So, you have a target and want to start a drug discovery program, do ya? How would you do it?
When I was at Brigham and Women’s Hospital, Harvard Medical School and the Broad Institute, I presented an idea from an early GWAS of rheumatoid arthritis (RA, see here) to Ed Scolnick (former president of Merck Research Labs, now founding director of the Stanely Center at the Broad Institute, see here). In this study, we found evidence that a non-coding variant at the CD40 gene locus increased risk of RA. The first questions he asked: How does the genetic mutation alter CD40 function? Is it gain-of-function or loss-of-function? What assay would you use for a high-throughput small molecule screen to recapitulate the genetic finding?
I was caught off-guard. Sadly, I had never really thought about all of the details. At the time, I knew enough as a clinician, biologist and a geneticist to appreciate that CD40 was an attractive drug target for RA. However, I was quite naïve to the steps required to take a target into a drug screen. That simple conversation led to several years worth of work, which ultimately led to a proof-of-concept phenotypic screen published in PLoS Genetics five years later (see here).
Since joining Merck, I have learned even more about the steps to advance a hit from an initial screen through to the clinic. Even if you have an assay for an initial screen, you need a hit finding package to follow-up on initial findings. In particular, you need a suite of assays to demonstrate that your therapeutic molecule engages the target and does what you think it is suppose to do.
Which brings me to the two key points addressed in this blog:
Human genetics not only takes you to a target, but also provides insight into mechanism of action (1) to develop high-throughput screening (HTS) assays for an initial screen and (2) to develop target engagement assays to ensure a hit compound recapitulates the genetic findings.
Consider the example of PCSK9 (see here). Early genetic studies showed that gain-of-function mutations increased LDL cholesterol levels and increased risk of heart disease (see here) and loss-of-function mutations lowered both (see here). While these studies indicated that PCSK9 inhibition should be beneficial, there was still the open question of HOW to inhibit function? How did PCSK9 influence LDL cholesterol levels? Should the mechanism of inhibition be through the catalytic activity of PCSK9, soluble levels of PCSK9, or another MOA?
It took years of research with animal models (circulating PCSK9 lowers LDL receptor levels), structural chemistry (the binding site on PCSK9 for the LDL receptor is distant from its catalytic site, see here), and cellular biology (PCSK9 functions as a chaperone to prevent LDLR recycling and/or to target LDLR for lysosomal degradation, see here) to generate testable therapeutic hypotheses about PCSK9. In these and other studies, the genetic mutations were useful biological tools to sort out relevance of different aspects of PCSK9 biology – and they served as anchoring points to focus assay development.
[ A nice exchange on the biology of PCSK9 and the challenge for drug discovery comes from Derek Lowe and David Grainger (see here). They go into interesting details about whether a small molecule inhibitor might be a good strategy to inhibit PCSK9. It is a good read for those interested in this topic. ]
Another example about the utility of genetics for assay development comes from cystic fibrosis (CF). The CF transmembrane conductance regulator (CFTR) gene, which encodes for a cAMP-regulated anion channel expressed on the surface of secretory epithelial cells, was identified in 1989 (see history here, here, here). The first mutation discovered, delta-F508, impairs CFTR folding and stability. In homozygous carriers, the CFTR protein fails to make it to the cell surface. If there is 0% CFTR function, then disease develops; if there is 50% function, there is no evidence of disease, as heterozygous carriers are healthy. For other mutations (e.g., G551D, observed in ~5% of CF patients), the CFTR protein makes it to the cell surface but functions less effectively. Considering the nearly 2000 CFTR mutations together, ~15-25% CFTR function relative to wild type is required to reduce disease severity or potentially even avoid disease.
These molecular observations led to a series of therapeutic hypotheses – CFTR “correctors” or “potentiators” – about treating the underlying cause of CF (see here). Corrector molecules would help mutant CFTR protein get to the cell surface, whereas potentiator molecules would help mutant CFTR proteins on the cell surface function more effectively. New phenotype assays based on the genetics of CFTR chloride channel function were developed to enable high-throughput small molecule screens. It took over 20-years, but in 2012 a first-generation drug to treat patients with a particular mutation (G551D) was finally approved. Matthew Herper of Forbes heralded the achievement as “a triumph of genetics and drug development, the first medicine to directly affect the genetic defect that causes the disease”.
What I also like about the CF story is it serves as a humbling reminder about the inconvenience of human genetics in drug discovery. Often, the underlying molecular defects may not point to a target or a mechanism that is easy to drug. In the case of CFTR drug discovery, new assays were developed to correct cellular misprocessing (“correctors”) and to find activators of CFTR function (“potentiators”; see here).
But what is key in both examples is that human genetics serves as an anchoring point to understand which assays are required to conduct an initial screen and to ensure target engagement to correct the underlying molecular defect in patients with disease.
Finally, consider two recent genetic findings in type 2 diabetes (T2D), which highlight the two key points of this blog (genetics to guide assays for an initial screen and assays for target engagement). The first, published in Nature (see here), identified SLC16A11 as a novel candidate gene for T2D with a possible role in triacylglycerol metabolism. However, the direction of effect of the disease-associated variant (GOF vs LOF) is not clear, nor is it understood how the disease-associated variant impacts function of SLC16A11 protein. Without this information, it is very difficult to develop robust assays for an initial drug screen.
The second T2D example, published in Nature Genetics (see here), demonstrates that SLC30A8 haploinsufficiency protects against T2D. This finding suggests that assays to measure function of ZnT8 (the protein product of SLC30A8) and screens to find ZnT8 inhibitors of may be an effective therapeutic strategy in T2D prevention. However, too little is known about ZnT8 to know how to measure target engagement in pre-clinical models. How much ZnT8 inhibition is needed to achieve therapeutic benefit (genetics would indicate at least 50% inhibition)? Is it sufficient to measure glucose levels in animals to know that the target has been engaged appropriately, or are there more specific molecular assays to measure target engagement?
Before concluding, I want to return to the theme of this blog series: decision-making in drug discovery. If you have target with multiple functions – or even worse, a target of unknown function – how do you know what assays should be used to initiate a drug discovery program? For example, if a protein has multiple functions (the PCSK9 protein has multiple domains with different functions, both intracellularly and extracellularly), which one is most relevant to human disease? Without clear causal human biology that links perturbation of the protein to an outcome relevant to human disease, it is difficult to know what is the right way to perturb the target.
In contrast, human genetics effectively tells you, “do this”. But even with human genetics as an anchoring point, the process is still not easy. The hope, as exemplified by PCSK9 and CFTR, is that a series of alleles at a gene locus can focus the effort to help define the right suite of assays to move a program forward.