The availability of large genetic bioresources creates the possibility of pursuing new approaches to drug discovery and development. Speakers in the second session were asked to describe selected current discovery activities enabled by new genetic cohort studies and to explore opportunities for cross-sector engagement.
The use of bioresources enables many novel approaches to research, including sophisticated analyses, new approaches to target validation and biomarker development, and studies of allelic variation in drug response. More broadly, large cohort studies are enabling a greater understanding of the biology of disease and of drug–target interactions, which facilitates the discovery of drugs to treat a wide array of diseases. This chapter also introduces ideas for incorporating patient perspectives and data in genomics-enabled drug development collaborations.
The Pharmacogenomics Knowledge Base (PharmGKB)1 developed at Stanford University is a comprehensive resource in which information about the impact of genetic variation on drug response has been curated for researchers and clinicians. The most popular feature of PharmGKB is the collection of pathways that it details, each of which has been curated from the literature and is supported by PubMed references, said Russ Altman, the Kenneth Fong Professor and a professor of bioengineering, genetics, medicine, and (by courtesy) computer science at Stanford University. The molecular pathways detailed in PharmGKB make possible a systems approach to drug discovery and understanding pharmacogenomics. PharmGKB provides a “platform to do new research projects on how drugs work, how they cause side effects, how they interact, and how we can discover new drugs,” Altman said. Informatics approaches allow researchers to integrate data on genes and drugs across many different scales, including the molecular, cellular, physiological, and population levels, he continued.
One example of the molecular information that PharmGKB provides is details about promiscuous drug binding, which can result in adverse side effects. Nonspecific drug binding can result from the interaction of a small molecule with proteins that have similar molecular structures to the
target protein, a phenomenon that can be detected through three-dimensional modeling (Liu and Altman, 2011). It is easier to understand side effects when the potential interactions of a small molecule with a wide array of proteins—and not just with the intended targets—are examined, Altman said. Another useful informatics approach featured on PharmGKB involves the text processing of millions of research abstracts to enable a high-fidelity extraction of information about the relationship between genes and drugs. As an example, Altman explained that variations in the ABCB1 gene can affect a patient’s response to verapamil, a calcium channel blocker; PharmGKB’s text processing feature distinguishes, identifies, and describes these variations in response.
Another approach that PharmGKB researchers have taken is to process electronic health record (EHR) data, thereby detecting harmful and often unanticipated effects of drugs and drug combinations at the population level. For example, a statistical model was used to recognize two drugs that adversely altered blood glucose levels based on the adverse event signature reported through the U.S. Food and Drug Administration’s Adverse Event Reporting System, a database that contains information on adverse events and medication errors. The statistical model revealed that pravastatin, a statin, and paroxetine, a selective serotonin reuptake inhibitor used to treat depression, can significantly increase blood glucose when taken together, even though neither is particularly associated with hyperglycemia (Tatonetti et al., 2011).
Together, these and other informatics approaches make it possible to test hypotheses in silico before scientists carry out experiments and prioritize them according to which appear most promising, Altman said. Discovering and developing effective and safe drugs requires an extensive amount of research in areas such as target structure and dynamics; drug recognition and binding; cellular response and molecular pathways; gene, drug, and phenotype associations; clinical responses; and population effect reporting, he said. This is in contrast to the approach that led to the discovery of PCSK9 as a target for lowering low-density lipoprotein (LDL) levels, which involved defining a relevant phenotype (low LDL levels), finding phenotypic outliers, identifying causative variants, showing that a loss of function has positive effects, and developing and refining a targeted inhibitor. Altman noted that in comparison to the PCSK9 example, drug discovery often involves a more comprehensive approach that includes defining relevant cell types and tissues; defining relevant molecular pathways; identifying genes and variants associated with pathway function and dysfunction; finding loss-of-function pathway
opportunities while seeking genetic evidence for modulation; conducting counter-screens for promiscuity, side effects, or efficacy issues; and developing and refining inhibitors of pathways.
In summary, Altman pointed out specific needs that could accelerate genomics-driven drug discovery (see Box 3-1). He also offered four primary observations about using informatics to advance drug discovery efforts. The first is that the majority of genetic contributors to disease consist of loss-of-function variants. Creating drugs to treat these diseases requires figuring out how to add back function, which is challenging. In contrast, when a loss-of-function mutation offers protection from a particular disease, it is relatively straightforward to develop a therapeutic strategy, he said.
Secondly, data integration across multiple scales—including molecular, cellular, physiological, medical (as represented in EHRs), and population levels—is essential for minimizing the risk of the drug discovery process. One reason why informatics is powerful, Altman said, is that it is not sensitive to the scale of the data, because “we can easily move between scales.” For example, he said, informaticists can study and curate data across molecular, cellular, systematic, and population levels.
The third point Altman emphasized is that drugs, like genes, have vastly underestimated pleiotropic effects that need to be considered during drug development. Drugs can cause unintended adverse effects when they bind to “off-target” proteins (Karczewski et al., 2012).
Finally, considering biological pathways instead of single targets will be helpful in understanding how pleiotropic genes and drugs modulate biology and for predicting side effects and poor efficacy. Altman said that
he and his informatics colleagues think in terms of networks rather than targets. “Looking just at the target won’t be the answer in 90 percent of the cases, but looking at its pathway may offer opportunities,” he said.
Drug discovery programs need to be set up so that targets in cell and model systems can be modulated and assayed in ways that recapitulate relevant parts of human physiology, said Sally John, the vice president of computational biology and genomics at Biogen. It is the translational piece, along with a better understanding of the biology of the targets, that is going to accelerate genomics-based drug discovery, she said.
John illustrated this point by describing work that she and her colleagues have done on salt-inducible kinase (SIK) inhibitors for the treatment of immune-mediated disease. Previous research demonstrated that inhibition of SIKs had the desirable effect of attenuating inflammatory cytokines while at the same time enhancing anti-inflammatory cytokines (Clark et al., 2012). Further research showed that these anti-inflammatory effects were mediated by one specific SIK isoform, SIK2. John and her colleagues went on to look for genetic data linking SIK2 with inflammatory diseases, and through a literature search they discovered that variants in SIK2 are associated with primary sclerosing cholangitis, an autoimmune disease of the liver (Liu et al., 2013). Meanwhile, data from the University of Cambridge Mendelian Randomisation Catalogue2 indicated that the same SIK2 variant had a protective effect for cardiovascular disease, indicating that drugs that inhibit SIK2 function may pose a potential safety concern, she said. Data from 23andMe also confirmed that the SIK2 SNP carried an increased risk of immune phenotypes and a decreased risk of cardiovascular disease, John said. Examining data from multiple genetic bioresources was extremely important, she said, because they supported the idea that the inhibition of SIK2 may be beneficial for immune-mediated disease, but the data also pointed out potential safety concerns. If researchers wish to move forward and design a selective SIK2 inhibitor,
2 The University of Cambridge’s Mendelian Randomisation (MR) Catalogue is a curated database with publicly available results from large-scale genetic association studies. The MR Catalogue has been recently renamed to PhenoScanner, and the website is available at http://www.phenoscanner.medschl.cam.ac.uk (accessed June 17, 2016).
they may need to enlist a team of chemists to try and avoid off-target effects, she said.
To gain a better understanding of functional coding variants, Biogen started collaborating with several other companies and nonprofit organizations as part of the Industry Partnership for Human Genetics, John said. Focusing on about 50 target genes of interest, the partnership carried out a comprehensive assessment of loss-of-function and gain-of-function variants and enrichment in the Finnish registries across the genes. The phenotypic data were queried according to specific hypotheses based on each individual genetic target of interest, and association analyses of distinct alleles were done across the entire phenotypic spectrum.
The results revealed the cardioprotective effects of loss-of-function PCSK9 variants. Individuals with these variants were much less likely to use lipid-modifying agents. However, they were more likely to use drugs prescribed for mental illness and were more likely to be hospitalized for psychiatric disorders. “Is this a potential safety indication?” John asked “I don’t know whether these types of drugs and indications would overlap with the newer cognitive effects we are seeing occasionally on PCSK9 inhibition, but this is the sort of data that we would pay attention to.”
Getting a complete picture of the role of allelic variation on target genes of interest is critical to improving success in early drug discovery, John concluded. In particular, she said, a better understanding of the biology leads to improved hypotheses.
Understanding the biology of drug targets can increase the likelihood of successful drug discovery and development, said Tim Rolph, the vice president of program value enhancement at Pfizer Inc. It takes considerable time to understand molecular targets and disease mechanisms well enough to devise effective treatments, he said. Many view the development of PCSK9 inhibitors as an ideal model for genetics-driven drug discovery; however, the knowledge about cholesterol metabolism contributed by Nobel prize–winning researchers Michael Brown and Joseph Goldstein greatly aided the process, Rolph said.
There are several critical stepping stones that arise early on the path to creating novel medicines, Rolph said. The first step is to design a primary
screen using small molecules or antibodies to detect a phenocopy of a genetic variant. For example, the existing knowledge of PCSK9 biology enabled the design of an in vitro screen to detect monoclonal antibody phenocopying of the PCSK9 loss-of-function variant. “We understood how PCSK9 interacted with the LDL receptor in terms of the extracellular domain,” Rolph said, “and we constructed and immobilized the extracellular domain binding assay.” Although the process seems very simple, it is an important step in the drug discovery and development process, he said.
The next step is to understand exposure–effect relationships by developing secondary screens (ex vivo or in vivo, or both) with a high confidence in the translation of an effect to the clinic. These secondary screens can be on either pathophysiological or physiological mechanisms, Rolph said, as long as they can detect the modulation that is being performed.
The third step is to demonstrate proof of pharmacology during first-in-human trials at a well-tolerated dose. A key aspect of this step, Rolph said, is to be able to translate preclinical pharmacokineticphamacodynamic (PK-PD) modeling to predict human PK-PD in the clinic so as not to rely on a “leap of faith” when trying a medicine in humans.
The fourth and final step is to demonstrate clinical proof of mechanism during short-duration studies in target patients across key safety biomarkers. “Positive proof of mechanism gives us great confidence to go and invest in clinical proof of concept,” Rolph said. If there is not a clear understanding of the phenotype of a particular genetic variant at the molecular and cellular level, the chances of developing an effective drug are low, he concluded.
The translation of 21st-century genetics into diagnostics, prognostics, and treatments has had many successes, especially in the areas of oncology, rare diseases, and drug safety, said Lon Cardon, the senior vice president of alternative discovery and development and head of target sciences at GlaxoSmithKline (GSK). However, in many areas the gap that exists between research results and medical applications is as wide as ever, he noted. “Where is genetics going to have its promise,” he asked, “and what are the first areas of translation where the impact is going to be beyond oncology, rare diseases, and adverse events?”
One way to bridge the gap between discovery and clinical use is through the use of more robust target validation, Cardon said. As noted in Chapter 1, drugs with human genetic information are more than twice as likely to be successful clinically as those without such information (Nelson et al., 2015). Drugs that successfully reach patients are more likely to have genetic validation, Cardon said, and failures at each stage are more likely to occur with drugs without genetic validation. However, only about 10 to 15 percent of targets currently being pursued by pharmaceutical companies have genetic support for a specific indication, Cardon said. If the number of new targets with genetic support increased to 50 percent, there would be a 13 to 15 percent reduction in the overall cost of drug development, he noted. This would result in significant savings, as it was recently estimated that the average cost of developing and marketing a new drug is $2.6 billion dollars (Tufts Center for the Study of Drug Development, 2014).
However, it is important to remember that not all genes can serve as good drug targets, Cardon continued. A gene may provide a good predictor of a disease, but it is often the case that multifaceted research is needed to turn a predictor into a valid drug target. Among the many factors that need to be determined are the mechanism of action, pleiotropy, gene regulation, the gene’s position within a pathway, tissue specificity, and its chemical tractability, Cardon said.
To address the need for research on a wide variety of factors to validate drug targets, the Centre for Therapeutic Target Validation (CTTV)3 was established. CTTV is a public–private collaboration designed to harness the power of big data and genome sequencing information to improve the success rate for discovering new medicines. The three original founding organizations of the CTTV were GSK, the European Bioinformatics Institute, and the Wellcome Trust Sanger Institute. Recently, Biogen joined the effort, and other potential members have expressed interest in joining. The CTTV was founded on the premise that the study of drug targets themselves belongs in the precompetitive realm, Cardon said. Pharmaceutical companies compete in other areas, such as chemistry, clinical trials, and linking targets to relevant phenotypes. But in the precompetitive realm, companies can share research results without losing their competitive advantage, he said. Furthermore, validating targets can sometimes require a combination of skills that are not available in any one organization, public or private. To that end, the CTTV brings
together people with different types of expertise to work collaboratively on projects. The concept of the CTTV is very exciting, Cardon said, because experts in the areas of drug discovery, functional genomics, and EHRs operate next to one another, all asking different questions of the data. Members of the center formally agree to pool their expertise and to share findings openly. The center is also a way to train a new generation of translational scientists, Cardon said. This collaborative concept appears to be taking off, he said, and although GSK provided the initial funding, it is reassuring that other organizations have now joined in as well.
Cardon also acts as an advisor to the Precision Medicine Initiative (PMI),4 which has the goal of enabling a new era of medicine through research, technology, and policies that empower the development of individualized care. He said that the CTTV is focused on the earliest stages of drug discovery, which complements the efforts of initiatives like the PMI. “The PMI will follow from these drug discovery target validation activities just about perfectly,” Cardon said, “and we need to find a way [to make] the PMI data equally accessible to all so that we can benefit all parties.”
The Michael J. Fox Foundation, the world’s largest nonprofit funder of research on Parkinson’s disease, has the mission of accelerating the development of improved therapies and, ultimately, a cure for people living with Parkinson’s disease. The foundation’s research focuses mostly on translational to early-stage clinical research, said Sohini Chowdhury, the senior vice president for research partnerships at the foundation. The foundation supports not just the development of therapeutics, but also the tools that are required to transform a finding into a therapeutic, such as models, assays, reagents, and biomarkers.
As an example of the foundation’s efforts in this area, Chowdhury cited the Parkinson’s Progression Markers Initiative (PPMI)5, which is an attempt to address the critical lack of biomarkers associated with dis-
4 For more information on the Precision Medicine Initiative, see https://www.nih.gov/precision-medicine-initiative-cohort-program (accessed June 17, 2016).
5 For more information on the Parkinson’s Progression Markers Initiative, see https://michaeljfox.org/page.html?parkinsons-progression-markers-initiative-get-involved (accessed October 14, 2016).
ease mechanism, drug mechanism, dosage determination, study eligibility, stratification into disease subtypes, and clinical signals. It is a precompetitive initiative, and the foundation is the primary funder of the study and receives additional support from industry partners, nonprofit organizations, and private individuals. The study population includes idiopathic Parkinson’s subjects, age- and sex-matched controls, individuals who clinically present with Parkinson’s disease but do not show a dopamine deficit, and individuals with and without genetic risk factors. All of the subjects are followed for a minimum of 3 years and a maximum of 8 years and undergo the same array of assessments, including clinical data collection, imaging, and biosampling.
All of the data collected through the PPMI study are accessible. “We share everything,” Chowdhury said. “You can go to the PPMI website and download all of the clinical data, all of the imaging data, all of the data that we collect. You can also request access to the samples. To date, we have had 500,000 data downloads, and 70 specimen requests.”
This is an exciting and optimistic time for Parkinson’s drug discovery and development, Chowdhury said. Several disease-modifying trials are in early-stage clinical testing, and genetic discoveries have provided tractable targets. But genetics can only take researchers so far in drug development, she said. Parkinson’s is a complex disease whose effects go far beyond what is occurring in a patient’s neurological system. As a result, Chowdhury said, all therapeutic development for Parkinson’s disease needs to begin with thorough studies at the epidemiological and clinical levels that take into account the patient experience as part of the biological characterization of the disease. “It has to all come back to the patient and the experience of the patient,” she said. Traditional disease research management holds that drug development unfolds in a linear fashion from basic research to the clinic. However, drug development should not be thought of as linear, Chowdhury said. Instead it should be viewed as a cycle, she said, and it is important to look for biomarkers in addition to drug targets because there is a critical need to measure the progression of the disease in actual patients.
The members of the foundation believe that achieving this perspective requires breaking down the siloes that exist between basic science, drug development, clinical research, and biomarker studies, Chowdury said. These fields should be communicating and working together, she said, and the research should all be grounded in the patient experience and seen through that holistic lens. Such a holistic view also
requires involving patients as partners, she added, elaborating as follows: “How are you getting the patient perspective? How are you getting access to patient samples? You have population-based studies, which are extremely important, but you also need studies that are focused on individuals with the disease that you are going after, because that’s a very unique perspective.”
Chowdhury noted that technology has made it possible for even small groups of patients with the disease to connect. The research landscape is changing, and integrating patients is important, she said. Patients who are living with a disease can be a valuable resource, she added, because they are often more willing to share their data than members of the general population, because they are in a position to benefit directly by doing so.
As discussed in previous sections in this chapter, one issue to be considered in a discussion of genomics-enabled drug discovery using data from large bioresources is the role of precompetitive research. (See Chapter 4 for more information on business models that support collaboration.) In the drug discovery process, there are research areas that certain groups view as precompetitive, but others see as competitive. In an effort to encourage collaboration and increase efficiency it may be useful to explore the early steps in drug discovery research and understand where all parties are willing to work together. Chowdhury advocated for “putting stakes in the ground” around areas that are precompetitive. “We need to start to think about what is competitive versus precompetitive,” she said. For example, both the development of tools and clinical datasets can be precompetitive. Although following individuals for long time periods is expensive, she said, working collaboratively on tools, reagents, and assays, while perhaps legally complex, is not that cost intensive.
As an example of research that is largely precompetitive, John cited research on cholesterol metabolism. Another example, she said, is developing additional biomarkers. “If we can do that precompetitively and come up with more robust biomarkers that we can use in the clinic, that would accelerate our ability to translate basic genomics,” she said. Genomics has been a field marked by collaboration, she noted, which facilitates precompetitive research.
One way to draw the line between precompetitive and competitive, Cardon suggested, would be to label the identification of targets as precompetitive while labeling experiments on those targets as competitive. However, he added, “where it gets complicated to me is when you start bringing molecules into it. Where do you draw that line, and how do you distinguish that, because molecules are also tools that can interrogate the targets themselves for validation?”
As an example of this distinction, Chowdhury noted that The Michael J. Fox Foundation spearheaded an initiative with three pharmaceutical companies that are all targeting the kinase LRRK2. All three companies shared their tool compounds with the foundation to determine whether a safety profile that was identified in one compound was actually relevant to all compounds that are targeting LRRK2. Precompetitiveness means identifying what is going to benefit the group as a whole, and in this case, all parties benefited from knowing the safety initiative data, Chowdhury said. This is an issue in academia as well as in industry, she continued. It is a field-wide challenge to understand that creative precompetitive arrangements can take place while still safeguarding proprietary interests, she said.
On the other hand, Stefánsson expressed some skepticism about the idea of making the target or research on the target precompetitive. “The target is not just a molecule,” he said. “The target is a molecule in a context. . . . The science is always going to be focused on the discovery and the characterization of the target in the context that you want. So I think it would be devastating if we would make the target precompetitive. That’s where we should be competing. That’s where it’s going to be exciting.” Scientific information becomes public eventually, but between discovery and publication, industry can use that information to compete, Stefánsson emphasized.
Cardon responded, “If you don’t want to share it, you don’t share it. There is no obligation to share anything you work on. [But] we are tripping over each other making the same mistakes, doing the same experiments, reproducing failure, not knowing where to look in the pathways. It’s more efficient from a societal and from a patient perspective to think of that part of the game as something we could share.”
As part of the Industry Partnership for Human Genetics, John said, companies submitted a number of genes of interest to academic researchers. All of the participants saw the full list of genes, but they did not know which company had submitted which gene. Furthermore, participants were not aware of the indications that companies were interested in
pursuing. Biogen was very happy with this type of arrangement, John said, because it received in return all of the phenotype associations with its genes of interest.
Relatedly, there is a distinction between genotype–phenotype correlations that come from patients’ DNA and downstream innovation, which is not necessarily performed by the same people uncovering the correlations, Daly said. “How to set that boundary in a way that maximizes the patient resources being used and available to everyone but retains the motivation that’s needed is worth fleshing out further,” he said.
For example, Cardon said, in the context of the PMI, all parties need access to information in the cohort study database. The initiative is patient-focused, but it also accommodates scientific and commercial interests. The generation of insights that lead to therapeutic hypotheses is going to remain precompetitive to some degree, Rolph observed. However, different models have been successful, such as those used in Iceland, in the U.S. commercial health care system, and with direct-to-consumer genetics companies, he said, and it is worth exploring the precompetitive space in each of these models.
Another area related to the precompetitive space that can present challenges is the relationship between industry and university researchers, said John Carulli, director of precision medicine at Biogen. In terms of precompetitive investment by the pharmaceutical industry, he asked, which topics are the most important to fund? Because of the complexity of drug development, Cardon said, it is no longer an efficient approach for industry to blindly give money to university researchers and hope that something comes back. Both basic science and regulation are getting more sophisticated and complex, he said, and there needs to be novel funding mechanisms and new ways to bring together the expertise of relevant individuals.
University researchers can be fearful of translational research at times, Altman said, but that attitude can change quickly once a research project reaches its full potential. Furthermore, many young researchers have gotten interested in translational research, including those with a background in computer science and engineering. However, a missing ingredient is the expertise of clinician scientists, whose numbers are declining (NIH, 2014). “That is something we have to work on in terms of the educational pipeline,” Altman said.
This page intentionally left blank.