It has been a good week for human genetics, with high-profile studies published in Science (here) and NEJM (here, here, here), and a summit at the White House on Precision Medicine. Here, I summarize the published studies and put them in context for drug discovery. But first, I want to briefly detour into a story about the Wright Brothers.
[Disclaimer: I am a Merck/MSD employee. The opinions I am expressing are my own and do not necessarily represent the position of my employer.]
In 1900, Wilbur and Orville Wright first began experiments with their flying machine. They defined three problems for manned flight: power, wing structure and control. As described beautifully in David McCullough’s book (review here), the brothers focused on the latter, control, which when sufficiently solved led to the first manned flight in 1903. Within ten years of solving the “flying problem”, aviation technology progressed to the point that manned flights were routine.
By analogy, I would argue that there are three key challenges for drug discovery: targets, biomarkers and clinical proof-of-concept studies. The key problem to solve is target selection. Today, we do not know enough about causal human biology to select targets, and as a consequence we have a crisis in cost (drugs are too expensive to develop because of failures at the most costly stage, late development) and innovation (for those drugs that work, there is insufficient differentiation from standard-of-care treatments to change health care outcomes).
The Science and NEJM papers published this week describe a path towards solving the “target problem”. The model, as I have described previously, is to identify genes with a series of alleles that serve as a biological mimic of therapeutic perturbation. These alleles can be used to estimate the efficacy-safety window of target perturbation at the very start of a drug discovery program. There are a handful of examples today of genes that satisfy the allelic series model, but I expect that many more will emerge as large-scale sequencing initiatives deliver data (see my previous blog on deCODE as one example).
The Science study, “Health and population effects of rare gene knockouts in adult humans with related parents” by Narasimhan and colleagues, performed exome sequencing on 3,222 British Pakistani-heritage adults with high parental relatedness. They discovered 1,111 rare-variant homozygous likely loss of function (rhLOF) genotypes predicted to disrupt (knockout) 781 genes. While this study was not designed to provide deeper insight into drug R&D, it does provide support for the idea. As stated in the manuscript: “Furthermore rhLOF genes that are drug targets have 11.4% phase I to approval rate versus 6.7% for all target-indication pairs (chi-squared P=0.046).”
The real value of the study, however, is that it quantitates the expected frequency of human knockouts in the population. Figure 1B provides an estimate for how many rhLOF variants are expected in an individual, depending upon the parental relationship for that individual. If the parents are completely unrelated, approximately 1% of individuals will carry rhLOF variants. In contrast, if the parents are first-cousins, then approximately 40% of individuals will carry rhLOF variants. Extrapolating to the entire genome, the authors predict “that in 100,000 subjects with first- cousin related parents of the same genetic ancestry we would expect at least one knockout in ~9,000 of the ~20,000 human protein-coding genes”.
How can human KO’s be used in drug discovery? As these populations are sequenced, and as rhLOF variants are tested for association with clinical phenotypes, new insights will emerge that will estimate efficacy-safety profiles at the start of a drug R&D program. If there is no phenotype in humans with rhLOF mutations, this should raise concern that the target is indeed a good target. Alternatively, if there is a clinical phenotype, and if this phenotype is related to phenotype that is a good surrogate for drug efficacy (e.g., triglyceride lowering and protection from cardiovascular disease), then it provides confidence in the drug target. Similarly, if human KO’s are associated with phenotypes that are concerning for adverse drug events, then this should raise concern about mechanism-based toxicities of therapeutic perturbation. To date, such datasets don’t readily exist, and this study provides a path towards identifying human KOs for many genes in the genome.
Two studies published in the NEJM provide evidence for a causal role of plasma triglycerides in coronary heart disease. In the first NEJM study, Stitziel and colleagues performed exome sequencing in more than 72,000 patients with coronary artery disease and 120,000 controls. They confirmed previous genetic associations at PCSK9 and LPA, and identified new associations between coronary artery disease and low-frequency missense variants in SVEP1 and ANGPTL4. Furthermore, they searched for mutations in the lipoprotein lipase gene, LPL (which is inhibited by ANPTL4), and identified both gain- and loss-of-function variants – an allelic series – associated with a decrease and increased risk of coronary artery disease, respectively. Figure 2 serves as a beautiful summary of their findings.
In the second NEJM study, Dewey and colleagues at Regeneron and Geisinger Health System sequenced the exons of ANGPTL4 in 42,930 participants, including 10,552 participants with coronary artery disease (CAD). They identified a series of alleles (low frequency E40K variant and 13 other monoallelic inactivating mutations) that were associated with both triglyceride levels and CAD. The direction of effect was consistent with the Stitziel et al study: variants associated with lower levels of triglycerides were also associated with lower risk of coronary artery disease. Thus, the therapeutic hypothesis is that ANGPTL4 inhibition should lower triglyceride levels (the biomarker) and protect from CAD (the clinical outcome).
What is very cool about the Dewey et al study is that they directly test this therapeutic hypothesis in mice and monkeys. They developed a monoclonal antibody that blocked the inhibition of LPL by ANGPTL4 in monkeys and mice with a human copy of ANGPTL4. They observed reductions in triglycerides, consistent with the human genetic data. But they also found evidence for lymphadenopathy, raising the concern that therapeutic modulation of ANGPTL4 might lead to mechanism-based adverse drug events (ADEs) in humans.
To directly test this ADE hypothesis, they performed a phenome-wide association study, or PheWAS. Specifically, they reviewed de-identified data from electronic health records for participants carrying E40K and other ANGLPTL4 inactivating mutations for information on abdominal or other lymphatic disease. While the study has limitations of power, they did not find any statistically significant associations, thus providing some level of comfort that the effects in the pre-clinical models will not translate into humans. Importantly, however, the PheWAS approach will not address off-target toxicities.
I truly believe we are learning to fly. Examples such as those published this week are starting to solve the “target problem”. Moreover, they are starting to translate into bona fide drug R&D programs. As with manned flight, once these problems are solved other advances will quickly follow. What will the next decade deliver? Buckle up, ladies and gentleman, as we will find out soon.