How the quality of GWAS of human lifespan and health span can be improved

Abstract

Most connections between phenotype and genetic variants detected in genome wide association studies (GWAS) of human longevity-related traits did not reach the genome-wide level of statistical significance. These estimates also suffer from the lack of replication of research findings in studies of independent populations (Deelen et al., 2011; Nebel et al., 2011). Critical analysis of these studies reveals underused reserves in the data that may improve the accuracy of genetic estimates. One such reserve deals with the proper use of genetic information contained in the age structure of study participants at the time of bio-specimen collection. As we will see later, this information has been ignored in genetic analyses of follow-up data. In this paper, we explain how such information can be used in analyses of follow up data and elucidate the benefits of joint analyses of both types of data. The proposed approach exploits the fact that participants in prospective studies often have different ages at baseline. The bio-specimen collection is often (but not always) done at baseline. Studies included in the CHARGE Consortium [except for the Original cohort of the Framingham Heart Study (FHS) where the bio-specimen collection was done some time after baseline] are typical examples. Since data on lifespan or other durations are often incomplete (e.g., censored), the Cox's type regression models is usually implemented in GWAS of these data (where, in addition to other covariates, conditioning on age at baseline is used). Note that using follow up data alone and conditioning on age at baseline may be required by the goal of the study (e.g., in the search for genes responsible for longest survival after reaching certain age, e.g., after 95 or 100 years). However, conditioning on the ages at bio-specimen collection when some of these ages are high enough may diminish or even totally exclude a substantial part of the genetic variation in longevity, i.e., eliminate the effects we are looking for. This is because the oldest old individuals participating in bio-specimen collection are precisely those who passed the process of mortality selection in the genetically heterogeneous population, and, therefore, are likely to carry genetic variants linked with “longevity” alleles (assuming that such alleles exist). Thus conditioning on the age at genotyping, especially when the oldest old study participants are in the sample, may leave little hope that associations of remaining genetic variants with human longevity will reach genome-wide significance. Such conditioning, however, seems to be a common practice in GWAS of human longevity-related traits dealing with prospective data. For example, a large group of researchers performed comprehensive genetic analyses of human lifespan and free of major diseases lifespan using data from nine studies of the CHARGE Consortium (Walter et al., 2011). The authors “conducted a survival analysis, adjusted for age at baseline and sex, to model continuous time to death or end of follow-up” [Walter et al. (2011), section “Methods”]. The Cox proportional hazards model was used to describe the connections between genetic variants and time to event. Note that, in addition to eliminating useful associations, adjusting for age at baseline may produce a bias in the analyses if the bio-specimen collection has been performed well after the time of the first examination (baseline). Such a situation characterizes the FHS data, which is the part of the CHARGE Consortium. It turns out, however, that in cases when the ages at bio-specimen collection include young adults and the oldest old individuals, the additional information about the role of genetic variants in lifespan can be obtained from the age patterns of genetic frequencies evaluated for any genetic variant even without using the follow up data. The approach based on comparison of genetic frequencies among individuals of different age categories is typically used in genetic studies of centenarians (Weir, 1996; Yashin et al., 1999, 2000; Tan et al., 2004). A monotonic increase in genetic frequency with age indicates that the corresponding variant is associated with lifespan increase (“longevity” allele). A monotonic decline in such frequency indicates that this variant contributes to shortening lifespan (deleterious, or “frailty” allele). In cases when genotyping involves individuals from a large spectrum of ages including young adults, old, and oldest old ages, there is an additional opportunity to improve the quality of genetic analyses which typically is not used in genetic association studies of follow up data. To realize such potential, new methods of association analyses are required.