Plenge Lab
Date posted: April 30, 2017 | Author: | No Comments »

Categories: Drug Discovery Human Genetics Immunogenomics

A recent study in the New England Journal of Medicine provides genetic support for a pharmacologically validated target, BAFF, in the treatment of systemic lupus erythematosus.  But can human genetics also be used to estimate the target dose and a therapeutic window?

As readers of plengegen.com know, I am constantly on the lookout for published studies that provide insight into the utility of human genetics for drug discovery and development.  This past week there was a great post from Francis Collins on the role of the NIH in the discovery (in part via human genetics) and development of tofacitinib (see here), anakinra and potentially novel targets (e.g., STING) for inflammatory diseases (here).  Nature Reviews Drug Discovery published a News & Analysis on PCSK9 as a “fertile testing ground for new drug modalities including long-acting RNA interference drugs, vaccines against self-antigens, CRISPR therapeutics and small molecules that control ribosomal activity” (here).  New York City released information about a new public health initiative, The NYC Macroscope, which will use electronic health records (EHRs) to track conditions managed by primary care practices that are important to public health..and one day may be linked to genetic data for discovery research (that is me just speculating).

In this blog, I want to focus on a very interesting study published by Steri et al in the New England Journal of Medicine (NEJM) on the discovery and functional characterization of an autoimmune-disease associated genetic variant in BAFF.  What makes this study interesting is that BAFF is the target of an FDA-approved therapy, belimumab (or Benlysta), thereby providing yet another “retrospective example” of human genetics for drug discovery.  What makes it even more interesting is the depth of functional data presented by Steri and colleagues, and the ability to compare the functional genetic data to functional data generated in clinical trials of belimumab.  These data can be used to address the question: can human genetics be used to estimate target dose and therapeutic window of a BAFF-inhibitor? Before tackling this question, I want to provide a brief summary of the NEJM study.

[Disclaimer: I am an employee of Merck, although will soon transition to a new role at Celgene.]

The study by Steri et al in the NEJM (here) provides genetic support that BAFF is involved in the pathogenesis of two autoimmune diseases, multiple sclerosis (MS) and systemic lupus erythematosus (SLE). The study identifies an insertion-deletion variant in the 3’UTR of the gene, TNFSF13B, which encodes for the protein B-cell activating factor, or BAFF (also known as BLyS).  The variant is associated with risk of multiple sclerosis (MS; increases risk by ~25% per copy of the risk allele) and systemic lupus erythematosus (SLE, increases risk ~40% per copy of the risk allele). The MS- and SLE-risk variant introduces a novel polyA signal, which shortens the 3’UTR of the transcribed TNFSF13B mRNA.  The portion that is removed contains a microRNA binding site. As a consequence, the shorter “risk” mRNA is more stable, is expressed at higher levels, and leads to nearly 2-fold higher levels of soluble BAFF in peripheral blood from healthy volunteers (effect size = 0.77 standard deviation units; P = 8 x 10-150; see Figure 1C, and Table S9 in the Supplementary Appendix).  The gain-of-function BAFF variant (referred to as BAFF-var) increases levels of circulating B lymphocytes and soluble immunoglobulin levels. As summarized in the manuscript: “an insertion–deletion variant, GCTGT→A (in which A is the risk allele), yielded a shorter transcript that escaped microRNA inhibition and increased production of soluble BAFF, which in turn up-regulated humoral immunity.”  These data provide a mechanistic link between BAFF and autoimmunity, which builds on observational data in patients, mechanistic studies in the murine models, and pharmacological data in humans.  In addition, the study references murine data and human epidemiology to suggest that the same BAFF-var that increases risk of MS and SLE might (emphasize might!) also decrease risk of malaria.  The study presents no genetic data to support this hypothesis, however.

Based on the genetic and functional data presented by Steri and colleagues, a therapeutic hypothesis is that inhibiting soluble BAFF should be beneficial in the treatment of both SLE and MS, but would put an individual at risk for malaria.   But of course, this therapeutic hypothesis has already been tested, at least in patients with SLE: a BAFF-inhibitor, belimumab (or Benlysta), was approved by the FDA in 2011.  Belimumab is a human antibody that binds to BAFF, thereby preventing BAFF from binding to B lymphocyte cells.  Belimumab, which is reported as only “marginally effective”, is most commonly used in SLE patients with active musculoskeletal or cutaneous disease that are unresponsive to standard therapy with glucocorticoids or other immunosuppressive agents.  It has not yet been tested in more severe SLE, including patients with renal or CNS involvement.

For the remainder of the blog, I want to focus on a question that is often raised once human genetics implicates a specific target: can genetics also be used to estimate optimal drug dose and a therapeutic window?  Human genetics provides evidence that perturbing a target will be effective in humans (see here).  However, it is unusual that human genetics – at least for complex traits – also provides evidence for the amount of pharmacological manipulation that will be safe and effective in humans.  That is, it is rare that human genetics provides clear evidence of a target dose or a therapeutic window.

I have argued previously that having a series of alleles provides a dynamic range of genetic perturbations – complete loss-of-function (or human knockouts), partial loss-of-function and gain-of function – which in turn helps estimate genotype-phenotype correlations that mimic pharmacological dose-response curves (here, here).  But what about when a genetic study identities only a single allele, as in the example of BAFF-var?  Let me walk through the logic using BAFF-var and belimumab as a case study, acknowledging (yet again!) that this is purely a retrospective example; human genetics had nothing to do with nominating BAFF as a target nor guiding dose of belimumab in clinical trials.  

I start by asking: what level of BAFF modulation is effective in treating SLE patients?  If this were known, it would be relatively simple to establish a connection between BAFF modulation with belimumab and the BAFF-var human genetic findings.  Unfortunately, belimumab interferes with accurate measurements of soluble BAFF (see here).  Thus, it is not possible to directly measure a dose-dependent relationship between BAFF levels and clinical effectiveness in belimumab-treated SLE patients.

An alternative is to use pharmacodynamic (PD) biomarkers to measure the downstream consequences of BAFF-inhibition.  A similar approach has been very successful in correlating the relationship between variation in genes influencing LDL cholesterol and risk of cardiovascular disease (e.g., NPC1L1 [target of ezetimibe] and HMGCR [target of statins]; see here).   Several PD biomarker measurements are robust in belimumuab clinical trials, as the drug does not interfere with these biomarker assays.

Rather than go through all of the PD biomarkers in detail, I focus on two  – naive B cell levels and soluble IgM levels – as both are also measured in the NEJM genetic study (see Supplementary appendix, Table S8 and S9).  This enables a comparison between the magnitude of BAFF-inhibition via a drug (belimumab vs placebo) and human genetics (wild type vs BAFF-var).

As shown below, the effect of belimumab is about 3-fold greater than the effect of the BAFF-var (although see my note below). Such comparison biological data offers a chance to benchmark the magnitude of effect for a genetic vs pharmacological perturbation.  To facilitate the comparison, I equate belimumab and the wild type BAFF allele, as both are associated with lower levels of soluble BAFF; and I equate placebo and the BAFF-var allele, as both are associated with higher levels of soluble BAFF.  The results are:

  • belimumab vs placebo: 60% decrease in naive B cells
  • wild-type vs BAFF-var: 20% decrease in naive B cells
  • belimumab vs placebo: 30% decrease in IgM levels
  • wild-type vs BAFF-var: 10% decrease in IgM levels

The effective size for the BAFF-var is actually quite large, but is still approximately 3-fold less than the level of therapeutic modulation achieved with belimumab.  Nonetheless, the functional data derived with these PD biomarkers is the first step towards establishing function-phenotype dose response curves.

It is worth pointing out that the magnitude of biological effect is relatively small for most risk alleles associated with most complex traits identified via genome-wide common-variant association studies (or CVAS).  As a result, disease-associated alleles identified via CVAS-based human genetics do not often provide enough dynamic range to understand the optimal therapeutic modulation that might be safe and effective.

[Note: It is possible that I did not adequately translate the B cell subsets across studies, as the flow cytometry panels used to in the belimumuab clinical trials and BAFF-var genetic study were different.  I have also made the BAFF-var the “reference” variant, so that a more direct comparison can be made between belimumab (which inhibits BAFF) and the BAFF-variant (which increases BAFF).  I believe, however, that the numbers are close enough for illustrative purposes.]

The next step is to understand how the BAFF PD biomarkers relate to clinical symptoms in lupus patients.  That is, how do changes in PD biomarkers translate into the effect observed on SLE patients, and can these biomarker data be used to gain insight into how human genetics may guide therapeutic dose?  Unfortunately, the clinical comparison is indirect, as the belimumab clinical trials measure disease activity, while the human genetic effect measures risk of disease.  In the BLISS-52 study, a Phase 3 registration trial, the response rate, as measured by the SLE Responder Index (or SRI), was 58% for 10 mg/kg belimumab compared with 44% in the controls. In another Phase 3 trial, BLISS-76, belimumab at 10 mg/kg led to a SRI response rate of 43.3% compared with 33.5% in controls.  In the Steri et al NEJM study, the odds ratio (OR) was 1.44 for risk of SLE in BAFF-var vs wild-type was across all cohorts.  At the risk of making statisticians squeamish, these numbers roughly translate into:

  • belimumab vs placebo: 30% improvement in disease activity at 1-year
  • BAFF-var vs wild-type: 30% decrease in lifetime risk of SLE

Honestly, I don’t know quite what to make of this clinical comparison, as the underlying biological processes for SLE disease activity and lifetime SLE risk of disease are likely quite different.  At a minimum, it suggests that the phenotype of “risk of SLE” is relevant for drug discovery and development (see deeper discussion on this topic here).  However, I don’t know if it is possible to extract much more than that.  If there were a clear relationship between the PD biomarkers and SLE disease activity, it might be possible to make this correlation (as has been done with LDL and risk of cardiovascular disease).  But alas, this is not known.

One additional take-away might be related to the therapeutic benefit of belimumab in MS vs SLE.  The effect of the BAFF-variant on risk of MS (OR=1.25) is less than that of SLE (OR=1.44).  This would suggest that the clinical benefit of belimumab on MS would be less than that observed in SLE.

My conclusions:

  1. Human genetics supports BAFF as a drug target, even though human genetics was not relevant in the discovery & development of belimumab, an inhibitor of BAFF approved for the treatment of mild-to-moderate lupus.
  2. Human genetics indicates that BAFF-inhibition should be beneficial in treating both SLE and MS, although the magnitude of benefit on disease activity is likely to be less in MS based on a crude interpretation of lifetime risk of both diseases.
  3. Human genetics support changes in B lymphocytes (as measured by flow cytometry) and soluble immunologlobulins (such as IgM) as reliable pharmacodynamic (PD) biomarkers in early development clinical trials.
  4. In this example, human genetics does not provide much guidance as to target dose or therapeutic window for a BAFF-inhibitor, as the range of effect of the risk allele is limited relative to the observed clinical effect of belimumab.

On this last point, here is where human knockouts and an allelic series (here, here) could be very beneficial for drug discovery and development.  If human knockouts in TNFSF13B/BAFF could be identified, and clinical data linked to genetic data, it would give an estimate of the magnitude of maximal genetic perturbation (and by extension maximal pharmacological perturbation) on clinical phenotype.  Similarly, if additional partial gain- or loss-of-function alleles could be identified in TNFSF13B/BAFF, then a more accurate estimate of genotype-phenotype correlation could be made.  Together, additional alleles could help estimate target dose and a therapeutic window.

I want to conclude by asking the community for help: we need better statistical methodologies to extrapolate human genetic data to estimate optimal therapeutic dose. As describe here, this is not easy.  However, I am sure the community can come up with some clever solutions to provide better estimates than we do today.

Any takers?

 

Leave a Reply