The appeal of precision medicine is the promise that we can understand disease with greater specificity and fashion treatments that are more individualized and more effective.
A core tenet (or “central dogma,” as I wrote in 2015) of precision medicine is the idea that large disease categories – like type 2 diabetes – actually consist of multiple discernable subtypes, each with its own distinct characteristics and genetic drivers. As genetic and phenotypic research advances, the argument goes, diseases like “type 2 diabetes” will go the way of quaint descriptive diagnoses like “dropsy” (edema) and be replaced by more precisely defined subgroups, each ideally associated with a distinct therapy developed for that population.
In 2015, this represented an intuitively appealing idea in search of robust supporting data (at least outside oncology).
In 2017, this represents an intuitively appealing idea in search of robust supporting data (at least outside oncology).
The gap between theory and data has troubled many researchers, and earlier this year, a pair of cardiologists from the Massachusetts General Hospital (MGH) and the Broad Institute, Sek Kathiresan and Amit V. Khera, wrote an important – and I’d suggest underappreciated – commentary in the journal Circulation that examined this very disconnect, through the lens of coronary artery disease (CAD).
(Disclosures: I’m Chief Medical Officer of DNAnexus, a cloud genomics company, and my training at MGH overlapped with Kathiresan’s.)
Do CAD patients “actually have one of many disease subtypes,” each with its own “distinct driving pathway” and ideal therapeutic approach, they asked, or is CAD a “quantitative blend of causal risk pathways” – essentially, a gimish, with slight defects in multiple pathways collectively contributing to disease manifestation? (See Figure 1.) Kathiresan and Khera tell me they’ve often described these as the “fruit salad” and “smoothie” models of disease, respectively.
Reviewing the CAD literature, Kathiresan and Khera conclude that while “precision medicine will in fact identify a small subset of individuals in whom an identifiable driving pathways accounts for much of their risk of CAD,” for “the vast majority of CAD patients, it is a quantitative blend of causal processes that underlies disease” – i.e. a smoothie.
In other words, for the vast majority of CAD patients, the disease reflects a collection of contributing genetic (as well as, of course, environmental) factors, a mix of slight defects that collectively nudge the body towards disease in what Kathiresan and Khera call “a probabilistic fashion.”
The implications of this, Kathiresan and Khera argue, is that there probably will not be a “taxonomic revolution in complex disease” – a clarifying subdivision of a complex disease into distinct subtypes – although there will be an opportunity to identify the “causal-risk pathways that contribute, albeit to varying degrees,” to complex diseases like CAD.
The authors point out that both rare and common genetic variants associated with CAD seem to involve the same molecular pathways, and the “vast majority of cardiovascular therapeutics in use today have demonstrated benefits across subgroups.” Perhaps the most visible example of this are the statins, which initially were demonstrated to improve the LDL (“bad cholesterol”) levels in patients heterozygous for familiar hypercholesterolemia (an example of a mutation in one gene profoundly impacting CAD risk); yet statins also proved to reduce CAD risk in the vast majority of patients whose disease was not caused by a single powerful mutation — sophisticated subtyping of disease was not required.
If Kathiresan and Khera are correct, and if their hypothesis extends beyond CAD, it means we should think carefully about reflexively assuming diligent data collection will permit us to subdivide complex disease – like multiple sclerosis and Alzheimer’s – into biologically and therapeutically distinct subcategories. At the same time, their argument suggests we should aggressively pursue robust molecular clues that have been identified, as pharmacologically targeting the products of genes with seemingly small effect could still have major impact, as the statin experience demonstrates. Individual rare variants with clear phenotypic effect might be particularly helpful in pointing out high value targets. Furthermore, based on Kathiresan and Khera’s hypothesis, drugs developed based on these targets might be expected to impact most patients, not just a select few. In short, while comprehensive genetic and phenotypic evaluation might be required to identify patients with high impact rare variants, once identified, these variants can lead to the development of a new medicine that has generalized utility. In other words: precision research, blockbuster development.
How far beyond CAD does Kathiresan and Khera’s hypothesis extend? For many conditions, after all, existing drugs seem to work for only a minority of patients; it was this very dilemma that motivated much of precision medicine. Surely, for many psychiatric conditions, some kind of improved “digital phenotyping” must be helpful, as Mindstrong Health President and Co-founder (and former NIMH Director) Tom Insel recently argued. And where do immune-related conditions like rheumatoid arthritis, inflammatory bowel disease, and multiple sclerosis fit in – are there biologically distinct, therapeutically relevant subtypes here?
For rare genetic diseases, precise, molecular-defined treatments will likely remain the goal, although here, too, we might be surprised. At this year’s JP Morgan meeting, for example, Vertex CEO Jeff Leiden expressed his hope that most patients with cystic fibrosis – a disease which can be caused by any of a staggering number of different mutations in a single large gene – could eventually be treated with a single (combination) drug regimen, a view reinforced by promising data that arrived this summer, as FORBES colleague Matthew Herper reported. This is similar to what’s occurred for hepatitis C, with the advent of single (combination) regimens that are effective across all viral genotypes, obviating the need for obtaining such molecular assessment in most patients.
At first blush, oncology looks like the most obvious exception to this hypothesis, and seems like a discipline where precise subtyping and identification of unique drivers has truly revolutionized the field, leading to relatively specific treatments that target specific “driver” genes and pathways. Yet, Kathiresan and Khera suggest that immune-oncology research may prove “potentially even more influential.” Immunotherapy, they argue, was born from “recognition of a key causal pathway,” and is “largely agnostic to the specific driver mutation of an individual cancer cell,” and thus may “have utility across a broad range of cancer subtypes.”
Until now, precision medicine has been animated by the belief that detailed genetic and phenotypic analysis will dissect complex diseases into distinct and more tractable subtypes. Kathiresan and Khera’s recent commentary asks whether we should question this assumption, and suggest that most complex conditions are likely to be caused by a collection of core causal factors, which blend in any given individual to cumulatively cause the disease. The authors endorse the detailed molecular identification of these factors, and anticipate that targeting some of these factors could result in generally effective treatments for many suffering from the condition – though how to pick which of the many potential factors to target seems less obvious; perhaps loss-of-function variants with a clear phenotype could represent a useful starting point.
If Kathiresan and Khera are correct, precision medicine may not produce customized cures for each patient – but instead offer the hope that in elucidating the complex genetic architecture of disease, we will be able to identify and develop novel therapeutics each offering benefit to large numbers of suffering patients.
It’s an interesting theory – but so was the one it hopes to replace. As always in science, we’ll enjoy the conversation, but look for the data.