Chemical hypersensitivity is a common method to probe gene dysfunction and can be deployed in drug screens to find new therapeutics. In this blog post, we will focus on models of Inborn Errors of Metabolism (IEM) and describe how these genetic conditions can lead to hypersensitivity to a metabolite. By using humanization techniques, we take advantage of the ancient biology between humans and other organisms to create stand-ins – patient avatars – for drug screening studies. These genetically engineered model systems enable fast and affordable phenotypic screens in whole organism format to enable researchers to find molecules that alleviate the metabolic stress occurring from an IEM deficiency. Ultimately, this report showcases how chemical hypersensitivity is generalizable to drug discovery tool applicable to many genetic disorders.

Clinical variants disrupting metabolic gene can lead to build up of toxic metabolites (Figure 1).  This phenomenon can be used to create a functional assay where the model system has hypersensitivity to metabolites upstream of the gene’s function in a metabolic pathway.

Chemical Hypersensitivity Due to Metabolic Block

In the above example, the second enzyme of the metabolic pathway, “ENZ 2,” is the cause of a genetic disease  that inhibits the enzymatic conversion of metabolite 2 into metabolite 3.  For patients with this condition, exposure to metabolite 1 leads to toxic metabolite 2 build up and activation of reactive oxygen species, ultimately leading to paralysis and death.  Because of this metabolic pathway blockade, patients can experience hypersensitivity with exposure to metabolite 1.

Model Systems in the Fluidics Paradigm

A variety of model systems are possible for use in testing for metabolite hypersensitivity.  Starting with the simple model organisms, a human gene associated with disease can be installed as a gene replacement. In the case of metabolic disorders, the high degree of sequence conservation in these ancient genes often enable the human gene to rescue the function of the removed ortholog (the animal’s version of the disease gene). When we add in iPSC and then install clinical variants in the human gene locus, we get three types of model systems (C. elegans nematode, zebrafish and iPSC to model clinical variants (Figure 2). These models are advantageous for drug discovery because they fit the fluidics paradigm – The zebrafish and nematode can live their entire lifecycle in liquid and differentiated iPSC can be hooked up in biocircuits with microfluidics. By being in fluid, assessment of oral bioavailability is simply done by adding drugs to the liquid growth media. In this liquid environment, The first step is to create the wt-Avatars. In the nematode, the evolutionary distance often renders a gene to gene comparison with low homology (sequence identity under 75%). As a result only a portion of the pathogenic alleles can be modeled as amino acid substitution in the nematode’s native locus.  For work around we use Whole Gene Humanization – CRISPR gene editing is used to remove the coding sequence at the native locus and replace it with the human gene coding sequence.  When the human sequence restores normal function, we know we are looking at a high degree of conservation of biology.

In zebrafish, the orthologous gene is often at a sequence identity that is equal to or greater than 75% of the human sequence. This often renders the zebrafish’s gene sufficient for modeling the patient condition as a single amino acid substitution, yet occasionally whole-gene humanization will need to be deployed. For iPSC, we source from a healthy patient (reference line – “wt Avatar”) and make the single amino acid variation to model the patient variant condition (var-Avatar).   Bottomline, modeling a patient condition requires the creation of a wild-type humanized animal or the use of an unmodified line (wt-Avatar) and then the insertion of the missense variant that models the genetic variant of the patient (var-Avatar). In regards to IEM deficiencies, this system enables detection of phenotypic abnormalities that are often accentuated by metabolite exposure.

Detection of Hypersensitivity Phenotypes

To provide a phenotypic screen that is linked to a mechanism of action, the var-Avatar lines can be examined for their hypersensitivity to specific environmental stressors. Applied to a metabolic gene, the environmental stressor is exposure to an upstream metabolite. Using the ENZ 2 example of Figure 1, increasing dosage of metabolite 1 can be monitored for its effect on paralysis in the var-Avatar model system. To measure activity in the nematode, the 96 well format of the wMicrotracker apparatus is used to track loss of locomotion. As the animal becomes more paralyzed light beam disruption rate decreases. By exposing a nematode var-Avatar to different concentrations of metabolite, an LD50 curve can be generated (Figure 3).

In the var-Avatar, the loss of enzyme function with ENZ 2 pathogenic variants creates pathway blockage and allows build up of metabolite 2. When the metabolite 2 reaches toxic levels and the cell becomes oxidatively stressed, which leads to cell apoptosis and eventually animal paralysis and death. In this example, 40 mM of metabolite 1 is enough to cause paralysis and lethality in the var-Avatar of ENZ 2 deficient animals, but it is not enough to create paralysis in wt-Avatar control animals. As a result, a drug screen can be developed that uses this 40mM concentration as a cut-off tool for finding drugs that can alleviate the metabolic block and enable survival of the ENZ 2 var-Avatar (Figure 4).

Drug Screen

A drug screen can be performed on repurposing libraries to find hits for use in clinical trials. In the above example, the 40 mM of metabolite 1 is used to detect hits as a var-Avatar animal that can survive when exposed to a drug from the repurposing library. Additionally, the hit is considered valid when the drug has minimal to no effect on the wt-Avatar control. For metabolic disorders screening,the first step (Phase 1) is a Stress Test to determine if upstream metabolites can lead to a hypersensitivity.  In the next step (Phase 2), prior to the library screening, testing for Metabolic Modulation is performed by observing if changes in hypersensitivity can be achieved with downstream metabolites, cofactors, oxidants and antioxidants (Figure 5).  In the final step (Phase 3), A Library Screen is performed to find modulatory effects from FDA approved drugs.

Phase 1 – Stress Testing to Detect Compound Hypersensitivity

In the first phase of a 3 phase process, an assessment is made for the detection of metabolite sensitivities to metabolites and ROS mediators. In this pilot screen, a small selection of chemicals are used to determine if metabolite sensitivity can be detected. First upstream metabolites are exposed to the model system. These will often lead to their build up to toxic levels which are achieved sooner in the var-Avatar line vs the wt-Avatar line.

Scope Steps: Hypersensitivity Screen (upstream metabolites)

1. Obtain wt-Avatar and var-Avatar: Generate humanized animals for modeling a metabolic gene deficiency – created as a prerequisite project.
2. Obtain Stress-Test Compounds: Select upstream metabolites (1 to 10 molecules)
3. Solubilize Stress-Test Compound: Resuspend the metabolites in 100% DMSO at 100mM concentration to mimic the sample conditions of most drug screening libraries.
4. Measure Activity: Determine suppression as retention of locomotion using wmicrotracker plate reader instrumentation.
5. Generate LD50 curves: Expose animals at concentrations 0.01, 0.1, 1, 10, 100, and 1000 uM (1% DMSO)
6. Select Optimal Separation: Examine LD50 curves for conditions creating a high degree of separation between wt-Avatar and var-Avatar.
7. Repeat With Combinations (optional): If necessary, repeat with combinations of chemicals to attempt to create at least 4x separation in chemical sensitivity between wt-Avatar and var-Avatar.

Phase 2 – Metabolic Modulation Effects

Once metabolic hypersensitivity is established, a second set of molecules are examined for their ability to alleviate metabolic defect and restore normal activity. Downstream metabolites and cofactors can be added to the system to determine if they can restore normal metabolite sensitivity.  Additionally, In many metabolic disorders the buildup of metabolite intermediates lead to disruption of normal levels of Reactive Oxygen Species (ROS) (doi: 10.1155/2018/1246069). As a result, two other groups of molecules can be screened for their effect on metabolite hypersensitivity. Various antioxidants (resveratrol, glutathione, metformin) can be tested for their ability to decrease sensitivity to metabolite build up. On the other hand, some metabolic intermediates may attenuate ROS which at low levels are necessary for homeostasis signaling. Various oxidants (paraquat, juglone and AAPH (2,2′-azobis-2-methyl-propanimidamide)) may lead to restoration of normal low levels of ROS and alleviate metabolic hypersensitivity.

Scope Steps: Metabolic Modulation (downstream metabolites, cofactors, oxidants, antioxidants)

1. Obtain metabolic-modulator Compounds: Select downstream metabolites, and oxidants and antioxidants (5 to 10 chemicals)
2. Solubilize metabolic-modulator Compound: Resuspend the compounds in 100% DMSO at 100mM concentration to mimic the sample conditions of most drug screening libraries.
3. Measure Activity: Determine suppression as retention of locomotion using wmicrotracker plate reader instrumentation.
4. Generate EC50 curves: Using a upstream metabolite at a concentration that leads to a strong deficiency (ie. death), add metabolic-modulator at concentrations 0.01, 0.1, 1, 10, 100, and 1000 uM (1% DMSO)
5. Identify Modulator Candidates: Examine the EC50 curves for the modulators effect on wt-Avatar and var-Avatar.
6. Repeat With Combinations (optional): If necessary, repeat with combinations of chemicals to attempt to decrease the separation between wt-Avatar and var-Avatar.

Phase 3: Library Screening Approach

For speed to clinical trials, a good choice is to use repurposing libraries. These are the FDA-approved drugs that are well vetted for ADMET issues, which makes them a safe choice for therapeutic consideration. We will apply the assay developed in phase 1 to scale across a commercially-sourced drug screening library. A variety of sources for compound libraries utilizing FDA-approved drugs are available (TargetMol, ApexBio, Chembridge, Microsource, Prestwick, Seleckchem, etc… – typically about 2000-3000 mlcs). As an example, the metabolite hypersensitivity assay is applied to the var-Avatar line (patient variant animal) at 10uM concentrations of compounds from the Apexbio FDA-approved library of 2726 compounds. The top hits (up to 200) are counter screened on wt-Avatar animals and the results are scored for minimal alteration of wild-type metabolite sensitivity. A rank is developed that balances strong response in the var-Avatar line against a strong response in the wt-Avatar line. The top 20 hits are examined at a range of doses to find the EC50 values in suppressing metabolite hypersensitivity. Data package is prepared and delivered to the client.

Scope Steps: Examine Rescue of metabolite Sensitivity in var-Avatar line with 2000+ Chemical Library

1. Obtain library: Multiple repurposing libraries are available – for example, ApexBio library of 2726 FDA-approved chemicals.
2. Drug Exposure: Using Stress-Test chemical concentrations optimized in phase 1, test The var-Avatar line against the library of 2726 molecules at 10uM concentration for their ability to suppress metabolite hypersensitivity.
3. Measure Activity: Determine suppression as retention of locomotion using wmicrotracker plate reader instrumentation.
4. Perform Primary Screen: Score compounds as positive or negative for ability to suppress metabolite hypersensitivity by retaining locomotion.
5. Identify Preliminary Hits: Select up to 200 compounds (≤ mlcs) from the positive category for use in repeat screen on wt-Avatar animals.
6. Perform Counter Screen: Using metabolite concentration optimized in phase 1, test the ≤200 mlcs at 10uM concentration for their ability to NOT suppress metabolite sensitivity in wt-Avatar animals.
7. Rank Hits: Select up to 20 compounds (≤20 mlcs) that are positive in the var-Avatar for suppression of metabolite hypersensitivity and negative for suppression of metabolite sensitivity in the wt-Avatar.
8. Characterize Hits: Perform EC50 assays on var-Avatar line for the 20 compounds.
9. Report Out: Send a report to the client.

RWE – Real World Evidence Applications

A real world example occurs in the propionate metabolic pathway.  Defects in the ECHS1 gene lead to altered propionate metabolism shunting propionate away from Acetyl-CoA production and instead only allow Succinyl-CoA production (Figure 6). 

Potentially hampering succinyl-CoA production is a vitamin B12 dependency of the last enzyme in the Succinyl-CoA production pathway. When both B12 is limiting and the Acetyl-CoA production is blocked by a ECHS1 defect, hypersensitivity to propionate occurs. Since the propionate pathway is highly conserved between the nematode and humans, creating a deficiency in the nematode equivalent of ECHS1 (the ech-6 gene) can create a propionate hypersensitivity (Figure 7).

Loss of function of ech-6 creates propionate hypersensitivity

RNAi was used to knock down expression from the ech-6 gene (Watson et al. Elife. 2016 Jul 6;5:e17670). The result was hypersensitivity to propionate – a 9 fold increase in LD50 with exposure to the metabolite. When B12 was added to the system, partial rescue of hypersensitivity occurred. From this data where the drug screen was performed at 30 mM propionate, the level of rescue of survival was 4 fold higher with B12 exposure. 40-50 mM propionate appears to be the ideal range for detecting chemicals capable of rescue of propionate hypersensitivity in ech-6 null animals (KO). Inserting the human ECHS1 as gene replacement of ech-6 and observing restoration of normal propionate sensitivity will establish that the animal model (wt-Avatar) can be used to create a background system for modeling human disease. Next, installing variants in the humanized strain will create a model system for specifically exploring a patient’s genetic condition (var-Avatar) (Figure 8). When good separation between var-Avatar and wt-Avatar is observed, a screen for rescue of propionate hypersensitivity can be performed and B12 can be used as positive control.

B3GAT3 Drug Target for Modulating Propionate Hypersensitivity

Reassuringly, other genes involved in propionate metabolism are known to have hypersensitivity when they are made defective. For instance, the PCCA gene, in combination with PCCB, is a heteromeric enzyme responsible for converting propionyl-CoA to D-methylmalonyl-CoA, which is an intermediate step of the metabolism of propionate to succinyl-CoA. When loss of function is created in the nematode ortholog pcca-1, propionate hypersensitivity occurs and these animals do not survive at above 50 mM propionate (Watson et al. Elife. 2016 Jul 6;5:e17670).  This same group used the propionate sensitivity assay to determine that  loss-of-function mutations in glct-3 (an ortholog of B3GAT3) create a propionate resistant animal (Na et al., PLoS Genet. 2020 Aug 28;16(8):e1008984). As a result, it appears inhibition of B3GAT3 may be useful in offsetting the hypersensitivity of defects within the propionate pathway.  In silico approaches can be used to dock molecules to the B3GAT3 structure (1FGG, 1KWS, 3CU0) (Figure 9).  Hits found can then be validated by testing in a humanized animal model.

Zebrafish Model for Study of RPE65 Defects

Switching to zebrafish models, vitamin A metabolite toxicity can be used to probe defects in the RPE65 gene. First, morpholinos or crispants can be used to dramatically reduce expression of this gene in the fish and lead to loss of function phenotypes.  Next, the addition of an mRNA to the injection mix can be used to rescue the loss of function and restore normal metabolic activity.  With regards to the loss of function activity, morpholinos in zebrafish work by shutting down translation of a gene transcript.  In the alternative approach, crispants work by disrupting the gene’s coding sequence in a large proportion of cells (>90%). Often the effect with either a pproach is a 10x or more decrease in gene expression which effectively models a loss-of-function variant. It is expected loss of function of the zebrafish isoforms for RPE65 will result in metabolite hypersensitivity.

In humans, loss of the RPE65 retinoid isomerohydrolase, an enzyme of the visual cycle involved in catalytic recycling of retinyl palmitate to 11-cys-retinol, will render the REP65 deficient animals hypersensitive to retinyl palmitate and its precursors. The all-trans-retinal (atRAL) is a precursor with established toxicity in the visual system (Chen et al. J Biol Chem. 2012 Feb 10; 287(7): 5059–5069; Gao et al., J Biol Chem. 2018 Sep 14; 293(37): 14507–14519). The RPE65 enzyme is critical in the last step of the visual cycle by creating the 11-cis-retionol needed by opsins to convert light into a biochemical signal. Deficiencies in this enzyme are likely to render the animals prone to light-induced blindness. In zebrafish, there are three genes (rpe65a, rpe65b, and rpe65c) that provide support for the RPE65 role in humans (Ward et al., Front Cell Dev Biol. 2018; 6: 37). A crispant CRISPR-based somatic knockout strategy can be employed where three sgRNAs each targeting the three isoforms are used to create removal of retinoid isomerohydrolase from nearly all tissues (~95%) in a zebrafish embryo. Exposing the resulting larvae to bright light and all-trans-retinal (atRAL) is likely to create a hypersensitivity that manifests as a high rate of blindness.

To rescue the blindness, an mRNA encoding human retinoid isomerohydrolase (hRPE65) is added to the crispant mix. This allows the visual cycle to remain active and toxic levels of atRAL are avoided. To determine if a variant in hRPE65 is pathogenic, any of the 177 missense variants identified in the clinical population database of ClinVar can be made as an mRNA and then be examined for their pathogenic potential (lack of rescue). The mRNA effectively becomes a var-Avatar model for testing a variant’s hypersensitivity to atRAL.  For the variants that exhibit accelerated blindness, small molecules can be explored for their ability to restore normal sensitivity to atRAL exposure.

An iPSC Model for Study of NOTCH3 Deficiencies

As an evolving model system, induced pluripotent stem cells (iPSCs) can be derivatized into tissues that are populated into microphysiological systems (MPS). Some of the MPS use iPSC-derived tissue that models the vascular system. An important disease to model in vascular formats is cerebrovascular disorders. CADASIL is one of the vascular diseases with an established genetic cause – variations in the extracellular domain of NOTCH3 gene result in small blood vesicle defects in the brain. Although NOTCH3 variations causative for CADASIL are rare, the disease is considered an ideal model for cerebrovascular disease, which affects 25% of all stroke patients and 45% of all dementias. Further we know there is a therapeutic antibody binding site on the extracellular side of the NOTCH3 protein that modulates the disease by promoting extracellular cleavage and turnover of the NOTCH3 gene. Loss or gain of cystines are the most common pathogenic variants in NOTCH3. A reference iPSCs can be modified by CRISPR to contain the patient variant (var-Avatar) and be compared to the unmodified cells (wt-Avatar). In CADASIL patients, oxidative stress occurs (Neves et  al., JCI Insight. 2019 Dec 5; 4(23): e131344.) Specifically, CADASIL patients exhibit Nox5-induced oxidative stress as measured by Lucigenin-enhanced chemiluminescence assay of NADPH-dependent ROS production. So, although the NOTCH3 gene is not considered a metabolic gene, disruptions in its function behave like many metabolic deficiencies and lead to ROS imbalance.  For a hypersensitivity screen, it is likely oxidative stressors such as paraquat will exhibit hypersensitivity in iPSC-derived MPS models of CADASIL patients.

Summary

Stressor hypersensitivity can be used to detect favorable drug effects.  Applied to Inborn Errors of Metabolism, the use of an upstream metabolite can be a stressor that leads to hypersensitivity to the metabolite. Patients with defects in an enzyme of a metabolic pathway experience unhealthy build up of a metabolite intermediates. We can take advantage of the ancient biology between humans and other organisms for the genes involved in metabolism where we create the patient’s genetic defect in the model system, either nematode or zebrafish (or iPSC). This humanized animal (or modified iPSC) then becomes the patient avatar for use in drug screening studies. For metabolic deficiencies, the animal can be used for phenotypic screens to find molecules that alleviate the metabolic stress occurring from the deficiency.  Ultimately the approach is likely to be widely generalizable to many genetic disorders.

Authors: Hannah Huston, Alexandra Narin, and Chris Hopkins

By definition, there are not a lot of patients for any given Rare Disease. A disease is only considered rare (or “orphan”) if it affects fewer than 200,000 people (NIH). Often, a rare genetic condition caused by a malfunction of a gene is even more infrequent. Maybe only a small handful of people on the planet will have genetic lesions in a given gene. Despite this, rare disease as a group is actually quite common. There are over 7,000 genes in everyone’s genome that can harbor a gene variation which leads to a disease. One way to conceptualize the variants in a rare disease is to imagine an inverted image of a galaxy, where black dot clusters of Rare variants have different levels of phenotypic severity (Figure 1). The variant clusters in the outer fringes are near normal activity, but in the center is the black null of lethality.

Although a pair of individuals with similar variations in a given gene is not often found, the likelihood that any two individuals will have a defective variation in any one of the 7,000 genes is much higher. The result is that about 1 in 15 persons are afflicted with a rare disease condition. So, in aggregate, it turns out that rare diseases are a highly common health care issue that suppresses quality of life for a large proportion of the planet’s population.

The challenge of today’s personalized medicine is to develop therapeutic approaches that can treat these rare disease populations in a cost effective way. In the traditional therapeutic development path, the cost to bring a chemical entity through the challenges of toxicity and efficiency and have it reach the market, has cost billions of dollars. This sizable sum of money can be challenging to justify when only a dozen or so individuals will stand to benefit from such a high cost. As a society, we must get inventive and find more affordable approaches to bring rare disease therapeutics to market. Answering this call to action are small biotech companies, such as Perlara.

Perlara, located in the San Francisco Bay Area, was founded in February 2014 by Ethan Perlstein. The company is “on a mission to accelerate the discovery of cures for rare genetic diseases and uncover underlying mechanisms that enable the development of treatments that work across a range of diseases and individuals” (Perlara Website). Through their PerlQuests™, Perlara partners with families, researchers, and patients to find treatments for these otherwise forgotten, yet very real, rare diseases. One such PerlQuest, focusing on PMM2 (Phosphomannomutase 2 deficiency), holds a special place in the heart of Dr. Sangeetha Iyer, the Director of Preclinical Development at Perlara.

Dr. lyer (currently a senior PM at Pfizer) has a background in neurodegenerative disorders and rare genetic diseases. She received her Ph.D. in molecular pharmacology from the University of Pittsburgh and then went on to do her postdoc research at the University of Texas, Austin. Sangeetha has worked with a plethora of models, starting her career with mouse models, and working with xenopus oocytes, drosophila, and C. elegans. Not only does she have a wide understanding of model organisms, but Dr. lyer also has had over 10 years of experience in model and assay development and drug screening for human genetic disorders. 

Dr. lyer developed nematode models of rare diseases and conducted several successful screen campaigns for rare diseases, one of which was the aforementioned PMM2, the Phosphomannomutase 2 deficiency. These efforts led to a clinical trial with one patient, Maggie (n=1). The trial outcome was a success! Maggie is doing quite well and the trial is currently being expanded to include other patients. Recently, these findings were published in a paper titled: Repurposing the aldose reductase inhibitor and diabetic neuropathy drug epalrestat for the congenital disorder of glycosylation PMM2-CDG.

What is PMM2?

PMM2-CDG, formerly known as congenital disorder of glycosylation type 1a, is a rare multisystem disorder that involves a normal, but complex, chemical process known as glycosylation (PMM2-CDG 2015) 

PMM2-CDG is caused by mutations of the phosphomannomutase-2 (PMM2) gene and is inherited as an autosomal recessive condition. The variation reduces the function of the PMM2 enzyme and leads to improper levels of glycosylation. The disease can affect any part of the body, though most cases usually have an important neurological component. PMM2-CDG is associated with a broad and highly variable range of symptoms and can vary in severity from mild cases to the severe, disabling or life-threatening cases. Most cases appear in infancy or early childhood, like in Maggie’s case, thus, this patient population became the focus of Perlara’s PerlQuest.

Maggie’s Quest

The PMM2 PerlQuest came about because, prior to meeting Maggie, Dr. lyer had started working on lysosomal storage disorders. The research team had just completed some work on a glycosylation disorder (NGLY-1), whose loss of function leads to developmental delay and seizures. In presenting this work, they were introduced to the Maggie’ parents, as PMM2 deficiency is one of the most common causes of glycosylation disorders.  Sangetha and the team at Perlara felt this was a good candidate for using humanizing mutations to create an animal model of the disease in a simple model organism.

“Some model systems work, but not a whole lot and we felt that it fit very well with our platform of model organisms. Perlara was working with yeast model systems, drosophila as well as C. elegans along with the basic model organism pipeline and on the other hand, we had patient fibroblast, which were also available for PMM2. So PMM2 basically came onto our radar because of Maggie, the girl who has PMM2 disorder. And after meeting with her parents and having some conversations about the utility of our platform, we decided to go ahead and model some of our mutations and see if we could conduct a drug screening campaign. That’s how that program came to be.” (Iyer, Podcast: 17 minutes of Science, 2020)

To pursue PMM2 treatment options, Sangeetha and her team at Perlara decided to go down the path of drug repurposing. This involves testing already known drugs and compounds in alternative ways to see if they are viable treatment options for other diseases outside of their initial purpose. One of the main benefits of drug repurposing is that it speeds up the traditional drug discovery timeline as the compounds are already known, and often have vast amounts of information already available for researchers to use. As a result, drug development can be achieved while cutting costs immensely: instead of the typical hundreds of millions it takes to reach clinical trials, drug repurposing trials can be conducted with millions or even sub-millions.

The process:

To begin their repurposing campaign, researchers at Perlara initially started testing with yeast models of PMM2. They then wanted to move into C. elegans, but hit a roadblock. C. elegans already had one PMM2 model, but it was lethal which meant it would not be a viable tool for their campaign. At this point, Perlara approached InVivo Biosystems for help building a new C. elegans model to use. “With InVivo Biosystems’ help, we were able to model another, a different patient mutation, one that does not have this severe lack of enzyme activity”(Iyer, Podcast: 17 minutes of Science, 2020). The new worm model that InVivo Biosystems built for Perlara has an ortholog for PMM2 in which the protein is 54% identical to humans. This may not sound like a lot, but it is very significant. Additionally, and possibly most importantly, the mutation sites were conserved between humans and C. elegans. Because of this conservation, Perlara was able to use the C. elegans models engineered with a specific point mutation that modeled the exact mutation seen in the patient.

“One of the reasons we believed in the power of model organisms specifically for rare monogenic diseases was because when you have a single gene ortholog and one that has high similarity to what one might encounter in humans, you can model the same mutation as you see in humans, in those model organisms.” (Iyer, Podcast: 17 minutes of Science, 2020). 

Dr. lyer performed a drug screen using a 2560-compound Microsource Spectrum library consisting of FDA-approved drugs, bioactive tool compounds and natural products. The top 20 hits were found to be either antidiabetic or antioxidant molecules. Remarkably, the dominant portion of hits were antioxidant flavonoids with known utility as aldose reductase inhibitors (ARIs). Next they checked the activity of the leads in yeast and patient-derived fibroblast to see if cross species conservation of activity can be observed. The result was identification of α-cyano-4-hydroxycinnamic acid (CHCA) as the most cross species bioactive for ARI activity. The structure activity relationship of the CHCA molecule was used to develop a drug pharmacophore profile which allows identification of a set of commercially-available ARIs (tolrestat, ranirestat, imirestat, zopolrestat, sorbinil, ponalrestat, alrestatin, fidarestat and epalrestat).  Testing these new molecules in worms and patient fibroblast revealed epalrestat as the best activity lead.

The aldose reductase is an enzyme that shunts glucose down the polyol pathway with its conversion of glucose into sorbitol. The inhibition of this enzyme activity has two favorable effects. It leads to an increase in glucose-1,6-bisphosphate production which activates PMM2. And it prevent activation of the polyol pathway activity which generates Reactive Oxygen Species (ROS) and leads to Advanced Glycation End-products (“AGE”).  The AGE are especially nasty because they create abnormal protein glycosylation and cause a normal protein to be recognized by the immune system as “foreign” protein. Then a cascade of auto-inflammatory response is initiated.

It is likely PMM2 deficiency set in motion quite a few cellular stress responses. Not only does it result in reduction in normal levels of N-linked glycosylation, it also results in increased shunt of glucose through the polyol pathway. This leads to high levels of sorbitol which readily alkylates with the amines in the body’s proteins rendering the proteins a “foreign” in appearance to the immune system and inflammation results. This ripple effect of cellular stress stress leads to the clinical presentations of the disease. 

“Until the time that we did this work, nobody had discovered that you could boost PMM2 enzyme activity through some other artificial shunt pathway, which is essentially what our model organism screens were telling us, that there was another way to increase PMM2 activity.” (Iyer, Podcast: 17 minutes of Science, 2020).

And while this was true, they had discovered another way to increase PMM2 activity in both their yeast and worm models, they were not certain how this would translate to the human enzyme. At this point, Dr lyer and her team at Perlara were able to incorporate the human fibroblasts which were known to have the defective enzyme activity. The team used both generic Fibroblasts from Coriell, a non-profit repository for patient fibroblasts, in addition to using the fibroblasts from Maggie. In both cases, the fibroblasts showed that PMM2 enzyme activity was in fact increased when exposed to epalrestat. According to Dr lyer, “that’s where the final piece of the puzzle came together.”

Maggie’s Cure

One of the benefits of drug repurposing over developing new compounds is that these drugs are already on the market and have had a wealth of safety data collected. While not always the case, this typically makes it easier to get approval to use the drugs. Although epalrestat had never been approved for use in the US, it had been on the market for over 20 years in other countries and was no longer under patent protection.

At this point, Perlara started talking with Maggie’s family and Dr. Eva Morava from the Mayo Clinic to see about treatment opportunities for Maggie using epalrestat. Dr. Morava, along with Maggie’s parents and Ethan Perlstein, put together an n=1 IND application which they submitted to the FDA in order to gain approval for the use of epalrestat. Thanks to the trove of safety data that was available for epalrestat and the body of data generated by Perlara substantiating that epalrestat increased PMM2 enzyme activity in a variety of modern systems, they were able to gain approval for the trial. 

Maggie has now been on the drug for over a year (Interview conducted when Maggie had been on the drug for about 10 months). She has gained weight and can have conversations, something she was unable to do pre-treatment. Her motor skills and coordination have skyrocketed — she can even ride a bike now. Maggie is continuing to take epalrestat, and her team is now working to expand the trial to a larger group in the hopes of helping others. Dr lyer credits the success of this drug repurposing study to the model organisms which were able to generate the data they needed quickly, efficiently, and affordably in order to gain their FDA approval. 

“To see a drug have an impact, a positive impact, on the child, in a child was just incredibly powerful” (Iyer, Podcast: 17 minutes of Science, 2020)

References

1. Sangeetha Interview: https://invivobiosystems.com/17-minutes-of-science/from-bench-to-bedside-using-model-organisms-to-find-rare-disease-treatments/?highlight=sangeetha
2. PMM2-CDG. (2015, August 06). Retrieved April 28, 2021, from https://rarediseases.org/rare-diseases/pmm2-cdg/#:~:text=Summary,chemical%20process%20known%20as%20glycosylation.
3. Iyer, S., Sam, F. S., DiPrimio, N., Preston, G., Verhejein, J., Murthy, K., . . . Perlstein, E. (2019, November 11). Repurposing the aldose reductase inhibitor and diabetic neuropathy drug epalrestat for the congenital disorder of glycosylation PMM2-CDG. Retrieved from: https://journals.biologists.com/dmm/article/12/11/dmm040584/223279/Repurposing-the-aldose-reductase-inhibitor-and
4. Aldose reductase inhibitors for the treatment of Diabetic polyneuropathy. (n.d.). Retrieved April 28, 2021, from https://www.cochrane.org/CD004572/NEUROMUSC_aldose-reductase-inhibitors-for-the-treatment-of-diabetic-polyneuropathy#:~:text=Aldose%20reductase%20inhibitors%20are%20a,or%20reverse%20progression%20of%20neuropathy
5. Aldose reductase inhibitors for the treatment of Diabetic polyneuropathy. (n.d.). Retrieved April 28, 2021, from https://www.cochrane.org/CD004572/NEUROMUSC_aldose-reductase-inhibitors-for-the-treatment-of-diabetic-polyneuropathy#:~:text=Aldose%20reductase%20inhibitors%20are%20a,or%20reverse%20progression%20of%20neuropathy

You know when that hunch seems to get reinforced over and over again, then your mind starts speculating it as a fact.

!Danger! Will Robinson… it’s time for a serious fact check.

My hunch was that the amino acid arginine (Aka: “Arg” or “R) seems to be showing frequent association with pathogenicity. It started with the observation that many of the established pathogenic variants in the coding sequence of STXBP1 seem to involve a preference for arginine. Extracting from ClinVar for missense that are pathogenic and likely pathogenic gives the following table:

Indeed arginine (R) is disproportionately represented. Assuming all amino acids as equals, then there should be 4.3 for each amino acid. Disproportionally low are things that make sense. Like methionine (M), only one codon (ATG) instructs for insertion of this amino acid in a sequence. Similarly tryptophan (W) also has only one codon (TGG). These two amino acids should be represented below the average. A little bit oddly, we have similar low levels from lysine (K), phenylalanine (F) and glutamate (Q) who each have two codons. If codon dosage was key to variant proportioning, then these should have been seen at least 2x more than M and W, so perhaps something more than codon dosage mediates amino acid choice in creating pathogenic variations.

Arginine has 6 codons which still could drive its outsized proportion in the graph. Yet Serine (S) and Leucine (L) also have 6 codons. But respectively they are at 7 and 3 for being involved in pathogenicity. Only mighty arginine accounts for 13 of the 43 pathogenic variants in STXBP1 (30%). Tempering my enthusiasm is the observation that for 3 amino acid positions R292, R406 and R451, we have multiple changes being called pathogenic. Yet no other amino acid in the STXBP1 pathogenics has this changling capacity, so why is it that arginine is at high proportion in the assigned pathogenics – perhaps it is just a consequence of a biased investigator focus specific to STXBP1 and they fixed their gaze onto the repeating de novo clinical variants at positions 292, 406 and 451.

Is arginine involved in fragility elsewhere in the genome?

To normalize for possible investigator bias and find a method that can be applied to other portions of the genome, I took advantage of the Ensembl database to list and rank a gene’s codon sequence variants by bioinformatics analysis. Ranking on CADD was used to list protein coding sequence variations by their severity.

Ensembl allows us to identify which variations are theoretically likely to be disruptive of protein function. The choice to rank by CADD (stand for Combined Annotation-Dependent Depletion) allows us to use a sophisticated algorithm that avoids investigator bias because it intentionally avoids using “known” pathogenicity databases when it creates it ranking. A key test is to see if CADD can independently observe the pathogenicity known to exist in STXBP1. To construct the test, we compare the top scoring CADD variants with the lowest scoring CADD variants.

With CADD, we get an independent call for possible pathogenicity that still picks up what you might expect. Nearly half the calls in the Top-30 CADD pull up known pathogenicity and no benign calls are found. In the Bottom-30 CADD we get one known benign call and no pathogenics.

Healthy population data also is consistent. STXBP1 is autosomal dominant. That means you only need one of your two chromosomal copies to be defective and disease will occur. Selection pressure has been very tight on autosomal dominant genes. Variants in healthy population cannot occur at higher than the known frequency of the disease in the population. Published frequency in STXBP1 for causing early-infantile epileptic encephalopathy is 1/90,000. The largest healthy population database is in GnomAD. At 141,456 individuals, and the fact that STXBP1 needs to distribute across at least 43 pathogenic alleles, the likeliness of even one pathogenic variant being in healthy populations is pretty close to zero. Some of our Top-30 CADD have 1x or more frequency in healthy populations. Most of these are unassigned. For these unassigned that are seen at 1x or more, the disease frequency argument strongly implicates that they are benign variants.

So the CADD is not perfect, the top scoring hits are a mix of known pathogenic and probably benign. But the bottom scoring CADD seems to be more efficient at pulling out benign. In the Bottom-30 CADD, only one variant, I271V, is labeled Likely Benign by ClinVar, yet nearly everyone of these alleles (27 of 30) is seen in healthy populations, so they too are probably benign.

At this point in the analysis, we can pinpoint an anomaly. Y264C is labeled in ClinVar as a Likely Pathogenic. But from the population frequency argument, this assignment is highly unlikely. Y264C has been observed to occur in healthy populations. So a a bare minimum, it should be downgraded to a VUS, but probably be called a Likely Benign for causing early-infantile epileptic encephalopathy.

Finding Arginine-associated Fragility Throughout the Genome

This top-30 / bottom-30 approach was applied to a large set of genes. As a form of internal control, we add isoleucine (I) in the screen. With less conviction, I have felt this amino acid was associating with benign variants. If true, it should show an enrichment in the Bottom 30 CADD scores. So in my gene set experiment, I measured 4 bins. 2 bins for how many arginine and isoleucine in the Top 30 and 2 bins for how many arginine and isoleucine in the Bottom 30.

30% of top 30 CADD scoring variants contain arginine???!!!

An assumption of even distribution of amino acids, combined with an even more absurd assumption of an average 3.05 codons per amino acid, gives us 4.3% as average amino acid fraction per each 30 (dashed line). Arginine is 7.2x more than this average number. Yet, we need to account for the fact arginine uses about 2x more than the average codon usage. A a result Arginine bias in the Top 30 is about 3.5x more than expected. For isoleucine, the enrichment in the bottom 30 appears to be about 2x more than expected.

Test dataset – 30% arginine in Top-30 CADD prevails

The noisiest data in the Top-30 CADD appears to be the Arginine data. A cumulative trending plot was used to see how many genes were need before the trend to 30% becomes apparent. After assessing 7 genes the trend starts to stabilize. A new set of 7 genes were chosen. This time the genes were chosen from the Undiagnosed Disease Network (UDN). The UDN recently listed 54 genes as in desperate need for animal modeling to provide gene function studies. A sub-selection of these were identified as having good sequence similarity to genes in the animal models which we hold dear to our heart and expertise (zebrafish and C. elegans). The Top-30 / Bottom-30 CADD selection was applied to these genes and plotted for Arg and Leu enrichment. 30% prevails for arginine – it occurs at least 3.5x more than expected for being the top CADD variants as hypersensitive to substitution.

This all assumes that the representation of amino acids is uniform across all proteins. But the that is a reach. Louis Gross at University of Tennessee, Knoxville, has observed the amino acid distribution in vertebrates has some anomalies.

Most notable anomaly is arginine. 6 codons are use by arginine, but the observed frequency is low at 4.2%. To illustrate how low, they calculated the expected frequency for each amino acid biasing only for the GC richness of vertebrate genomes.

The expected frequency for arginine is quite high at about 10.5% due to its GC richness in its codons. Yet the actual observed frequency is quite low at about 4%. Based on this observed frequency, we bounce back – we now assess that we are observing arginine in the top 30 at 8x more than expected. No explanation for the anomaly and it just became more pronounced!

Taking a different approach, we can ask what percentage of ALL known pathogenic and likely pathogenic variants in a gene involve arginine substitution. 7 genes analyzed and we get the same 30% for arginine. Yet the calculations are that it should be below 4%. 8x more than expected prevails.

Are your arginines special too?

This analysis has uncovered a unique phenomenon. It appear everyone’s arginines are special. Exactly why arginine has this special status is not entirely clear. It is highly likely arginine has been strongly selected against its random incorporation during evolution. As a result of this strong negative selection (much more than what is happening for all other amino acids), arginine’s frequency in all proteins is much lower than predicted. The observed pathogenic sensitivity may be a read out of this hyperselectivity of evolution. Basically, arginine’s use in any given protein is very particular. A possible driver for this is arginine’s amazing capacity to bring high order to neighboring side chains in most protein structures. When it is gone, chaos reigns. When it is introduced where it should not be, chaos still reigns.

Arginine is special. I suggest we need to ditch Douglas Adam’s “42”.

Instead, we make like a pirate and just say:

“Arrrrrrrrg”

How prevalent are Variants of Uncertain Significance?

ClinVar database for variant interpretation was analyzed for its levels of ACMG-AMP assessments. With help from the data dumps from ClinVar Miner, the yearly distribution of assessments was plotted. Since 2016 and shortly after the ACMG-AMP guidelines came out in 2015, the number of assessments assigned to the VUS category has grown rapidly. These are the variants that clinical genetics researchers have examined, but cannot decide if they are pathogenic or not.

How big will the VUS problem get?

To estimate how large the VUS problem will become, we must first understand how big is the human genome. Controversy abounds, but current estimate are there are 21,306 protein coding genes and 21,856 non-coding genes. To be conservative, and for simplicity sake, let us use 20,000 genes as the number. The next question is how many of these are disease associated. When we look to ClinVar the number of “genes with variants specific to one protein-coding gene” we get 7221 genes. More conservatively, we can look to ClinVar’s “gene_condition_source_id” which list 4242 genes as being associated with a diagnostic condition. This lower number is reinforced by OMIM in which the “Total number of genes with phenotype-causing mutation” is 4162 genes. These list have been growing rather steady at 5% per year, so in a few years the likely number of gene-disease associations will probably approach 5000 genes, or roughly 1/4 the human genome.

VUS problem may eventually approach 7 Million variants

A recent attempt to preload the human genome with pathogenicity assessment potential has been made. InterVar database applied ACMG-AMP guidelines to ~80,000,000 amino acid positions in the genome to provide a database for easier variant interpretation. Since at least 20% of these positions are likely to be in genes with known disease association, there are roughly 16,000,000 variants that will eventually occur in patient-derived genome sequencing. If the current trend of 44% VUS translates across that number, then there will be close to 7,000,000 variants in need of functional studies to resolve their pathogenicity.

A novel animal model systems for rapid variant interpretation

The team at Nemametrix just produced a wonderful set of preliminary data that we showed at the recent American Society of Human Genetics. It shows it is possible to use a training set of known benign and pathogenic alleles in a gene to “teach” a ML algorithm to determine if pathogenicity is present in a VUS. When applied to the STXBP1 gene, a set of 5 benign and 5 pathogenic was sufficient to train for segregation in an LDA plot and the Y75C was assessed as pathogenic.

Once this type of system is trained with a set of known pathogenic and benign variants, the assessment of pathogenicity can be achieved in a soon as 10 days from start of a VUS transgenesis project.

In a prior blog post, the presence of dominant alleles in my genome gave me pause when trying to interpret the data from sequencing my DNA. Dominant alleles can be the cause disease when only one pathogenic variation occurs in only one gene copy of the chromosome pair. Contrast this to a recessive allele where you must get a defect in both chromosome copies of the gene to cause disease. In the recessive condition, if you only have one defective copy, you can expect to remain healthy, but you are a carrier of a disease allele. With the lack of immediate consequence to being a carrier status, many more individuals should be walking around with variations that are recessive towards disease. In fact, the CFTR gene variation (p.Arg117His) for Cystic Fibrosis that was highlighted for me in my Veritas Genomic sequencing report is quite common. It occurs globally at 1 per 2,500 persons, and that increases close 1 per 1,000 for northern europeans, which is a dominant portion of my ancestral genomic composition. In contrast, the CACNA1S variant (p.Arg419His) that most concerns me in my genome, has a prevalence of 1 in 25,000. Thats of low enough to be Rare Disease in Europe, but still probably way to high for disease manifestation rates.

Rare domination in CACNA1S needs to be rare enough to cause Hypokalemic Periodic Paralysis.

Dominant disease causality with the Arg419His variation in CACNA1S is unlikely because it is too frequent for the 1 per 100,000 population frequency for the disease of Hypokalemic Periodic Paralysis. Yet there are two variations known to be causative in CACNA1S, Arg528His and Arg1239His. Arg528His occurs at close to 1/100,000, while Arg1239His has yet to be detected in healthy populations. Clearly the Arg11239His is low enough population frequency to be causative for Hypokalemic Periodic Paralysis. Yet for my Arg419His, the frequency is too high for it to be causative. A variant effect that is Autosomal Dominant (AD) is extremely unlikely for my lone Arg419His allele.

If dominant alleles need to be rare in the population, how frequent is dominant status for variants of a disease?

The frequency of Autosomal Dominance (AD) for any given disease gene appears to be quite high. It is estimated that there are about 7000 Rare Diseases. If we assume the On-line Mendelian Inheritance in Man (OMIM) already represents most of these genes, then rare disease variants will map to the 4346 gene entries in OMIM with published allelic variations. Next, I listed these variations in blocks of 100 to reveals the number of genes for which they are known to exclusively Autosomal Dominant (AD) or Autosomal Recessive (AR), or some kind of hybrid.

When one runs down the inheritance pattern and tabulates them per gene, the first 100 variants have about twice as many genes in the AR category when compared to the AD category.

Running thru the another 400 more variants in the 100 variant blocks shows the trend continues – Dominance of a genetic conditions occurs for about 1/3rd of the disease genome.

Axiom for the individual : “I am not very dominating, but there are lots out there who are.”

So at the individual basis, it appears the AD status of pathogenic or likely pathogenic variants in your genome is very rare. Yet, at a population level, a large proportion of Rare Disease is caused by Autosomal Dominant variation. Rare disease calculate to occur at about 1 per 15 persons. So, for about 1 in 50 (150 million persons), their disease casing variation is likely to be Autosomal Dominant.