One of the biggest challenges of drug discovery is to determine which targets, when perturbed, will have an acceptable efficacy-safety therapeutic window in patients. In fact, the success rate at choosing the right target and developing a safe and effective drug is quite low: less than 10% of drugs that enter Phase I are approved by regulatory authorities (see recent Nature Reviews Drug Discovery article here, Derek Lowe blog here). Most of the failures in Phase II and III are due to lack of efficacy or unexpected toxicity.
Human genetics offers one potential solution to identify new drug targets with an acceptable therapeutic window. A study published this week in Science Translational Medicine (STM) provides genetics support for an established therapeutic target in type 2 diabetes (T2D), glucagon-like peptide-1 receptor, GLP1R (link to STM article here). What is surprising, however, is that human genetics suggests that GLP1R agonists may also protect from coronary heart disease (CHD).
[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.]
There are three points that I want to make in this blog. First, the STM study provides general support for the model that human genetics is useful to predict efficacy & safety in drug discovery. Second, it is possible to infer the functional consequences of an associated allele (e.g., gain-of-function [GoF], loss-of-function [LoF]) without actual functional “wet lab” data. And third, I pose a hypothetical question about GLP1R as a drug target: if we did not know about human pharmacology of GLP1R agonists and/or human biology of GLP1 as an incretin, how would we interpret the human genetic findings from this study for a drug discovery program?
For context, GLP1R agonists are an established therapy for T2D. The target was discovered by a non-genetic “experiment of nature”, based on the observation that oral glucose provokes a greater insulin secretory response than the same amount of insulin injected into a vein (here). Recently, it has been suggested that GLP1R agonists may not only lower glucose in patients with T2D, but may also protect from CHD (see here).
The STM study by Robert Scott and colleagues provides genetic support for GLP1R as a drug target. A protein-coding variant in GLP1R influences fasting glucose (P=2.6×10-10) and risk of T2D (P=9.4×10-5), consistent with a previous study (here). More specifically, the minor allele of the GLP1R Ala316Thr variant is associated with lower fasting glucose and protection from T2D.
But a good drug target is not only effective but also safe. Human genetics can be used to assess on-target safety signals. Note that human genetics cannot predict off-target toxicity, which represents at least half of the clinical toxicities observed with small molecule therapeutics.
Scott et al. test the same GLP1R Ala316Thr variant for association with other phenotypes that may be considered adverse events. What they found, however, was not an adverse event, but rather an association with a clinical phenotype that may be beneficial to patients. The same variant that lowers blood glucose and protects from T2D also protects from coronary heart disease (CHD). This genetic result is consistent with an unpublished observation that GLP1R agonists may protect from CHD (see here). It is important to note the both the effect size and level of statistical significance are modest (OR=0.93, P=9.2 x 10-3). The sample size required to detect this association was large (61,846 individuals with CHD and 163,728 controls), making additional replication difficult.
The second point I want to make is about inferring the functional consequences of an associated allele without actual functional data. Ideally, once a disease-associated allele is identified, functional wet lab experiments would be performed to determine the impact of the allele on gene function. At the simplest level, it is important to determine gain-of-function (GoF) or loss-of-function (LoF) relative to risk of disease, as a drug discovery program needs to determine whether to agonize or antagonize the target or pathway implicated by the genetics study. For example, a LoF mutation that protects from disease would indicate that a target should be antagonized for a beneficial therapeutic effect.
A genetic short-cut to get at function is to identify alleles that are bioinformatically predicted to be complete LoF. Because of natural selection, most of these LoF variants are rare in the general population. Such LoF variants could then be tested for association in humans to establish a relationship between gene function and disease risk. The STM study tried to do this, but could not find enough LoF alleles to perform a sufficiently powered genetic association study.
Although no functional studies were performed, Scott et al. were able to infer function from human genetics and the known pharmacology of GLP1R agonists. It is known that GLP1R agonism with approved drugs enhances glucose-mediated insulin secretion and thereby lowers glucose level. The STM study demonstrated that that the minor allele of the GLP1R Ala316Thr variant lowers fasting glucose and protects from T2D (Figures 3 & 4, respectively). Therefore, it is reasonable to conclude that the minor allele acts like a natural GLP1R agonist and is likely a GoF allele.
As it pertains to cardiovascular outcome, it is also reasonable to make assumptions about direction of effect from a therapeutic perspective. We know that the same allele that protects from T2D also protects from CHD (Figure 4). Thus, it is reasonable to conclude that GLP1R agonism will lower glucose and protect from CHD (if the drug is given for a long enough time).
Finally, if nothing were known about GLP1/GLP1R biology, and the GLP1R Ala316Thr variant were discovered from an unbiased genetic association study (e.g., minor allele is associated with protection from T2D and CHD), how would one go about a drug discovery program? Under this scenario, it would be unclear if GLP1R should be agonized or inhibited for therapeutic benefit. To deduce function, either functional experiments in the lab or genetic association studies with rare LoF alleles (e.g., human knockouts) would be required to assess function relative to disease risk.
Unfortunately, this is the outcome of most human genetic studies today. That is, there is convincing evidence of a statistical association with a trait of interest, but it is unclear with how to proceed with a drug program because of lack of functional data. Making matters worse, even if the biology is understand, it is often difficult to therapeutically modulate the target in a manner consistent with the genetic perturbation (see recent blog here). Yep, drug discovery is hard, even with solid human genetic data such as that presented in the STM study.
Nonetheless, the STM study by Scott et al. provides a very nice example of how human genetics can be used to predict efficacy-safety therapeutic windows and also adds to the growing list of validated drug targets implicated by human genetics.