Replicating genotype–phenotype associations

NCI-NHGRI Working Group on Replication in Association Studies The study of human genetics has recently undergone a dramatic transition with the com­ pletion of both the sequencing of the human genome and the mapping of human haplo­ types of the most common form of genetic variation, the single nucleotide polymorphism (SNP). In concert with this rapid expansion of detailed genomic information, cost-effective genotyping technologies have been developed that can assay hundreds of thousands of SNPs simultaneously. Together, these advances have allowed a systematic, even ‘agnostic’, approach to genome-wide interrogation, thereby relaxing the requirement for strong prior hypotheses. So far, comprehensive reviews of the pub­ lished literature, most of which reports work based on the candidate-gene approach, have demonstrated a plethora of questionable geno­ type–phenotype associations, replication of which has often failed in independent stud­ ies. As the transition to genome-wide asso­ ciation studies occurs, the challenge will be to studies because of issues in either the initial conclusion from the literature because followseparate true associations from the blizzard of study or the attempted replication. Small up studies have not consistently analysed the false positives attained through attempts to rep­ sample size is a frequent problem and can result same markers or those in perfect linkage dis­ licate positive findings in subsequent studies. in insufficient power to detect minor contri­ equilibrium (r = 1.0). Other recent examples The purpose of a replication study is to evalu­ butions of one or more alleles. Similarly, small for which initial reports of association have ate a positive finding from a previous study, sample sizes can provide imprecise or incor­ been inconsistently replicated include insu­ to provide credibility that the initial finding is rect estimates of the magnitude of the observed lin-induced gene 2 (INSIG2) and obesity, valid. Replication is essential for establishing effects. Poor study design — particularly a lack and cyclic-AMP-specific phosphodiesterase the credibility of a genotype–phenotype asso­ of comparability between cases and controls (PDE4D) and stroke. These have been ciation, whether derived from candidate-gene — can increase the risk of biases because there accompanied by controversies about what or genome-wide association studies. However, can be heterogeneity in exposure to environ­ actually constitutes replication. there is a lack of agreement about what consti­ mental challenges and population stratifica­ This paper presents the conclusions of a tutes a finding deserving of replication, what tion. The latter arises when investigators fail working group on the replication of geno­ constitutes an adequate replication study and to account for case–control differences in the type–phenotype associations — whether what constitutes a replication or refutation. genetic structure of the underlying population. identified in genome-wide or candidate-gene Investigators and journal editors have offered Heterogeneity in classification of outcomes studies — convened by the National Cancer guidelines for how to address this problem, across studies can undermine the opportu­ Institute and the National Human Genome but these initial efforts have been hampered by nity to compare among them. Similarly, data Research Institute. The group was composed limited experience and conflicting empirical ‘dredging’ can be a major problem, especially of experts from diverse disciplines, including data. However, as evidence has accumulated, when criteria for defining phenotypes are biostatistics, clinical medicine, epidemiology, several instructive examples have emerged altered to achieve statistical significance worthy genetics and scientific publishing. The purpose of genotype–phenotype associations being of publication. was to review the current state of the field and reproduced reliably in follow-up studies. These Another challenge arises when follow-up propose best practices for the design, conduct include peroxisome proliferator-activated studies analyse different variants. An example and publication of replication studies that aim receptor-γ (PPARG) and the transcription is the reported association between DTNBP1 to follow up notable findings, particularly in factor TCF7L2 (refs 14–19), related to diabetes; and schizophrenia, initially identified in Irish genome-wide association studies. The group nucleotide-binding oligomerization domain pedigrees and ‘confirmed’ in independent addressed three topics. First, assessment of the containing 2 (NOD2) and Crohn’s disease; European studies. Unfortunately, different validity and limitations of any single genetic complement factor H (CFH) and age-related risk alleles and haplotypes were reported in association study. Second, criteria for establish­ macular degeneration; and chromosome each study, making comparison difficult. ing replication in genetic association studies. region 8q24 and prostate cancer risk. Although it is plausible that more than one Third, points to consider for publication of Many instances have arisen in which initial variant could contribute to schizophrenia risk high-quality genotype–phenotype association findings have not been reproduced in follow-up at the DTNBP1 locus, it is difficult to draw this reports (Box 1).