coal, oil, and gas, which are all not sustainable sources of energy (Fossil fuels, 2017). To supply the energy demands of the rapidly increasing global population, it is essential to transition to a more sustainable energy source that does not negatively affect the environment. Hydrogen, widely touted as a clean and sustainable energy carrier, has garnered significant attention as a potential alternative to conventional fuels, thanks to several factors including its high energy density of up to 39.39 kWh/kg, its high heating value that is triple that of petroleum, and the fact that it has the highest energy content of any fuel by weight. With these factors in mind, it is of little surprise that Goldman Sachs projects the clean hydrogen industry to be worth as much as $10 trillion by 2050.

However, the practical implementation of hydrogen gas as a fuel faces significant hurdles, specifically in cooling and storage; hydrogen liquefaction effectively addresses many of these problems. Liquified hydrogen has an energy density 853 times higher than its gaseous counterpart. This condensed state also allows for more efficient storage and transportation, overcoming some of the major hurdles associated with gaseous hydrogen's volumetric inefficiency. The potential advantages of liquid hydrogen in these domains, as opposed to gaseous hydrogen, are driving ongoing research and development efforts in the hydrogen liquefaction process.

Moreover, advancements in materials are paving the way for innovative solutions. Researchers are exploring novel materials for cryogenic applications; for example, breakthroughs in the design of new materials, including G10 and austenite stainless steel, allow for more effective insulation and transportation for fuel cells.

Meanwhile, the increasing emphasis on green hydrogen production methods, specifically utilizing renewable energy sources, also holds promise for improving the environmental sustainability of hydrogen liquefaction. Electrolysis powered by renewable energy, such as water electrolysis, can produce hydrogen without relying on fossil fuels, aligning with the goal of creating a cleaner energy ecosystem.

However, the foremost impediment in the current state of hydrogen liquefaction technology is its cost. Traditional methods, such as the Claude cycle, Linde-Hampson cycle, or the modified Brayton cycle, involve intricate processes and substantial energy consumption, making them economically burdensome and inefficient in terms of space for a car. The energy-intensive nature of these techniques contributes to a considerable carbon footprint, diminishing the eco-friendly appeal of hydrogen.

The energy efficiency of these methods also leaves much to be desired. The liquefaction process demands extremely low temperatures, typically below -250 °C, to transition hydrogen into a liquid state. Achieving and maintaining such cryogenic conditions necessitates elaborate refrigeration systems, leading to increased operational costs.

Efforts are underway to design compact and decentralized liquefaction systems, particularly crucial for applications like fueling hydrogen-powered vehicles. Miniaturized liquefaction units, equipped with advanced cooling technologies, are being explored to make hydrogen more accessible and practical for diverse applications. The OHLM draws upon the work of ongoing research done by Nam et al, Aasadania et al, and Yuksel et al in order to create more cost-effective hydrogen liquefaction systems for practical use in personal vehicles. We adapted Nam et al’s pre-cooling mechanism using liquid nitrogen and their use of an HTS coil to ensure temperature stability (Conceptual Design of an Aviation Propulsion System Using Hydrogen Fuel Cell and Superconducting Motor, 2021); we used Asadania’s material choices, G10, and a strengthened copper alloy, in our design; and we used miniaturization techniques from Yuksel to create a practical final design. However, none of these papers focused on economizing the cost of hydrogen fuel cells, a key aspect that we aim to address.

History

The history of hydrogen liquefaction traces back to the late 19th century when scientists first began exploring the possibilities of transitioning hydrogen from its gaseous to liquid state. The initial breakthroughs were achieved using rudimentary methods, involving compression and cooling to extremely low temperatures. Notably, in 1898, James Dewar, a Scottish chemist, succeeded in liquefying hydrogen using a combination of compression and cooling called regenerative cooling. He used the Hampson-Linde process to do so; this process is split into four key stages: isentropic compression (elevating the gas pressure and enthalpy), expansion (significantly lowering the gas’s temperature through the Joule-Thompson effect), heat exchange (where the chilled gas cools the next batch of incoming compressed gas), and cycle repetition (back to the top, again, until the gas lowers temperature to sufficiently become liquid).

As our understanding of cryogenics deepened, the early 20th century witnessed the development of more refined liquefaction processes. Claude's cycle, introduced by Georges Claude in 1902, became another pioneering method for large-scale hydrogen liquefaction. These advancements laid the foundation for the industrial use of liquid hydrogen in various applications, including aerospace and scientific research.

In subsequent decades, researchers continued to refine and optimize hydrogen liquefaction technologies. The modified Brayton cycle, introduced in the mid-20th century, further improved the efficiency of the liquefaction process. These technological strides allowed for the production of liquid hydrogen in larger quantities, albeit with inherent challenges such as high energy consumption and environmental concerns.

In the 1960s, General Electric developed a hydrogen-powered fuel cell system for NASA's Gemini and Apollo spacecraft. This marked an early application of hydrogen fuel cells in a real-world setting. However, it was not until the late 20th century that hydrogen fuel cells started to attract attention as a potential solution for automotive propulsion.

The 1990s witnessed the development of hydrogen fuel cell vehicles (FCVs) by various automakers. Toyota, Honda, and other pioneers introduced prototype vehicles, showcasing the feasibility of using hydrogen as a clean and efficient fuel for transportation. In 2004, the first commercial hydrogen fuel cell vehicle, the Toyota Mirai, was unveiled, signaling a new era in the automotive industry.

Despite the technological advancements, challenges persisted. Limited hydrogen infrastructure, high production costs, and concerns about the overall environmental impact hindered the widespread adoption of hydrogen fuel cell vehicles. Currently, city buses are the most promising application as they only need to refuel in one place, thus requiring only one hydrogen refuel station.

Future Technology

OHLM’s cooling system comprises three main components: a pre-cooling system, a hydrogen cooling system, and an HTS (high-temperature superconducting) coil cooling system. To minimize the overall heat load, room-temperature hydrogen gas undergoes pre-cooling to 77K using cost-effective liquid nitrogen before entering the hydrogen cooling system. The cryo-cooler within the hydrogen cooling system then cools the hydrogen to 21 K, funneling it down into the inner tank. The HTS coil, a liquid hydrogen–helium heat exchanger, plays a crucial role in maintaining stable temperatures within the system and feeding liquid hydrogen to the engine. This hydrogen is sent to the blower of the car, providing a total engine capacity of 12 Wh/ml of hydrogen used, a stark upgrade compared to 9.7 wH/ml in gasoline.

Pre-Cooling

A copper pipe undergoes direct cooling through immersion in a pre-cooling system containing liquid nitrogen at a temperature of 77 K. Half of the interior of the copper pipe contains Iron II Oxide (FeO), serving as a catalyst for the ortho-para conversion (OPC) of hydrogen simultaneously during the cooling process. The ortho-para conversion is important because parahydrogen (p-H2) has significant advantages compared to orthohydrogen (o-H2), specifically in heat-related properties such as enthalpy, thermal conductivity, thermal diffusivity, and specific heat capacity. A p-H2 concentration >95% is ensured to reduce the chance of boil-off losses, which is critical to prevent mass vaporization and energy losses. Hydrogen gas, at room temperature, flows through the interior of the copper pipe and undergoes cooling. We anticipate that this gas will come from pressurized containers, which are easily bought online for relatively economical prices. The ortho-para conversion of hydrogen takes place, and the rate of conversion is dependent on the temperature of hydrogen. The target outlet temperature of the pre-cooled hydrogen is 78 K.

The diameter of this pipe will be 12.7 mm, and its thickness will be 1.24 mm. The diameter of the pre-cooling tank will be 175mm; its entire length will be 345mm. The container itself will be made of G10, a carbon composite. G10 is made up of an epoxy resin binder and a glass epoxy composite that contribute to its strength, durability, chemical resistance, and effective insulation (low thermal conductivity value of 7.0 x 10-4 °C/cm). G10 retains all these characteristics for temperatures as low as -270 °C, rendering it useful for cryogenic applications. The total G10 apparatus of the precooler will be minimal in cost — this entire portion will have costs of under $339, including the Dewar flask apparatus and G10.

We have designed this system with the assumption that the inlet temperature of the hydrogen gas will be 300 K, and it will come in at 2250 torr. The inlet volume flow will be 45 liters per minute. There is finally a pipe for N2 to be released once it becomes a gas.

The requisite materials for this would include copper piping, of dimensions 183.65x1x12.77mm. Using current standard rates, we approximate that this part of the system would cost $7100 to manufacture, most of which is devoted towards the OP converter; and it will have running costs of $143.12 to replace the liquid nitrogen.

Hydrogen Cooling

The hydrogen cooling system is carefully engineered to address various aspects of heat management, consisting of a liquid nitrogen shield layer, buffer layers, a cooling pipe, and an O-P converter (Ortho-Para converter). The outer wall is equipped with a 175x175x250mm liquid nitrogen shielding layer, strategically placed to minimize radiant heat within the hydrogen cooling system. Additionally, multiple buffer layers are incorporated in the inner tank, effectively mitigating convective heat loss. The O-P converter, a critical element, again contains FeO as a catalyst for the ortho-para conversion of hydrogen, with the rate contingent on the temperature of hydrogen.

To facilitate the comprehensive analysis of the system, specific parameters and conditions are considered. The temperature distribution of the inner wall, outlet temperature, and the pressure and flow rate of hydrogen liquefied through the O-P converter play pivotal roles. The hydrogen cooling system's O-P converter has an outer diameter of 100 mm and a total length of 75 mm. Featuring 20 pins, each with a thickness of 2 mm and a length of 40 mm, the converter optimizes heat transfer within minimal volume. Operating at an input flow rate of 0.55 lpm, the converter efficiently transforms hydrogen from an inlet temperature of 78.6 K to an outlet temperature of 21.3 K, highlighting its critical role in the controlled cooling process of the overall system.

To minimize contact of the liquid N2 with air to prevent condensation of atmospheric oxygen on the liquid N2, our model implements a Dewar flask around the liquid nitrogen containers. This will ensure that oxygen can not enter the container while also allowing gaseous N2 to escape. The gap between the two vessels in the Dewar flask will have a high vacuum, providing good thermal insulation and reducing heat conduction, and convection. The Dewar flask will trap the eventual gaseous N2 above the liquid N2 and hold it at high pressure. This will increase the boiling point of the liquid N2, allowing it to be stored for longer periods. In case of excessive vapor pressure, the Dewar flask will have pressure-relief valves to release the gas automatically.

HTS Coil

The HTS coil system consists of a current lead, a support structure, and a copper plate. To ensure optimal performance, the support structure also employs G10, effectively minimizing conduction heat load and maintaining the stability of the HTS coil. The copper plate, serving as the heat transfer pathway between the HTS coil and the heat exchanger, is constructed using OFHC (oxygen-free high-conductivity copper) for enhanced cooling efficiency due to its high heat transfer rate. To further reduce thermal load, an HTS wire is employed as the current lead.

The coil comprises six turns, and the heat exchanger's dimensions are 300 × 140 mm, with a pipe diameter of 12.7 mm and a thickness of 1 mm. It operates with a current of 150 A, featuring an inlet pipe temperature of 23 K. The current lead reaches a temperature of 80 K, while the support maintains a temperature of 300 K.

The cost of this system is $4806; most costs are for the oxygen-free copper. With the other two sections, it brings us to a grand total of $9651.76 as our cost.

Breakthroughs

The widespread adoption of hydrogen fuel faces several significant roadblocks, which preclude the usage of OHLM. While advancements in technology have addressed some challenges, various barriers persist, hindering the seamless integration of hydrogen as a mainstream energy source.

One primary roadblock is the production of hydrogen itself. The most common methods of hydrogen production involve natural gas reforming and electrolysis. Natural gas reforming emits carbon dioxide, undermining the environmental benefits of hydrogen as a clean energy source. On the other hand, electrolysis relies on electricity, often sourced from fossil fuels or other renewable energy sources, contributing to the overall carbon footprint and raising the question if it makes sense for hydrogen to be the primary power if we just have to use another energy source to extract it.

The transportation and storage of hydrogen to fueling stations pose additional hurdles. Hydrogen has a low energy density, requiring significant storage and transportation infrastructure. Existing pipelines designed for natural gas are unsuitable for hydrogen due to their unique properties, necessitating extensive modifications or the construction of new infrastructure. The development of efficient and safe hydrogen storage technologies, such as advanced composite materials or liquid carriers, is crucial for overcoming these logistical challenges. While OHLM allows cars to store LH2 easily, the bulk transportation of the fuel is another matter entirely.

Another roadblock lies in the high upfront costs associated with establishing a hydrogen infrastructure. Building hydrogen production facilities, storage tanks, distribution networks, and refueling stations requires substantial investment. Governments, industries, and investors must collaborate to allocate resources and incentivize the development of a comprehensive hydrogen ecosystem. Without financial support and a clear regulatory framework, the transition to hydrogen may be economically unviable for many stakeholders. We have seen, however, an early example of a moderately successful infrastructure base for hydrogen fuel in California, proving that it is possible to create that infrastructure base given sufficient investment into the field. A potential solution to increase hydrogen refuel station prevalence is to use electrolysis at refueling stations to convert electricity from the grid into hydrogen. If electrolysis is used, there may have to be an increase in electricity generation, though the increase would be small, since hydrogen can be produced during off-peak times.

The lack of a standardized regulatory framework is another significant obstacle. Governments and international bodies need to establish consistent regulations and safety standards to facilitate the global adoption of hydrogen. Harmonizing codes and standards would streamline the development of hydrogen technologies and infrastructure, ensuring interoperability and enhancing public confidence in the safety of hydrogen applications.

Design Process

We considered incorporating a thermal energy recovery system into our hydrogen cooling system design with the aim of harnessing and repurposing excess heat generated during the cooling process. The concept was to capture this thermal energy and convert it into usable electrical power or redirect it for other heating purposes. While this idea sounded promising in theory, we refrained from its inclusion due to practical constraints. The current technology and system dynamics may not efficiently facilitate the recovery and conversion of the generated heat. The intricate nature of the hydrogen cooling process, coupled with challenges associated with converting excess heat into useful energy, made implementing this feature impractical at the moment.

Another feature we contemplated was an adaptive cooling control system, designed to dynamically adjust cooling parameters based on real-time conditions and demands. This system could optimize the cooling process, ensuring energy efficiency and precise temperature control. However, we opted against its incorporation due to the complexity it introduced, and the resultant increased prices that consumers would face. Seamless integration would demand advancements in control algorithms and sensor technologies.

Finally, we explored the possibility of introducing an advanced hydrogen purity monitoring system to ensure the quality of the liquefied hydrogen. This system would continuously monitor hydrogen purity levels and adjust processing parameters to maintain high-quality liquid hydrogen. While this feature seemed beneficial, the current limitations in sensor technologies and the complexity of precisely controlling hydrogen purity within the stringent conditions of the liquefaction process presented significant hurdles. Achieving real-time monitoring with high accuracy and implementing precise control measures would require advancements in both sensor capabilities and process control methodologies, making the inclusion of this feature impractical for our current design.

Consequences

An anticipated future where OHLM is adopted would not be without its drawbacks. The environmental impact of mass hydrogen production could lead to a situation where hydrogen fuel cells are just as ‘dirty’ as fossil fuels, even if the invention gets adopted. Additionally, the many millions of jobs in the fossil fuel industry would almost certainly be disrupted with the widespread emergence of hydrogen technology, as we are already seeing in the status quo. The fossil fuel industry is deeply entrenched in global economies, providing the primary source of energy needs, particularly in transportation and power generation. This shift could lead to economic challenges for countries heavily dependent on fossil fuel exports and trigger workforce adjustments within the industry.

Nonetheless, let us not forget that the alternative to the adoption of renewable fuels is furthering climate change, which is far worse in its impact given that warming temperatures and more severe storms are an existential threat to humanity. OHLM holds the potential to usher in a green energy transition. By leveraging hydrogen as a clean and sustainable energy carrier, it could replace traditional fossil fuels, offering a promising solution to reduce carbon emissions and address climate change concerns. Additionally, integrating such systems into the automotive sector could revolutionize transportation, providing a cleaner alternative and overcoming limitations associated with electric vehicle range and battery technologies. It could also spur more investment into other forms of renewable energy, leading to greater institutional funding from the automotive industry and other stakeholders in the transportation sector.

If you liked this article, have any questions, or want to provide any feedback, please email me (natank929@gmail.com) or check out my <a href="https://www.linkedin.com/in/natan-kramskiy-1a3999294/">LinkedIn</a> to connect further.

Sample Web Pages

Here is a mockup of a potential website if we were to really make this product: https://www.figma.com/file/7Ho9YZIgZzyqcnMckp3PR7/Figma-Website-Template---Landing-Page-(Free)-(Community)?type=design&node-id=0-88&mode=design

Bibliography

Ritchie, Hannah, Pablo Rosado, and Max Roser. “Fossil Fuels.” Our World in Data, January 5, 2024. https://ourworldindata.org/fossil-fuels.

GenH2 Staff. “Defining the Hydrogen Economy from A to Z: D is for Density.” Hydrogen A-Z Series, January 20, 2022. https://genh2hydrogen.com/blog/defining-hydrogen-from-a-to-z-3/#:~:text=At%20standard%20atmospheric%20pressure%20and,through%20phase%20change%20is%20incredible

Manoharan, Yogesh, Seyed Ehsan Hosseini, Brayden Butler, Hisham Alzhahrani, Bhi Thi Fou Senior, Turaj Ashuri, and John Krohn. “Hydrogen Fuel Cell Vehicles; Current Status and Future Prospect.” Applied Sciences 9, no. 11 (January 2019): 2296. https://doi.org/10.3390/app9112296.

Qiu, Yinan, Huan Yang, Lige Tong, and Li Wang. “Research Progress of Cryogenic Materials for Storage and Transportation of Liquid Hydrogen.” Metals 11, no. 7 (July 2021): 1101. https://doi.org/10.3390/met11071101.

Michael Lapides, Insoo Kim, David Fishman, Chao Ji. “Green Hydrogen: The next transformational driver of the Utilities industry.” Equity Research, September 22, 2020. https://www.goldmansachs.com/intelligence/pages/gs-research/green-hydrogen/report.pdf

“Online Course and Simulator for Engineering Thermodynamics.” https://direns.mines-paristech.fr/Sites/Thopt/en/co/cryogenie.html.

Hydrogen Solutions. “LIQUEFACTION.” https://cryostar-hydrogen-solutions.com/liquefaction/.

“Hampson–Linde Cycle.” In Wikipedia, August 17, 2023. https://en.wikipedia.org/w/index.php?title=Hampson%E2%80%93Linde_cycle&oldid=1170776139.

Chang, Ho-Myung, Bo Hyun Kim, and Byungil Choi. “Hydrogen Liquefaction Process with Brayton Refrigeration Cycle to Utilize the Cold Energy of LNG.” Cryogenics 108 (June 1, 2020): 103093. https://doi.org/10.1016/j.cryogenics.2020.103093.

Aasadnia, Majid, Mehdi Mehrpooya, and Bahram Ghorbani. “A Novel Integrated Structure for Hydrogen Purification Using the Cryogenic Method.” Journal of Cleaner Production 278 (January 1, 2021): 123872. https://doi.org/10.1016/j.jclepro.2020.123872.

Dzhafarov, T. D., and S. Aydin Yuksel. “Porous Silicon-Based Direct Hydrogen Sulphide Fuel Cells.” Journal of Nanoscience and Nanotechnology 11, no. 10 (October 2011): 9012–15. https://doi.org/10.1166/jnn.2011.3456.

Nam, Gi-Dong, Le Dinh Vuong, Hae-Jin Sung, Seok Ju Lee, and Minwon Park. “Conceptual Design of an Aviation Propulsion System Using Hydrogen Fuel Cell and Superconducting Motor.” IEEE Transactions on Applied Superconductivity 31, no. 5 (August 2021): 1–7. https://doi.org/10.1109/TASC.2021.3064526.

“Energy Density.” https://chemistry.beloit.edu/edetc/SlideShow/slides/energy/density.html.

Energy.gov. “Hydrogen Storage.” https://www.energy.gov/eere/fuelcells/hydrogen-storage.

Riaz, Amjad, Muhammad Abdul Qyyum, Arif Hussain, and Moonyong Lee. “Significance of Ortho-Para Hydrogen Conversion in the Performance of Hydrogen Liquefaction Process.” International Journal of Hydrogen Energy, Special Issue on HYPOTHESIS XVI, 48, no. 68 (August 8, 2023): 26568–82. https://doi.org/10.1016/j.ijhydene.2022.09.022.

“G-10/FR4 - Current Inc,” May 20, 2017. https://currentcomposites.com/industrial-laminates/g10-fr4/, https://currentcomposites.com/industrial-laminates/g10-fr4/.

East, Sara. “G10: Your Go-To Material for Electronics, Robotics, and Mechanical Applications.” SendCutSend (blog), January 10, 2024. https://sendcutsend.com/blog/g10-your-go-to-material-for-electronics-robotics-and-mechanical-applications/.

Egan, Max. “G10 Material: Complete Guide to Epoxy Glass Composite.” Atlas Fibre, November 26, 2023. https://www.atlasfibre.com/g10-material-complete-guide-to-epoxy-glass-composite/.

“Vacuum Flask.” In Wikipedia, January 22, 2024. https://en.wikipedia.org/w/index.php?title=Vacuum_flask&oldid=1198054311.

“Cryogenic Storage Dewar.” In Wikipedia, September 30, 2023. https://en.wikipedia.org/w/index.php?title=Cryogenic_storage_dewar&oldid=1177987298.

While I know I have plenty of time to decide, it’s always fun to speculate and think about where I could go. However, whenever I do this thought exercise, many of questions arise that I ultimately cannot answer. Only through experiencing it myself will I know if a route through academia or industry is truly worth it.

There is one other way though to answer all of these questions: talking to people who are already on this journey. There are more pharmaceutical and biotechnology companies now than ever before, and an increasingly number of people are excited about working in these fields. One of the centers of biotech, Boston, has over 1,000 biotech companies alone!

Just like technology witnessed a renaissance with the dotcom era and now the AI era, science, particularly the biotech sector, is experiencing its own phase of exponential growth. Less than ten years ago, in 2015, the biotechnology industry was worth $330.3 billion. Now its worth $1.76 trillion. Now more than ever is the time to learn about what is happening in science and what it is like to work in the field.

For anyone interested in this, my conversation with Georgia offers a perfect glimpse into the life sciences industry.

Georgia’s interest in medicine and science sprouted from her parents, who were both doctors, and her innate desire to run experiments. She described running “experiments” on her older brothers, observing different aspects of their daily lifestyle.

She watched what they ate and how it made them feel, what sports they played and how their bodies reacted, and how much they slept and their behavior the following day. Bill Nye, a super famous science communicator, believed that most people who became scientists knew that they wanted to be one at an early age. While I don’t think people necessarily have a calling to science, Georgia definitely is an example of someone with an innate interest in experiments and the general scientific method.

While she had this keen interest, Georgia hadn’t yet conducted a full-length experiment with proper objective data and a hypothesis. This opportunity arose in high school when she was able to work with a Fitbit to track patients’ recovery after total hip implant surgeries. Instead of subjective observations with her brothers, she was now participating in a proper study. Georgia was able to observe the X-rays of patients and how they correlated with the patients’s progress days, weeks, and months after surgery. This experience was the point where she knew she enjoyed science and the experimental process.

Having this experience so early on was definitely crucial for Georgia as it gave her early exposure to what it means to conduct a scientific experiment. This is something most people don’t experience until college, but getting that early validation that you truly like science is important for when times get difficult and you want to quit. When those moments eventually arise, having experiences where you truly loved science is vital to remember why you chose this path in the first place.

College experience

Combining her love for science and medicine with her skill in math, Georgia decided to pursue engineering in college - a natural choice given her skillset. Georgia attended USC mainly because she really wanted to be in California and get out of her environment in Maine. At USC, she chose biomedical engineering as her major because someone suggested it and said that it was a great major for someone who was good at math and science. I love this story because it shows that people don’t normally have everything meticulously planned. Often times, people pick a path because someone suggested it or because of an impulse; not every decision has to be perfectly rational and calculated.

With the biotech boom, biomedical engineering has also gained increased popularity. The major is a perfect combination of practical engineering classes and the premed science-focused classes.

Georgia’s experience at USC was amazing, and she certainly enjoyed her degree. Specifically, she loved having tons of practical experiences such as building an EKG monitor from scratch without any instruction or creating a medical device that had yet to exist and marketing it. These real-world scenarios made classes seem less like school because you are not just learning the material but also putting it into practice. These experiences also give you key intangible skills that can only be taught through experiences like these. Georgia adopted the mindset of “lets do it” and figuring stuff out no matter how impossible the task seemed. These activities also taught Georgia how to push herself and work even when there is no clear solution in sight and she has already been working for hours.

She remembers working in a lab for 6 hours to the point where she was unable to think and yet still continuing to push through. Then when her group finally uncovered the solution, Georgia also recalls how satisfying and rewarding it felt to finally get it. Developing these mindsets and having these experiences are arguably just as important as the hard skills because it is so easy to see something fail and quit, but having the ability to see that failure and push through to find a solution is truly irreplaceable.

“Don’t use failure as an excuse to stop working.”

While everyone develops some set of skills in college, I think what really separates someone is their mindsets and attitude when it comes to approaching problems. Instead of facing challenges and saying “lets fix them tomorrow,” the mental resiliency to continue working when you are demoralized will make you work faster and accomplish more than anyone else.

In addition to her biomedical engineering classes, Georgia took many business courses because she loved the business aspect of science companies. It's important not just to take classes that align with your major but also to take ones that simply interest you. This gives you more dimensionality when going into the workplace because you can bring in different skills and attitudes from those courses. Often, those different ways of thinking is what allows you to come up with creative solutions that other people who just took the courses in the major won’t be able to see.

Listening to this story, Georgia’s college experience seemed perfect. She learned skills while also putting them into practice, setting herself up for success in industry.

Industry Life

When presented with the fork in the road, Georgia chose the industry path. Her reasoning for doing so was that she wanted to work closely with patients, developing therapies that would directly impact them. Unlike academia, where research often focuses on publications that might one day be applied to patients, industry work is far more direct in its contact with patients. Another factor that influenced her decision was that academia often has very limited funding. While prestigious labs do receive a lot of funding, industry generally has a lot more funds at their disposal. Her final reason for was her experience during college at Novartis.

During her final year, Georgia interned at Novartis, one of the largest medical companies developing drugs to treat society’s most challenging diseases. As an intern, Georgia had an incredible time because she felt valued and was treated seriously. Normally, interns are neglected and not taken seriously, but at Novartis, people were eager to teach her, and she was given important work. She was able to present in front of senior scientists and directors and take on significant projects. If a project wasn’t being pushed, Georgia had the freedom to just take the lead and get work done. She was able to try things, find papers that could influence the direction of the research, and it felt like she was a director even at the lowest level of the company. During her research, Georgia did not feel the slightest concerned about money. She could use expensive reagents and equipment without worrying about the cost. Being able to take a project all the way from the lab to presenting it is actually super rare and the fact that she was able to do this speaks to how unique Novartis is.

On top of all this, what was special about her work at Novartis was that it was an industry lab pursuing cutting-edge research. It was essentially an academic lab with industry funding - you could pursue early-stage research without worrying about funds. This is all to say that you are still researching what the company focuses on. Furthermore, your project could get suddenly closed if the company decides to move on to different things they think will generate more profit. This can be viewed in two ways. On the one hand, a Ph.D. definitely offers more stability in that there is little risk of losing your research project; however, having projects suddenly close down means that you end up dabbling in different research areas. As a result, you end up learning a bunch of different topics during your time in industry.

After her internship and graduation from college, Georgia wanted to return to Novartis. However, most positions required at least a masters or Ph.D. This is the irony of the situation: many industry professions actually require a Ph.D., meaning there is sometimes only one path you can take. Because Georgia really wanted to work at Novartis, she found one position that did not require a Ph.D. or masters. Thus, if you truly want to work in industry or do anything else for that matter, be proactive like Georgia and get that opportunity.

Upon returning to Novartis, Georgia gradually took on a larger role, eventually managing other scientists. She moved from the microscopic view of just working on her experiment to now managing others and running a significant part of the project. Novartis was constantly doing collaborations with Harvard and MIT for their projects, so the environment felt like a breeding ground for science. ‘

Hearing about her experience at Novartis made me incredibly excited about the prospect of working there. All the collaborations and opportunities make for a really exciting and innovative environment, especially given that a lot of the projects are in early drug discovery and thus super cutting-edge.

Transition to Charles River Lab

After around 3 years, Georgia switched to Charles River Laboratories, which focuses on delivering solutions to accelerate the development of drugs, chemicals, and medical devices.

Her reasoning for doing so was quite interesting. She called it the “30,000 foot view.” Imagine a cliff: the scientists are the people at the bottom, focused on just their work and tasks in the lab. Then we have the president of the research branch, he sits atop the cliff overseeing the entire research project. Finally, we have the CEO, who sits atop the mountain looking at all the cliffs because he is focused on everything - the overall research and the business strategy of the company. The problem for Georgia is that no one at Novartis had both the business perspective and the boots on the ground research point of view.

What do all of these people do for our organization?

What research areas are making money?

How do we stand up with the competition?

Georgia wanted to be able to answer all of these business questions while still having a foot in the research sector. On top of that, remember how I said that research projects often get closed. Well, with that comes the fact that people often get laid off with the projects closure. Georgia experienced that fate after her area, respiratory diseases, was closed. The major issue for her was that no one knew why the department was getting laid off, not even the director. People were saying it was strategy saying, but what does that actually mean? Georgia really wanted to understand the thought process behind the strategic team and why the company deemed it necessary.

Industry Part 2: Charles River Laboratories

Georgia was able to get those two perspectives at Charles River Lab where she was a strategy analyst for the biologics executive team. Essentially, she worked with the top dogs at the company and was able to understand the backend of what made these companies successful. She learned about the process of hiring and firing, the competition present, and the impact of new technologies on the business. Ultimately, she got to understand why Novartis laid her off and the thought process behind the strategic team.

Finally, she then moved on to being a project manager at a startup inside Charles River Lab called RightSource Solutions Georgia is specifically managing building two quality control labs, making sure everything is running smoothly in accordance with FDA and MHRA regulations. Its very interesting how her job is now fully remote which is a massive change from her work at Novartis in a lab; it really shows the variety of industry experiences that one can have. Her skills at USC now come in especially useful because her job involves a lot of effective planning and troubleshooting when something inevitably does not work out. It goes to show that while hard skills can get you the job, it is those intangibles that come in real handy during your work.

Conclusion

My mission has always been to show the lives of people working in science, both to expose myself to the possibilities and to show others interested in science what you can do. With our understanding and ability to harness biology continuing to ever-expand, there is an increasing need for people to get involved and this space and for them to continue to push the frontiers of science. Stories like Georgia’s are a perfect example of all the promise there is working in this space, and I hope that her story inspires and excites you just as it did for me.

With our previous digital revolution being marked with the formation of the World Wide Web and connecting all the devices in the world, this one is focused on the development of tools that can optimize our lives.

In the world of biotech, we have three new categories of important technology: lab automation and AI, DNA synthesis and CRISPR, and genomics. With AI, we are now able to make far more precise measurements and model biology more efficiently. AI, if used correctly, may be the best tool available for predicting the complex and nonlinear dynamics of cells.

Furthermore, DNA synthesis technologies make it possible to write new DNA. CRISPR, a Nobel Prize winner in Chemistry, gives us “genetic scissors'' or the ability to edit specific bases of DNA programmatically. CRISPR allows us to imagine curing a genetic disease by editing the dysfunctional gene back to a functional state.

Finally, between 1990 and 2003, scientists around the world coordinated to produce the first reference map of the human genome, costing nearly $3B. Since then, DNA sequencing has decreased in cost faster than any other technology in human history. DNA sequencing is the process of obtaining the exact sequence of nucleotides, or bases, in a DNA molecule, meaning that it is becoming ever easier to collect data and understand our body.

With these tools, we can clearly see how the development of new tools has greatly improved our industries and made research far easier.

Another application for technology in this digital revolution has been giving a voice to the scientific community.

Problem:

Most research is done by universities with huge amounts of funding and resources available to them. Leading investigators are becoming increasingly older while far less young people are able to assume this position, revealing signs of bureaucratic stagnation and that people are unwilling to accept change and new leadership. Research and funding should be democratized, with labs outside of academic institutions having their fair share of funding and opportunity to pursue their research.

Solution:

Establishment of the DeSci community, seeking to use blockchain to provide funding for research groups. While blockchain and cryptocurrency may not be as exciting of a topic with the FTX crisis, blockchain has many more uses beyond the speculative cryptocurrencies and the many scam projects that we typically associate it with. The DeSci community has a goal to: “build public infrastructure for funding, creating, reviewing, crediting, storing, and spreading scientific knowledge fairly and equitably using the Web3 stack.” The Web3 stack refers to all of the technology and services used to build an application. While Web2 applications rely on centralized databases, Web3 applications are built on top of blockchain architectures for trustless and permissionless access.

Several questions that the DeSci community is trying to tackle include:

1. Is it possible to utilize crypto capital markets as a means of financing ambitious scientific projects?

2. In what ways can NFTs contribute to simplifying and accelerating the process of scientific commercialization?

3. Can Decentralized Autonomous Organizations (DAOs) effectively facilitate the organization of productive online research communities?

One project in the DeSci community that is attempting to solve this problem is Molecule’s IP-NFT. Intellectual Property (IP) is a crucial method for turning scientific research into marketable products that people can get behind funding. Patents are specifically the way in which labs commercialize their research. A patent is a type of intellectual property that gives its owner the legal right to exclude others from making, using, or selling an invention for a limited period of time. Currently, patents are primarily generated by university labs because they are the main groups that perform research in the scientific community. If the patent has commercial potential, companies can negotiate with the university’s tech transfer office (TTO) to license or sell the IP. University spin-outs are specific companies created to transform technological inventions developed from university research that are likely to not be used to their maximum benefit otherwise. University spin-outs recognize the potential of a lab's research and can provide the proper resources and time required to transform this research into a marketable product that can generate a lot of profit. This entire process of licensing an IP has many real inefficiencies and most IP’s remain dormant, never becoming an actual product. Thus, Paul Kolhaas, co-founder and CEO of Molecule, proposed the idea “What if instead of using NFTs for trading pictures of monkeys, we used them to solve a real problem like creating a more fluid IP marketplace?”

With the core IP-NFT protocol being a complicated legal and technical innovation, it is perhaps better to examine its use cases to better understand the technology. The Scheibye-Knudsen Group at the University of Copenhagen ran a large-scale analysis on the impact of prescription drugs on survival. Using a dataset containing 1.5 billion prescriptions from 4.8 million individuals, they identified a set of over ten drugs correlated with extending human lifespan. They followed this by minting or posting their findings as an IP-NFT using the Molecule protocol. Their NFT was then bought by VitaDAO, a web3 community focused on funding longevity research. This transaction will provide the funding for a new study focused “on optimizing, repurposing, and re-formulating the three drugs with the strongest effect on human lifespan.” A DAO or Decentralized Autonomous Organization is a form of legal structure that has no central governing body and whose members share a common goal to act in the best interest of the entity. People who own the token of the blockchain participate in the management and decision-making. All votes and activity through the DAO are posted on a blockchain, making all the actions of users publicly viewable. VitaDAO is funded by venture capitalists, showing how there is still interest in the blockchain space despite the overall bearish attitude towards cryptocurrencies.

With infrastructure like the IP-NFT protocol, more science DAOs are starting to emerge. Now, way more people are able to get involved where previously only university students or professors received adequate funding for their ideas. Looking at the work of IP-NFT and many other projects such as LabDAO, there is a clear motive to create a web3 toolkit that allows communities to easily form around whatever they want to research and actually make things happen. DeSci is working to give anybody with an Internet connection the ability to participate in cutting-edge research. No longer will your location matter in regards to your resources, the DeSci community is working to accomplish a vision where all of your resources are online and readily available. There will be no barrier between your ideas and actually starting a research project, working to solve problems that you care about. Combining the ideals of the crypto community in permissionless entrance and community ownership with the excitement of the scientific community, there are infinite potential experiments that can be done and I am for one extremely excited for what our future holds.

Looking at the intersection between blockchain and science, I was surprised by the variety of projects that are already starting to form. One project that caught my eye was a paper from a program in Computational Biology and Bioinformatics at Yale University. In this paper, they outlined using smart contracts on the Ethereum blockchain to securely store and share pharmacogenomics data. Smart contracts are simply programs stored on a blockchain that run when predetermined conditions are met. They are typically used to automate the execution of an agreement so that all participants can be immediately certain of the outcome, without any intermediary’s involvement or time loss. Pharmacogenomics data, which is just simply genetic information that is relevant to understanding how an individual's genes influence their response to medications, is extremely vital for prescribing the correct drugs to a patient. Thus, it is of the utmost importance to store this data securely in robust places like smart contracts. Smart contracts allow for researchers and physicians to share their data with select people while also ensuring that their data is not unintentionally leaked or lost. Blockchain is thus growing in popularity as a promising solution to solve secure data storage problems because of its decentralization, distributed architecture, and immutable linking. The reason why decentralization is important is because it prevents any single user from controlling all the data; similarly, distributed architecture means that many components of the system are located on different networks or computers, thus eliminating a single point of failure. Finally, immutable linking prevents alteration of past records, preventing anyone with malicious intent to corrupt the data or any accidental mistakes.

This group aimed to develop an Ethereum smart contract for storing and querying pharmacogenomics data with time and memory efficiency. While blockchain technology offers many useful features, it is notoriously inefficient and slow when it comes to storing and sending data. They overcome this issue by using an index-based, multi-mapping data structure in a Solidity (the language that smart contracts are written in) smart contract to store the data. Specifically, they stored all pharmacogenomics data in a database mapping in which the key is an index and the value is a part of the data. They also stored three other indexes where the key is a different field (gene name, variant number, and drug name respectively), and the value is an array holding the indexes that go into the main database and match the particular field of the keys.

The group also developed an alternate solution to account for scalability called “fastQuery,” which makes use of a similar query algorithm as mentioned above, but instead organizes the gene-variant-drug relations every time someone searches for it rather than when the data is first added. This fastQuery solution exhibited significantly increased time efficiency, about one to three times faster.

When testing both of their algorithms, they found that their first solution required around 70 seconds, 500 MB of memory, and 80 MB of disk space to insert 1000 entries into the dataset. They also found that it took 400 milliseconds and 5MB of memory to send 1000 entries. This constant memory for insertion and querying meant that their algorithm is capable of handling extremely large datasets without an issue. For their alternate fastQuery solution, it required 6o seconds, 500 MB of memory, and 80 MB of disk space to insert 1000 entries while only taking 83 milliseconds and 5 MB of memory to send 1000 entries. These results confirm their theory that their second algorithm was indeed a faster solution for querying data.

Reflecting on their study, it again showcased the potential of blockchain as a resource for the scientific community. This group's approach could be used for storing data beyond pharmacogenomics such as anything that is highly important to a research team and needs to be stored securely. While blockchain is a nascent technology, it is showing signs of not just being a fad and actually having a place in biotechnology and the scientific community at large.

With large-scale gene expression data collected from millions of cells, it is possible to learn the relationship between genetic circuits and the traits of the cells they control. With genetic circuits proving to be effective tools in the field of synthetic biology, a main question in the field of synthetic biology is how can we systematically design genetic circuits for new tasks?

Many modeling frameworks and software tools have been developed for biological circuit design, but the sheer complexity of biology has limited their effectiveness. For now, these tools are still nascent. Thus, instead of pushing the implementation of new tools, it seems necessary to now take a step back and focus on better understanding our body.

We need to first obtain better quality data on the types of genetic circuits in order to later effectively model complex cellular processes. After all, a lot of meticulous work has gone into developing a framework for circuit design, but these biologists building the most complex systems are still relying on empirical intuition.

Thus, the main question for the field should shift towards: how do we collect the data necessary to accurately model and design new genetic circuits? In order to do this, scientists have begun to examine the parts of a genetic circuit to understand its meaning.

A group of scientists at the department of Bioengineering at Rice University have developed a platform called CLASSIC. CLASSIC attempts to characterize a large number of genetic circuits by combining multiple types of sequencing technologies. In characterizing numerous genetic circuits, the scientists can create mappings that reveal rules for how different genetic parts can be combined and quickly identify genes that would perform the desired functions or exhibit the desired behaviors most effectively.

So how does CLASSIC actually work? CLASSIC uses both long and short read next-generation sequencing (NGS) techniques to accurately analyze mixed libraries of DNA constructs. These libraries serve as the input for the model, and are created by combining genetic parts and a pool of barcode sequences. This library is then stitched together using a Golden Gate assembly protocol, and then later sequenced using a long-read nanopore sequencing technology. Sequencing refers to the process of determining the precise order of nucleotides or bases in a DNA or RNA molecule. This complex process ultimately establishes a mapping or index between each construct (genetic circuit) and its corresponding barcode, hence the name: construct-to-barcode index.

Next, these constructs are introduced into mammalian cells, which are just a type of cells found in the tissues of mammals. These cells contain the necessary machinery to process and express genes. Once the constructs are inside the cells, they can undergo a process called gene expression. This refers to the conversion of genetic information stored in the genes into functional products, typically proteins.

The phenotype or observable characteristics of the circuits is then measured using an approach called flow-seq. This process works by first measuring the range of observable characteristics or traits exhibited, creating bins of cells within that range using flow cytometry. Each bin then gets sequences using short-read illumina sequencing, which reads out the barcode and creates a phenotype-to-barcode index.

After this entire rather complex process, a given barcode is now linked to two pieces of information: The composition of a genetic circuit and the phenotype of the genetic circuit. Linking these two pieces of information now forms a composition-to-function relationship as we now know how the composition of a genetic circuit affects its observable traits, allowing us to determine its function.

Through this entire process, we have fully unpacked the acronym of CLASSIC: Combining Long And Short range Sequencing to Investigate genetic Complexity.

Putting this system into practice, the team optimized a system called synTF which is a circuit that is influenced by the presence or absence of a particular drug, affecting the activity of the transcription factor. A transcription factor is a class of proteins that regulates how fast the transcription process takes or the “factor” at which it performs. Using CLASSIC, they examined and analyzed a circuit design space of 165,888 possible distinct configurations. The circuit design space refers to a set of potential configurations or arrangements of a genetic circuit.

Their pooling screening approach was able to assign barcodes to 95.3% of the dataset, showcasing the incredible accuracy of their model. The scale of the circuit data that can be collected with a system like CLASSIC greatly speeds up and expands the field of synthetic biology, providing a foundation for designing complex genetic systems based on data-driven approaches.

In the labs concluding paragraph of their research paper, they mentioned an incredible application of CLASSIC.

They stated that using the immense data acquired with CLASSIC could allow them to train high capacity deep-learning algorithms to predict or even create models for extremely complex genetic circuits.2 With OpenAI’s language model, ChatGPT, showing us that AI algorithms become more accurate and efficient with the more data that they have, CLASSIC’s ability to obtain huge datasets moves us in the right direction to eventually have AI become an integral part of the synthetic biology space.

If we are eventually able to do this, it will allow us to design new medicines and other biological systems extremely quickly, propelling our society into a more digital and ultimately exciting future.

Sources:

1. https://www.biorxiv.org/content/10.1101/2023.03.16.532704v1

2. https://centuryofbio.com/p/accelerating-genetic-design