Personalized Medicine: Temper Expectations

See allHide authors and affiliations

Science  24 Aug 2012:
Vol. 337, Issue 6097, pp. 910
DOI: 10.1126/science.337.6097.910-a

The 1 June Policy Forums, “The ultimate genetic test” (R. Drmanac, p. 1110) and “Whole-genome sequencing: The new standard of care?” (L. R. Brunham and M. R. Hayden, p. 1112), discuss clinical breakthroughs that might be possible through whole-genome sequencing (WGS). We offer a cautionary note about the interpretation and expectations of personalized medicine and its subset, individualized drug therapy, specifically those that pertain to risk prediction in the individual patient.

As first shown in 1918 (1), a complex quantitative trait can be explained by Mendelian inheritance if multiple genes affect the trait. From this analysis, one can infer that accurate statistical predictions of a complex trait require identification of many small-effect variants which, in combination, can explain a large fraction of variance in the phenotype. For most complex traits, this is an unachievable goal. Although we can obtain WGS data from a large number of patients, effect sizes for the majority of small-effect variants are simply too miniscule to be detected, even with any practicably attainable sample size. The anticipation of personalized medicine and individualized drug therapy thus seems unrealistic. We might be able to obtain accurate genomic data from an individual patient, but our ability to tailor treatment will be limited to only a small fraction of variants that have relatively large (“identifiable”) effect sizes.


Before 1990, a number of examples of pharmacogenetic traits, usually binary, were published [e.g., (26); reviewed in (7)]. Most of them adhere to simple Mendelian inheritance and are controlled by one or a very small number of large-effect genes. These breakthroughs in genotype-phenotype associations helped to establish expectations of individualized genetic risk prediction.

Numerous complexities of the genome—and what constitutes a gene—have emerged during the past two decades. We now realize that most examples of pharmacogenomic traits (adverse drug reactions, as well as drug efficacy) resemble complex diseases and other multifactorial traits such as height or body mass index. These traits reflect contributions from innumerable low-effect genes. Consequently, genome-wide technology and large-cohort studies (on the scale of N = 20,000 subjects) might enable us to uncover many genetic variants significantly associated with a complex trait. However, these variants are just the tip of the iceberg. Even in combination, they usually explain only a small fraction of phenotype variation and, consequently, have limited predictive value or clinical utility.

This same dilemma applies not only to all DNA-sequence analyses (genome-wide association studies, copy-number variant patterns, and WGS), but also to transcriptomics, metabolomics, epigenomics, and other analyses that evaluate multifactorial traits. Hence, all of these methodologies might lead to individual discoveries of significant disease associations when large cohorts are studied. These findings may well lead to identification of novel drug targets and, hence, new inroads into greater understanding and possible treatment of complex diseases. However, the idealistic goal of personalized medicine and individualized drug therapy, which needs a holistic understanding of each individual patient's unique -omics read-out, is most likely unattainable—for the vast majority of complex traits—by advances in technology alone.


Stay Connected to Science

Navigate This Article