(differing with respect to their relationship to obesity, fat distribution, and lipid metabolism), and a sixth cluster (designated only in the Mahajan et al. paper) had less clearcut phenotypic features [20,38] (Table 2). The T2D-risk variants assigned to the three insulin sensitivity clusters displayed the most obvious overlap across the two approaches. Variants near FTO, MC4R, and NRXN3, all loci known to have substantial impact on variation in BMI, mapped to a cluster of T2D-risk variants thereby assumed to be driven primarily by obesity. Variants at IRS1, PPARG, and KLF14 implicated in effects on adipocyte differentiation and body fat distribution, were colocated to a cluster of T2D-risk variants featuring lipodystrophy-like effects on insulin sensitivity, partly overlapping with the set of “favorable adiposity” loci identified by others [34-36]. Finally, variants at GCKR and TM6SF2, known for their profound impact on ectopic fat accumulation in liver and altered circulating lipid levels [86,87] were members of a cluster which seems to be driven by alterations in hepatic metabolism. Though there was broad agreement concerning the variants deemed to influence beta-cell function, disposition across the pair of beta-cell clusters was less consistent, particularly for variants with less dramatic effects on the continuous glycemic traits that distinguished them. T2D-risk variants at SLC30A8, TCF7L2, ADCY5, HNF1A, and MTNR1B consistently mapped to a cluster characterized by an association between T2D-risk, reduced insulin levels but elevated proinsulin levels, whilst those at ARAP1, IGFBP2, DGKB, and CCND2, combined T2D-risk and reduced beta-cell function with reduced proinsulin levels. Some of the variation in the assignment of other variants across these two clusters reflects differences in the traits included in the respective analyses, compounded by substantial differences in the size of the GWAS data sets available across traits (which has an impact on discriminatory power). Nevertheless, the replicated subdivision of beta-cell function variants into two clusters distinguished by the direction of the association to proinsulin speaks to two distinctive mechanisms whereby T2D-associated variation results in beta-cell dysfunction [88]. Despite some of the differences in the assignment of individual variants across clusters, the mechanistic basis of these clusters appears robust, mapping as it does to current understanding concerning the major pathophysiological processes influencing T2D development. Allocation of variants to these physiologically-defined clusters is also broadly supported by orthogonal analyses of tissue-specific patterns of chromatin accessibility, histone modification, and transcriptional regulation. The various subsets of T2D-risk variants identified by clustering of GWAS data demonstrate clear evidence of genome-wide enrichment with respect to tissue-specific active enhancers and promoters [9,38,44,89-91], cis-eQTL signals [90,92], and enhanced connectivity in tissue-specific protein-protein interaction networks [93]. As anticipated, these link variants in the insulin secretion clusters to altered transcriptional regulation in the islet, and those in insulin action clusters to events in liver, fat and muscle. Beyond the ability of these efforts to identify disease pathways, a critical question in terms of clinical translation is whether or not the pPS generated from these clusters show associations with clinically relevant outcomes: early results are encouraging. For example, differential cluster associations have been observed for coronary artery disease, stroke, and the renal complications of diabetes [38,94,95], each emphasizing enhanced risk associated with T2D predisposition mediated through insulin resistance. In the case of macrovascular disease, of course, this is likely to reflect the pleiotropic impact of these variants on nonglycemic risk factors such as lipids. A specific role for pPS-captured defects in insulin secretion and altered gut microbiome has also been reported: those microbiome changes include an effect on butyrate-producing pathways shown to play a causal role with respect to diabetic and obesity phenotypes [8]. These findings support the notion that whilst, by definition, all cluster-defined pPS associate with T2D risk, differential effects can be detected with respect to aspects of mechanism, phenotype, and clinical outcomes. However, further effort is needed to validate and extend these findings, and to define the contribution that these can make to the delivery of more personalized management in diabetes. So far, clustering analyses have been restricted to a subset of the most robust genome-wide significant T2D-associated variants, primarily those discovered in Europeans, and for which association statistics are available across multiple related traits. More complete analyses (delving deeper into the list of T2Dassociated variants, and embracing a wider range of traits) capable of generating more powerful pPS will become possible as GWAS efforts for those other traits scale up. Inclusion of additional phenotypes should provide more granular clustering, attributing mechanism to ADVANCE ARTICLE: Endocrine Reviews Downloaded from https://academic.oup.com/edrv/advance-article-abstract/doi/10.1210/er.2019-00088/5535575 by 81225740 user on 24 July 2019 ADVANCE ARTICLE Endocrine Reviews; Copyright 2019 DOI: