What’s a Genome Worth?

See allHide authors and affiliations

Science Translational Medicine  09 May 2012:
Vol. 4, Issue 133, pp. 133fs13
DOI: 10.1126/scitranslmed.3004208


A recent study (Roberts et al.) explores considerations in estimating the current and potential clinical utility of whole-genome sequencing for individual patients.

Both in the public eye as well as within the scientific community (1), much of the current enthusiasm for genomics rests on the prospect that assaying a patient’s genetic information will improve the effectiveness of health care—that is, it will directly inform their medical care in ways that improve outcomes and reduce costs. However, it remains controversial whether these high hopes for what is termed “genomic medicine” are well founded. In this issue of Science Translational Medicine, Roberts et al. (2) describe their use of epidemiological data from twin registries to attempt to answer a pressing question in genomics: How much predictive value for disease risk will actually be obtained when the genomes of healthy individuals are routinely sequenced in a clinical setting?

This is a timely topic not only because a week does not pass without the publication of a diagnostic result obtained through whole-genome sequencing but also because the price of whole-genome sequencing has decreased sufficiently that it is now comparable to the costs of other high-tech diagnostic tests such as high-resolution cranial magnetic resonance imaging. The convergence of increasing utility and plummeting costs highlights the relevance of the specific question posed by Roberts et al. (2): What is the maximum potential benefit of whole-genome sequencing with regard to the ability to predict future diseases? We can refer to this predictive task as the quantification of the lifelong risk for a specific disease in an asymptomatic individual without any clinical evidence of disease.

To address this question, Roberts et al. leverage the overall disease prevalence and the rates of concordance between monozygotic twins for 24 common diseases, including coronary heart disease, Alzheimer’s disease, and breast cancer, which they estimate by analyzing clinical data from several large twin registries in Scandinavia and the United States. They introduce the concept of “genometypes”—groups of genomes conferring an identical genetic risk. Because monozygotic twins essentially share a genome, they, by definition, reside within the same genometype. However, only a subset of all possible distributions of genetic risk of disease is compatible with the epidemiological data. For example, the prevalence and monozygotic twin concordance rate of a given disease might equally well be explained by a small fraction of genometypes conferring a high genetic risk or a larger fraction of genometypes conferring a modest genetic risk. Roberts et al. assume a best- case scenario—that is, the distribution of genometypes that is both compatible with the epidemiological data and maximizes the clinical utility of whole-genome sequencing. Assuming that there will come a day when genomics researchers will be able to comprehensively decode the heritability signal present in whole genomes, the result is an upper bound of the performance of whole-genome sequencing for the prognosis of common diseases among asymptomatic individuals.

Given the aspirations of the many teams engaged in this area, most notably the direct-to-consumer genomics companies of the past decade, the results may appear disappointing at first glance. For the majority of the diseases tested (23 out of 24), most individuals would receive negative test results and these negative results would not meaningfully decrease their estimated risk for developing that disease. However, this perspective obscures a more positive result: 90% of tested individuals would in fact have at least one disease for which they were predicted to have a higher risk at a threshold that was selected to be of clinical utility. In other words, although a whole-genome sequence is highly unlikely to serve as a crystal ball across most diseases for most patients, it still has value by identifying subsets of patients that are at a clinically significant increased risk for specific diseases.

Unfortunately, we remain far away from being able to reliably interpret the entirety of the estimated heritability for common diseases encoded by variation in the human genome (3). Nearly all of the millions of genetic polymorphisms observed to date remain of unknown clinical significance. Moreover, many of the variants for which there is published evidence for impact on disease risk may not confer the same risk in an asymptomatic population of interest as they do in the population where they were originally implicated. Consequently, during the interval in which we sort out which variants and combinations thereof contribute to disease risk, not only will whole-genome sequencing fall short of the upper bound estimates for asymptomatic prognosis provided by Roberts et al., but also there may be substantial violation of the medical imperative to do no harm. Specifically, there is the threat of the “incidentalome”—that is, the likely proliferation of false positive incidental findings consequent to performing genetic testing on a genome-wide scale (4).

Notwithstanding the results of Roberts et al., there are an increasing number of published examples of whole-genome sequencing demonstrating its value in specific clinical scenarios. These include whole-genome or exome sequencing that has facilitated unanticipated diagnoses of Mendelian disorders (57), as well as aiding in therapeutic decision-making for cancer patients (8). For example, Worthey et al. reported the lifesaving diagnosis of a rare but treatable Mendelian disorder, X-linked inhibitor of apoptosis deficiency, in a pediatric patient by exome sequencing (5). Meanwhile, Jones et al. have used genome-wide mutational analysis to guide the selection of specific kinase inhibitors to treat a patient with adenocarcinoma of the tongue, a rare cancer with no established treatment protocol (8). Furthermore, the large catalog of non–whole-genome sequencing genetic tests (i.e., targeted to specific variants, genes, or gene panels) that are already a part of medicine is clear evidence that genetic information can and does add value to medical care. The broad set of clinical tasks that might well benefit from whole-genome sequencing that are not addressed by Roberts et al. include genetic testing for all aspects of reproductive health. This includes preconception carrier screening to detect, for example, the Tay Sachs disease mutation, preimplantation genetic diagnosis to select healthy embryos, prenatal diagnostic testing, and newborn screening to definitively diagnose all Mendelian disorders at birth (Fig. 1). In the context of cancer, whole-genome sequencing could be used to provide a rational selection of targeted therapeutics based on the precise molecular pathways disrupted in an individual patient’s tumor and also to monitor the evolution of the tumor in response to those treatments.

Fig. 1. Insights from genomics: From the cradle to the grave.

With the cost of whole-genome sequencing plummeting, the DNA in a blood sample obtained at a single time point (e.g., at birth) could be sequenced and deposited in a database accessible to the individual and designated clinicians. This whole-genome sequencing data could be used prognostically, while the individual is asymptomatic, to calculate the lifetime risk of that person developing common diseases, such as coronary heart disease, Alzheimer’s disease, and breast cancer. An analysis by Roberts et al. (2) suggests that taking this approach might typically reveal a significantly increased risk for the individual for only one out of 24 common diseases. However, it may turn out that the real value of whole-genome sequencing in the clinic lies in areas other than asymptomatic prognosis—for example, the precise diagnosis of a disease at presentation or the selection of an appropriate therapy.

CREDIT: U.S. Air Force photo/Staff Sgt Eric T. Sheler

The differences between these clinical scenarios and that considered by Roberts et al. are manifold. But a key difference is the fact that the probabilistic dependencies invoked while interpreting and acting on genetic information in the context of a specific clinical situation are very different from those when the same task is carried out in an asymptomatic population. These clinical scenarios, rather than prognosis in an asymptomatic population, may well end up becoming the raison d’être of clinical whole-genome sequencing. However, it is important to recognize that the value of whole-genome sequencing in these and other contexts cannot be estimated from the model that Roberts et al. have developed.

Could the analysis of Roberts et al. have significantly underestimated the value of whole-genome sequencing even for the task of asymptomatic prognosis? As the authors themselves recognize, there are some limitations of their study that cannot be fully addressed at this time. First, several of the 24 diseases considered may not be independent—for example, a diagnosis of breast cancer increases one’s risk of developing ovarian cancer. Such dependencies would likely alter the informativeness of genetic profiling. Also, the populations in which many of the twin studies were conducted, such as those in Sweden and Finland, are considerably less heterogeneous than those in other countries, such as the United States. This might affect some of the prevalence and heritability estimates that the current study is based upon. Lastly, a single (and necessarily somewhat arbitrary) definition of clinical utility is applied by Roberts et al. on a per-individual basis. However, it may be the case that whole-genome sequencing is extremely useful (i.e., lifesaving) for some genometypes, such that the “average utility” makes its use for asymptomatic prognosis worthwhile.

How much value must whole-genome sequencing add in order for it to make sense to introduce it into routine clinical care? A back-of-the-envelope estimate suggests that the bar is remarkably low. Let us assume that the all-inclusive cost for ordering a “genome test” for a newborn in the United States drops to $1,000. Given that one’s germline genome is essentially static (and a genome sequenced at birth can effectively be consulted throughout one’s lifetime), this cost can be amortized over the individual’s 78-year life expectancy for an effective cost of just $13 per year. In the United States, we currently spend about $9,000 per capita per year on health care, to which this $13 would add about a tenth of one percent. As the cost of whole-genome sequencing may well drop even lower than $1,000 within the next few years, the “value added” threshold that whole-genome sequencing must achieve to justify its widespread implementation may be even lower.

It is a measure of the rapid maturation of this field that even as we are still sorting out the genetic basis of human disease, we are already able to make informed, if tentative, estimates of the potential clinical utility of whole-genome sequencing as a routine test that will be applied to every citizen. Many more refinements of these estimates will have to be elaborated if we are, as a society, going to make an informed decision as to when and for whom to incorporate whole-genome sequencing into the routine practice of clinical medicine.


View Abstract

Navigate This Article