Drug Discovery & Development
LabRoots Virtual Event

WEBINAR TRANSCRIPT

Hello, everyone! Thanks for joining me. My name is Mark Kiel, and I’m the chief science officer and founder of Genomenon. Today I’d like to present to you results from our comprehensive amyotrophic lateral sclerosis genomic landscape and its implications for drug development and discovery.

The outline of my talk today is to begin by introducing ALS, its disease presentation, and its pathophysiology with the discussion of the clinical and laboratory diagnostics, then a view into the current and future landscape of treatments for ALS, followed by a discussion of what genomic landscapes are, how they’re produced, and how they’re used. Then we’ll go into some detail talking about the characteristics of the ALS genomic landscape that Genomenon has developed, focusing on some specific genes and their variants and calling out some representative insights from that landscape. I’ll conclude with implications and applications of this new data.
So to begin, let me talk about the clinical features of ALS. This is a rare disease with a prevalence of about five total cases per 100,000 individuals across many different ethnicities. It is a progressive paralytic neurological disorder characterized by degeneration of motor neurons in the central nervous system. The symptomatology includes cognitive and behavioral impairment, and eventual respiratory compromise and death, usually within three to ten years of initial presentation. In terms of disease mechanisms, ALS is characterized by neuronal cell death and dysfunction that occurs through a variety of different pathways, including altered protein homeostasis, neurotoxicity mediated by RNA deposition, as well as altered cytoskeletal dynamics and other aspects of neuronal biology. These usually culminate in neuroinflammation, hyperexcitability, glial dysfunction, axonopathy, oxidative stress, etc.

In terms of the causes of ALS, there are two forms: sporadic, which comprises 90 percent of cases and is often attendant by presumed environmental factors, and a less frequent but real familial form that comprises 10 of cases. There have been many genes that have been implicated, each with their own unique functions, in both forms of ALS. This was highlighted in a recent New England Journal of Medicine review that I’m showcasing here, where you see on the x-axis, the year of discovery of each of these genes in the graph, and as those genes accumulate over time, you see that graph increment up on the y-axis. Specifically, the size of each of those icons is a reflection of the relative proportion of cases that are marked by mutations in any one of those genes, with the genes C9orf72 and SOD1 causing the majority of these familial cases, followed by a long tail of multiple additional genes, as many as 25 or more.

In terms of diagnosis, ALS is a complicated diagnosis with a vast differential diagnosis of additional distinct neurological disorders, some of which I’m highlighting here. Part of the challenge with understanding ALS diagnosis involves a lack of specific criteria that can help differentiate from among other neurological disorders. Those criteria that have been established are largely clinical, and they comprise positive criteria as well as negative criteria. For positive criteria, patients must exhibit both upper and lower motor neuron degeneration and evince progression of disease. For negative criteria, this involves a rule out of any one of those additional diseases in the differential diagnosis for ALS, usually through somewhat invasive laboratory testing, including electrophysical and pathological laboratory findings, but also including less invasive radiographic techniques.

In addition to being a challenging diagnosis, there are categories of diagnostic certainty ranging from “definite” all the way through to “probable,” “possible,” and/or “suspected.” Part of the challenge here is this lack of objective laboratory biomarkers, the need for which is urgent for both diagnostic as well as prognostic purposes.

Moving on to the clinical complexity and diagnostic challenge associated with ALS is the lack of effective current therapeutic strategies. So these are not my words, but to date, there have been few, if any, commercial therapies for ALS that offer a substantial clinical benefit. Most patients are treated by managing symptoms and benefit greatly from timely intervention to improve quality of life and to forestall demise. There have been a number of recently repurposed drugs applied to the treatment of these ALS patients, but the most exciting recent developments in treating these patients involve genomic therapies that are targeted to some of those specific genes that I alluded to as causing ALS. In particular, as recently as summer of 2020, two New England Journal of Medicine studies (one, a case report, and one, the results of a phase 1-2 trial) have demonstrated that patients benefit from specific genetic targeting, specifically, patients who have mutations in the SOD1 gene. This includes adeno-associated viruses, or AAVs for SOD1, as well as an antisense therapy called Tofersen, also used to specifically target SOD1.

These advances are signaling a new beginning for ALS therapeutics in which some forms of the disease may become treatable for subgroups of patients with specific genomic features or genotypes. That will obviously require sequencing of these patients’ genomes using a panel or a whole exome platform or a whole genome platform. As thousands of these new sequences become available, it’s feasible to imagine that all of that data across all of those patients, with a better understanding of all of the unique gene variants that cause or are associated with their diseases, not just for familial cases, but as well as sporadic cases, will open up new lines of inquiry and a broader understanding of this complex disease to better inform novel therapeutics.

This is a review that was published as recently as a month ago in early 2021. There is a shift in the “one-size-fits-all” mode for clinical trials in ALS to more of a nuanced patient stratification approach with precision therapeutic application. This will obviously require detailed and carefully recorded phenotypic and genotypic understanding of each of these patients to therefore guide trials and treatments under the auspice of these precision medicine approaches. There are challenges associated with applying a precision medicine model to these clinical trials. I mentioned the genetic complexity and clinical heterogeneity of ALS, which is attendant by an incomplete understanding of all of the disparate disease mechanisms that culminate in symptoms of ALS and the urgent need for and lack of sensitive genetic biomarkers, as well as the challenges associated with accurately diagnosing ALS patients, relying as we do just on clinical parameters at present, and rule-out of other diseases through laboratory techniques. When we’re talking about the future of ALS diagnosis and treatment, all of these challenges culminate in a sub-optimal clinical design paradigm, which in the future will lead to challenges to adequate treatment of these patients.

Into this picture enters a genomic landscape. I’ll go into a much greater depth about what a genomic landscape is shortly, but you can imagine now a genomic landscape at the disease level is a thorough understanding of every gene associated with causation of one disease, and every one of its genetic variants, and all of the idiosyncrasies and nuanced associations, even down to the variant level, for each of these individual patients. A genomic landscape for that disease would be a thorough understanding of all of that complexity and would benefit researchers, trial designers, clinicians and patients by uncovering more targets, allowing us to identify more patients and de-risk clinical trial outcomes, in addition to allowing us to gain a more thorough understanding of all of the different disease mechanisms and how they may converge or diverge.

So specifically, we’re talking about a comprehensive genomic landscape for ALS, that is to say, a landscape that comprises every ALS-associated gene and each of their genetic variants that has ever been published, all of which are interpreted after curation of every relevant piece of clinical and/or functional information which are usually patterned in the empirical published literature. That is exactly what Genomenon has done for multiple genes and multiple diseases over the past many years. The results of our work for ALS are what I want to showcase today. The way that we produce these datasets is through a unique combinatorial approach that first leverages artificial intelligence to organize, annotate, and cue up all of this evidence very comprehensively and very sensitively across every paper ever describing any ALS patient. We then follow that organization of evidence with an expert, expedited manual review that allows us to do this comprehensively and in very short order — in this case, in a matter of a few months.

Just a little bit more detail about that approach: especially in the context of neurological disease, a genomic diagnosis for patients may be new territory for clinicians and trial designers. It’s not new territory for other diseases, particularly, rare diseases, and in fact there are frameworks codifying the way that genetic or genomic sequencing information for a particular patient needs to be interpreted using that evidence to ensure the utmost reliability of those interpretations. What I’m showcasing here is built on the framework devised by the American College of Medical Geneticists and Genomicists, or ACMG guidelines. I’m illuminating just a little bit behind the scenes about Genomenon’s application of that set of guidelines, relying as it does on a triad of evidence, including the population frequency, the predictive models of pathogenicity for each of these variants based largely on evolutionary conservation, and then the apex of that triangle is all the evidence in the published literature that usually speciates into clinical information, patients, cohorts, and their association with disease and their clinical phenomenology, as well as the functional studies that have tested those variants in various in vitro and in vivo models of protein activity. I wanted to clarify here a little bit what that ACMG framework is. I won’t go into any detail about how genomic has semi-automated this entire process, again, culminating in an expert manual review of all of that evidence to ensure the utmost accuracy of the results.

So again, we’ve done this across every gene associated with ALS. In the previous slide from the New England Journal of Medicine review, now, about four years ago, say, which had an apex at about 25 or 26 genes understood to be associated with ALS. Since that time, and also resulting from our unique efforts, there are about 10 more genes that are now known to be associated with ALS at some penetrance. I’m listing many of them at the upper right, but again, when we produce this genomic landscape, we look not just at the gene level, but also in detail at the variant level. So across those 36+ genes that have been fully curated according to the ACMG standards providing evidence and annotations for all of the clinical characteristics of those patients, as well as the pathogenic mechanisms of the functional studies, all of that supported by the scientific evidence available to us — and as I’ll showcase in a slide later, summarized for efficient reporting and kept up to date — all of that work has culminated, after review of 21,000 variants, in 7,000 variants across these three dozen genes that are associated with ALS, including 1,800 that meet the clinical grade criteria for ascribing a variant to causing a patient’s disease. That’s referred to as “pathogenic” or “likely pathogenic” by the ACMG criteria.

In a little bit more depth, this slide is a reflection of how exhaustively annotated each of these variants and their associated reference citations are articulated in the ALS genomic landscape. You can see the provisional call according to those ACMG criteria after review of every one of the single, dozens, hundreds of articles that mention any one of these variants. There’s an annotation of all of the clinical and functional features as you can see there in the middle on the left. There’s also categorized and enumerated evidence categories from the ACMG criteria, in all of their associated rigorous guidelines and rules. Then there’s very detailed and thoroughly annotated explication of the clinical and functional findings for each of those cases or functional studies.

So I alluded to the overall yield of the ALS-associated genes. I’ll highlight here, as I mentioned before, that there’s a vast number of variants from among those genes that are now clinically associated with ALS, based on the evidence that we’d assembled and reviewed. But there’s also a very large number of variants that were found in patients and associated with disease, but for which there’s yet insignificant evidence to pass that clinical threshold, but which nevertheless and especially in aggregate are very relevant to better understanding some of the idiosyncrasies of the ALS disease. Obviously we have all of this information thoroughly annotated for each of those individual genes, as I’m showcasing here for one of the more prevalent genes associated with familial ALS, which is SOD1. To take you a little bit deeper into the nature of that evidence, I want to showcase some of the findings for each of these representative genes that I’ll be calling out on the next slides.

So in particular, at the upper portion of this slide, I want to draw your attention to the fact that the ALS genomic landscape produced by Genomenon has uncovered three times more variants, each of which is associated with a vast body of evidence, compared to the number of variants for SOD1 that are found in ClinVar, which is a collection of user-submitted variants that are seen in clinical practice. When you look at the blue crescent and the grey moon at the right, you can see how the data that Genomenon has produced in the ALS dataset is actually a superset of those ClinVar variants, and in many cases, can reconcile discrepancies or conflicts based on ClinVar user submissions. So three times more variants available in the genomic landscape produced by Genomenon than otherwise would be available in ClinVar.

Then, if you’ll allow me to draw your attention to the lower panel, you can see the linear axis of the SOD protein and its functional domain at the bottom. Then, in the two panels in gray and blue for ClinVar and Genomenon, respectively, I’m showcasing as each one of those dots a unique variant found in a paper or hundreds of papers and how they distribute across that linear axis of the protein, and in this case, how they influence the functional domain of SOD1. I’m suggesting here that only with a complete genomic landscape will you be afforded a view into the patterns that may emerge and the unique activity of each of these variants as they influence protein functionality.

Quickly, for TARDBP, there are four times more variants in the Genomenon ALS landscape compared to ClinVar. In particular, if you look at the protein diagram at the bottom, there’s a dearth of variants represented in ClinVar in multiple of the functional domains in TARDBP, whereas in the genomic landscape from Genomenon, there’s a significant number of variants associated with patients and often with functional studies that can indicate how that domain in that protein can be mutated to cause disease.

Lastly, for a less prevalent gene which is found to vary in ALS patients, the TBK1 gene is five times more represented in the Genomenon ALS landscape compared to ClinVar, and again, you see the value of that representation in a more comprehensive landscape being reflected in the patterns that you see across the different domains of TBK, including the TANK domain there at the right.

There have been a number of genomic associations associated with ALS in the literature before. However, they have been small in number, and the evidence supporting those associations has been meager. Nevertheless, the data that has been presented does suggest that there may be genomic associations with ALS, as I’m highlighting in this table from a recent New England Journal of Medicine review. Specifically, I’m drawing your attention to two such purported genetic associations, each from a single study. The first is a FUS variant, the p525l variant, which was found in a series of 11 patients to suggest that patients with that variant present many years earlier and have a shorter survival course. Disease progression is also accelerated in patients who have a mutation in the SOD1 gene, the a4v mutation. These hint at the idea that there’s value in better understanding a comprehensive genomic landscape in ALS to divine what these clinical associations may be.

Armed with all of the data that comprises the ALS genomic landscape across those many tens of thousands of references, and the 36 genes and all of their many thousands of genetic variants, we took a closer look at all of that data, specifically focusing on each of those clinical cases that were described. Those number more than 2,000 individual patients in total. This is in contrast to the content in the table that I showcased earlier, which was looking at single studies and couldn’t possibly have aggregated all of that data without an artificial intelligence assisted platform, such as I described earlier employed by Genomenon.

Just to orient you to the data that we were able to understand here from our dataset: this is now a distribution of the ages of onset by decade of life on the y-axis, for each of those patients on the x-axis out of the 2,006 patients, and the distribution of those ages of onset for all patients in our dataset with familial ALS. What you can see here from the histogram is a normal distribution of those age ranges with a mean of 56 years for all patients in our data. Obviously, we have all of this data precisely defined at both the gene and the variant level, and so specifically when we’re talking about the distribution for patients with SOD1 mutations, what you see is an earlier age of onset, about a dozen years earlier across the 609 patients in our dataset that had SOD1 variants.

There’s an opposite trend toward an earlier presentation of disease for patients who have mutations in the FUS gene, as you can see here. A mean of 38 years across those 190 patients in our dataset with FUS mutations. Calling your attention back to that table, table one, with the suggestion of genomic associations particularly for FUS p525l from that single case series of 11 patients. This suggests that, not only is there corroborating evidence that the FUS variant is associated with an earlier age of onset, but even more broadly, any variant in the FUS gene, based on this data, you might interpret to be associated with an earlier age of onset, and perhaps a worse pace of progression for the patients, conferring a worse prognosis. So, to our knowledge, this is a unique assembly of such a comprehensive dataset to be able to understand these associations at scale, again, to leverage the quantity and quality of that data to better understand these patterns that might reconcile some of the clinical and functional heterogeneity of ALS patients.

In contrast, the TBK1 gene, which I spoke to on a preceding slide, this has a more average normal distribution, with an age of onset of 54 years across all 128 patients who have TBK1 mutations. I should mention here, TBK1 is associated with other diseases in the spectrum of neurological diseases that have substantial overlap with ALS, specifically, for TBK1, that’s frontotemporal dementia, high on the list of the differential diagnosis for patients with mutations in TBK1.

Not only do we have information at the gene level, but we also have more precise information at the genetic variant level, which I’m showcasing here for SOD1. Again, just to orient you, on the x-axis is the linear axis of the SOD1 protein with its functional domain, and then on the y-axis are those decades of life for age of onset. What I’m showcasing there in the bubbles is a reflection of the number of patients that we’ve identified that have mutations at specific residues or positions along the SOD1 protein, and how those different residues confer a different age of onset for those patients. Specifically, I’d like to draw your attention to the box in the middle there, under the residue at the 85 position. This is a reflection of the l85f or v variant in SOD1, which is among the more prevalent variants in SOD1, which seems to have an earlier age of onset across the many patients in our dataset with enough clinical information for us to determine that pattern. This is in contrast to, say, patients who have mutations at residue 114.

We talked about age of onset and potential genomic associations; let’s talk about the rate of progression of disease. This is the overall view for those patients with sufficient longitudinal information depicting the rate of progression for their disease. That’s a smaller number than the 2,006 that I alluded to earlier, again, because it requires more longitudinal understanding. That’s still a substantial number of patients, here numbering 587. What you see here is the distribution of speeds of progression for those patients, broken down in “fast,” “moderate,” and “slow,” demarcated by years three and ten of initial onset to demise. SOD1, for whom we have 232 patients, has a relatively similar spread to the overall distribution of rate of progression for patients at large with familial ALS. Similarly, the rate of progression is exacerbated in patients who have mutations in FUS, which you can see here as a preponderance of those patients with FUS, out of the 87, have a very rapid rate of progression of disease. So FUS is not just fast in its onset, but as I showcased earlier, it presents in younger and younger patients, which collectively suggests that variants in FUS confer a more aggressive pattern of disease.

So in contrast to the fast and early presentation of ALS for patients with mutations in FUS, here’s the distribution for our patients who have mutations in the VCP gene that are associated with neurological disease including ALS. What you can see here is that there is a more indolent course, a slower rate of progression across those 46 patients who have ALS, or who may have inclusion body myositis, which is commonly used in the differential of ALS, and might suggest that there would be benefit in more precise definition of those patients, based on genotype of genes associated with ALS to rule in or rule out different diseases in the differential.

Just like I showcased for age of onset, for specific protein positions in SOD1, this is a depiction of that distribution of unique variants at different residues along the linear axis of the protein on the bottom there, as compared to the rate of progression of disease for each of those patients. Again, the size of the bubble indicates the number of patients that were identified in our dataset with those particular mutations. I’ll draw your attention to that point in the upper left, that five, that’s actually a reflection of the a4v variant that we alluded to earlier, that’s an idiosyncrasy of the SOD1 variant nomenclature, but they are the same variant. You see that trend proved out in our data as well, in addition to having been suggested by that single study that was cited from a couple years ago. That lends credence to the reliability of the other data that I’m presenting here, most notably, that although SOD1 is overall associated with a sort of uniform distribution of rate of progression just like all familial ALS patients have, in contrast, patients with a mutation at the 91st position in SOD1 have a slower course, with a significant number of patients demonstrating that trend. Without going into any biochemical details, this may suggest that different variants in the SOD1 protein confer different functional effects.

Speaking of functional effects, this is now a different gene the TARDBP gene, where for each of the patients where there were variants in the TARDBP gene, I’m showcasing here on the bottom the number of variants which have a description of their functional consequence based on empirical studies of in vivo or in vitro functional studies for TARDBP. I’ve just listed the top ten, though there are a number of different functional consequences that were annotated here to showcase the deep and sensitive annotation comprising the ALS genomic landscape, even among empirical studies that may or may not be associated with a clinical case, to better understand the disease mechanisms that play here across all genes that are associated with ALS disease causation.

Lastly, in terms of description of the data and the results of the ALS genomic landscape, I want to talk a little bit to the prevalence of gene mutations among familial ALS patients as reflected here. I think this is a Nature Reviews review about ALS, but you can see here for each gene on the left, you can see on the right their relative prevalence or the proportion of cases for familial and sporadic ALS where mutations in each of those genes have been found. For various reasons that I don’t have time to go into, we’re relegating C9orf72 to a later talk, but when you look at SOD1, TARDBP, FUS, TBK1 and VCP, genes that I’ve talked about in other slides, you can see the relative proportion of patients based on some small-scale sequencing studies put together to to try to guesstimate what the prevalence is. That allowed me to think a little bit more about the ability of our data in the ALS genomic landscape to be applied in the service of better predicting what the prevalence, the relative prevalence of gene mutations in these different genes is across all familial ALS patients.

The reason that that came to my mind is, we’ve seen that there’s a correlation between the reference citations for each of these variants and the prevalence of patients with those variants across disease cohorts. So this is a different disease circumstance for a different gene, in this case, it’s cancer and P53 mutations, but what you can see here in the plot of article count for each of those variants, reflected by dots, against patient count in a representative and very large cohort of disease patients, you can see on that log scale a nice direct correlation between citation count and patient prevalence.

Now turning our attention back to the ALS data, across each of these different genes, we have a certain number of patients out of those 2,006 who had mutations in any one of those specific genes. When you look at that data in aggregate, again, relegating the C9orf72 patients for a separate analysis, you can see here that the prevalence of mutations in familial ALS patients that were projected from that review hold true, based on an assessment of the representation for each of those cases with variants and specific genes in the empirical evidence in the medical literature, such as has been aggregated and annotated in the genomic landscape. There’s much more detailed information available about each of these cases, but to take a step back and look at the value of this information that we’ve aggregated, I like to say that the patients that are represented in the body of scientific literature and clinical studies is actually a reflection of the types of information that one would find in a large-scale sequencing study of new patients. When I do some back of the envelope calculations, the number of patients with mutations in each one of these genes is sufficient in our dataset to reflect what would only be available in a sequencing study of three to ten thousand familial ALS patients. That’s, as I’ve suggested throughout, a reflection of the unique value that can only be extracted when you look at this data at scale in a very comprehensive and exhaustive way.

Just to summarize here by walking through what we have learned: there’s a great deal of information available in the empirical literature that we’ve aggregated, annotated and analyzed here for ALS across multiple different genes, and we’ve found a number of putative genomic associations that have bearing on clinical presentation, and quite likely and especially with emerging genomic therapies, on how we’re going to better treat these ALS patients. The implications of this data, the possible implications that I articulate here are that there may very well be, given the clinical heterogeneity of ALS and the diagnostic challenges attendant with that heterogeneity, there may very well be distinct syndromes or sub-syndromes that could be better predicted with a more comprehensive understanding of genotypes for all of these ALS patients, either to speciate different patients with ALS defined by those clinical parameters, or otherwise to better understand which patients have ALS or some other disease in the differential diagnosis. Having this understanding from this data allows one to better understand all of the idiosyncratic disease mechanisms across all of these different genes, and how those disease mechanisms at the biochemical level may be influencing the clinical features that we alluded to in those trends that I highlighted.

The possible applications of this data in the ALS genomic landscape are in resolving that diagnostic uncertainty that I spoke to earlier, which would be a sea change in the way that neurology patients in general and ALS patients specifically are currently diagnosed, as well as, as was suggested before about the change from a one-size-fits-all clinical trial model to a more personalized medicine approach, using the ALS dataset that I’ve discussed here will allow clinical trial designers to better segregate patients a priori by genotype to ensure a more likely successful outcome of those trials meeting the endpoints, especially as the therapies are predicated on specific disease pathologies or specific gene mutations in those patients. Finally, forward looking, when all of this trial data has been accumulated, I foresee that having a better understanding of the genomic underpinnings of ALS disease causation will allow us to dictate much more precisely which patients should get which treatments so that we can ensure improved outcomes for all patients, regardless of their unique genomic makeup.

In conclusion here, I’d like to just state that Genomenon produces these landscapes across all different diseases, and specifically what I’d like to highlight here is that we’ve done such an exhaustive analysis, as I’ve showcased here for ALS, for other neurodegenerative diseases, including Parkinson’s disease with I think more than four dozen, five dozen genes associated with Parkinson’s disease and Parkinsonism or Parkinson’s disease risk, all exhaustively annotated in the ways that I described for ALS. We’re making great progress on Alzheimer’s disease as well, a number of genes associated or presumed to be associated with Alzheimer’s are being analyzed in this way, as I described for ALS, and so too for Wilson’s disease. Altogether, the promulgation of this data across all of neurodegenerative disease I think will reveal unprecedented insight into the disease mechanisms of these different disease entities, as well as the unique clinical presentations that may be speciated by these genomic findings and how that has bearing on how we test and treat patients in clinical trials and moving on into clinical practice in the future.

With that, I’d like to thank you for your attention, and encourage anyone of you who’s interested in learning more to reach out to me via the Genomenon website.