Drug Discovery: Empiricism vs Insight
How do you go about discovering new medicines?
You might begin empirically, and look around for something that seems to work — willow bark for pain, for instance, or an anti-tuberculosis pill for depression. This approach relies on what author Morton Meyers calls “happy accidents,” and accounts for a remarkable number of existing therapeutics (most of them, Nassim Taleb asserts). Supporting this seemingly radical conjecture, a 2011 paper in Nature Reviews Drug Discovery reported that between 1999 and 2008, more FDA-approved first-in-class small molecules were derived from phenotypic screening (an empiric approach) than from insight-oriented, target-based drug discovery.
In this age of Big Data, there’s considerable interest in trying to industrialize serendipity in silico, and use large datasets and smart algorithms to generate empirically-derived novel insights and new therapies. While this nut hasn’t yet been cracked, many clever data scientists are busily working on this challenge.
The second way to approach drug discovery is mechanistically: figure out the cause of a disease, and use this knowledge to shape treatment. This represents the promise of molecular medicine, though realizing this dream has proved much more difficult than many anticipated, due to the sheer complexity of biology, and the challenge of moving between DNA and disease. However, when this approach works, the success can be so compelling, so logical, so validating that it motivates researchers to continue trying.
The elegant triumph of the cancer drug Gleevec, for example, inspired cancer researchers to seek other mechanistically-informed treatments — although for years, Gleevec remained one of molecular medicine’s few uncontested success stories.
Captivated by PCSK9
More recently, the protagonist role in the molecular medicine narrative has been assumed by the family of lipid-lowering drugs now in development that target PCSK9. Amgen and Regeneron/Sanofi lead this race, with many others not too far behind.
What makes PCSK9 so captivating is that the discovery and development process actually worked the way it’s supposed to, but in practice rarely does. The study of two French families with unusually high cholesterol levels and a strong history of heart disease led to the identification of a factor called PCSK9, which was subsequently shown to increase cholesterol when overactive. Additional research – nicely described by Gina Kolata in The New York Times — demonstrated that people with a reduced activity of PCSK9 had low cholesterol levels and a markedly lower cardiac risk. Extensive animal studies strongly reinforced these findings. Collectively, these data suggested that a medicine that could reduce PCSK9 activity or function would represent a potent therapeutic. The results of clinical trials performed to date seem to support this optimism.
The PCSK9 experience – particularly its elegance and relative ease — seems to have transfixed every drug developer who’s touched it, at once reaffirming their belief in the promise of molecular medicine and stimulating their drive to re-create this apparent success.
The key starting point – not surprisingly – is phenotype, ideally, an extreme phenotype of obvious medical interest. As the noted clinical researcher Stephen O’Rahilly has eloquently observed, “The study of rare individuals with extreme disturbances of their physiology has long had an impact in terms of scientiﬁc understanding grossly disproportionate to the infrequency of the particular disease being investigated.”
Studying these patients, O’Rahilly writes, can help both these specific patients while potentially providing insight into broader physiological questions.
Archibald Garrod’s classic study of a patient with black urine in 1901 led to the “inborn error in metabolism” concept, anticipating by 40 years the “one gene, one enzyme” concept, as Brown and Goldstein point out in their Nobel address.
Brown and Goldstein’s own transformative work on cholesterol transport was similarly rooted in a small population of patients – homozygous familial hypercholesterolemia [FH]– with an extreme phenotype — extraordinarily high lipid levels. Their discoveries contributed to the development of the statins, medicines now taken by millions of people with elevated cholesterol to lower lipids and reduce cardiac risk. (Sadly, the patients with homozygous FH not only have the highest levels of lipids, but these lipid levels are only minimally impacted by statins; however, there’s hope these patients could derive additional lipid-lowering benefit from the new drugs targeting PCSK9.)
New Economics of Drug Development
There’s an important economic driver at work here as well. Historically, drug developers were interested in targeting the largest population they could find – big market, big impact, lots of revenue. Over the last decade or so, this has changed, primarily as a result of two contrasting forces. First, increased regulatory and reimbursement pressure has made it harder and more expensive to get approval for a drug offering modest benefit to a large population; the studies need to be huge, and even after approval, payors are reluctant to pay for effects perceived as small.
In contrast, the economics of targeting small, well-defined populations seem increasingly favorable. The ability to precisely target drugs to the patients means that you treat only the patients most likely to benefit, resulting in proportionally larger impact, and often relatively small and more manageable trials. Such niche products often command premium pricing, in some cases nearly half a million dollars a year (per patient), raising the question of whether the current approach is sustainable (and what will happen to patients with rare diseases if it’s not).
Chasing Success – Three Approaches
Three companies – including two with leading PCSK9 programs – have set out to replicate the success of the PCSK9 story, each in a slightly different ways. In each case, the key issue is the starting point – where do you find the phenotypes of interest?
Amgen Looks To Iceland
The approach taken by Amgen, the Thousand Oaks, CA-based biotech with a current market cap of around $93B, is arguably the most direct: buy it. Their 2012 acquisition of deCODE Genetics reflects their belief that there’s tremendous value in the dataset that Kari Stefansson has built by combining DNA sequencing information and phenotypic information on much of the population of Iceland.
Amgen’s acquisition surprised many observers, who had left deCODE for dead. Originally founded in 1996, deCODE rode enthusiasm for the platform and an ebullient market to a successful IPO in 2000, netting a solid return for early investors such as Polaris. However, unable to turn promising science into commercial success, and further challenged by global economic conditions, deCODE struggled, ultimately declaring bankruptcy in 2009. Its core assets were then auctioned off – and picked up for $14M by a pair of VCs including Polaris. Their efforts to reinvigorate the company evidently paid off – deCODE was acquired by Amgen in December 2012 for $415M in cash.
According to Polaris’s Terry McGuire,
“In the face of tremendous rational and irrational societal pressures related to medical privacy, Kari was able to build a coalition in support of transforming genomics research. He was able to tie together large amounts of Iceland’s conserved genealogy, healthcare records, and over 140,000 blood samples volunteered by Iceland’s countrymen.”
These efforts, McGuire writes, have resulted in the publication by deCODE researchers of “over 400 major studies in peer reviewed journals.”
The question now is whether Amgen will be able to succeed where deCODE (and others with whom they’ve previously partnered) seem to have failed: turning these data and insights into drugs and revenue, while responsibly addressing the issues around use of data that continue to dog the company.
Regeneron Looks to Geisinger
Last week, Regeneron – a rapidly climbing, science-driven biotech based in Tarrytown, NY, with a market cap now at nearly $30B – announced a slightly different approach to the same question, revealing a new partnership withGeisinger Health System, one of the country’s most innovative healthcare organizations. In this new relationship (well-characterized by Andrew Pollack in The New York Times), Regeneron will sequence the DNA (technically, just the coding region, or exome) of 100,000 volunteer patients from the Geisinger, at an estimated cost of about $100M over five years, according to the Times.
Geisinger will get a lot of free sequencing out of this deal, as well as some stake in any Regeneron products that eventually come out of this initiative.
Regeneron, meanwhile, will gain access to phenotype (as well as the DNA sequences the company is bankrolling). Geisinger has a well-established electronic medical records system, which provides valuable phenotype information on each patient (though of course the data will be de-identified before it’s shared with Regeneron). While the data is not as dense as many would like (generally capturing just information from episodic visits, rather than continuous measurements, as discussed here), it’s a great starting point.
In particular, according to the Times, Regeneron scientists are likely to be searching for low frequency, high impact variants, rare genetic variations associated with extreme phenotypes, and thus may provide immediate mechanistic insight into specific clinical conditions.
UCB Looks To The Crowd
Perhaps the most interesting approach to the identification of extreme phenotypes was announced last month by Belgium-based pharma UCB, a company with a current market cap of about $13B. Their idea: partner with the open innovation facilitator Innocentive to crowdsource the search.
The goal of this project, according to Duncan McHale, VP of Global Exploratory Development at UCB, is “to identify individuals, families, groups or communities who possess rare phenotypes – for example people who possess great self-healing abilities, incredible memory or who are protected from disease.” The best submission (as determined by an expert panel convened by UCB) will receive a $10,000 award.
“We have a completely open mind about what phenotypes might be most appropriate to explore,” McHale says.
“For example, someone who exhibited exceptional wound healing after surgery or trauma might really help our understanding of wound healing. Equally, a person or group of people who have consistently displayed exceptional resistance or immunity to infections, or someone who, after a robust clinical diagnosis, displayed unusually fast or spontaneous disease remission might lead to a new therapy for that disease. These submissions could all be winners of the competition.”
Ultimately, the idea is to use these phenotypes to identify interesting genes, and from there, develop promising medicines; McHale cites the study of South African patients with unusually strong bones leading to the identification of sclerostin as a promising drug target for osteoporosis drug, and the study of a Pakistani family insensitive to pain driving the identification and prioritization of SCN9A, encoding the Nav1.7 sodium channel, as a target for novel analgesics.
The extreme phenotype to gene to medicine approach is compelling – when it works, which is far less often than you might think. Many important phenotypes may not be driven by a single gene, and even when you can identify the specific molecular cause of a disease, a personalized treatment doesn’t inevitably follow.
Consider the challenges of coming up with a cure for sickle cell anemia – considered the first molecular disease, and a condition for which early initiation of penicillin is now regarded as “the single most important disease-modifying therapy,” according to the Chief Medical Officer of the Sickle Cell Disease Association of America.
Cystic Fibrosis stands out as another prominent example of the challenges of translating genetic information – the most common causative mutation, deltaF508, was first identified by Francis Collins and colleagues in 1988 – into effective drugs, although the recent progress reported by Vertex appears encouraging.
Even many orphan disease companies that are focused on enzyme replacement strategies for diseases caused by a single defective gene are discovering that these mechanism-based treatments are often less effective than might have been predicted based on seemingly precise nature of the intervention.
Final Thoughts: Primacy of Phenotype
Perhaps the most enduring lesson to be drawn from these three efforts is the demonstrated importance of phenotype, and specifically, of a carefully phenotyped population. As the cost of genotyping continues to decline, it will become increasingly apparent that a key sticking point in medical discovery – a key opportunity, as legendary MGH clinical investor Bill Crowley would undoubtedly frame it – is the need for meticulously phenotyped populations.
Traditionally, this has been the provenance of expert clinical investigators such as Crowley, the inquisitive physicians described by Judah Folkman, the patient-oriented scientists revered by Brown and Goldstein.
With the rise of the technologies of digital health, there are now important opportunities for citizen scientists to play increasingly central roles in medical innovation – as well as a new opportunity (really, the obligation) for providers to partner with patients and together collect the dense phenotypic data that, in conjunction with genomic data, biological insight – and yes, muscular, empirically-driven analytics — will power the medical innovation of the future.