10.1210/er.2019-00088 15 variants which currently show only weak phenotypic features, and bringing to light new pathways involved in T2D development. Integration with tissue- and cell-type-specific regulatory annotation maps will continue to support mechanistic inference [38,44]. Greater access to association data on T2D and other traits from non-European ethnicities will enable broader exploration of ethnic-specific variants and the heterogeneity of clinical presentation and course across major ethnic groups. As confidence grows in the mechanistic basis of these variant clusters, it will become possible to use trait-specific GWAS data to “build out” cognate pPSs and generate more powerful genetic instruments. For example, the pPS formed from the handful of genome-wide significant T2D variants in the “obesity” cluster could be superseded by using a polygenic score constructed from the BMI GWAS efforts themselves, and a pPS capturing islet autoimmunity generated from existing polygenic scores for T1D. For diseases such as T2D, the characterization of clinical phenotype using genetic measures alone is constrained by the fact that individual variation within each of the endophenotypic axes is also influenced by non-genetic factors. Diagnostic and predictive accuracy would be much improved, and the ability to track an individual’s journey from health to disease much enhanced, if the genetic contribution to phenotypic variation (as captured by the pPS) can be integrated with robust longitudinal measures of relevant features of the external environment (e.g. related to diet and physical activity) and internal milieu (e.g. metabolic memory and microbiome). Integration of this “predictive” information with evolving measures of the individual’s clinical state would add another dimension. In the context of T2D, the latter would involve capturing anthropometric data, and glycemic and metabolic state, forming an integrated profile of that individual that can be tracked over time. It would be particularly valuable in this regard to develop process-specific biomarkers that provide clinical readouts for each of the endophenotypic axes that corresponds to a particular pPS. The best illustration of this concept is the use of LDL-cholesterol as an integrated biomarker for that component of cardiovascular risk attributable to genetic and environmental influences on lipoprotein metabolism. The growing availability of large, publicly-available metabolomic and proteomic datasets makes it possible to use pPS as instruments to identify biomarkers correlated to pPS-defined risk as candidates for further prospective testing [96,97]. A key focus of ongoing research relates to understanding how these pPS might be deployed in clinical practice. One interesting possibility is that pPS profiling will allow identification of individuals whose diabetes is mostly attributable to defects in a single process. In the analysis by Udler et al. [38], one third of individuals fell within the top decile of T2D-risk for at least one cluster and, of these, 75% were not placed at the top decile of any other cluster. These individuals would be obvious recruits for the testing of targeted interventions. An alternative, possibly complementary, approach would make use of the full range of scores for a given individual to assign risk, and optimize management. In either case, much will depend on the extent to which these various ways of representing etiological heterogeneity (with or without additional environmental and clinical state information) can be shown to optimize clinical management (for example, the selection of therapeutic agents). One important corollary is that, by conceptualizing a disease such as T2D as arising from the coming together of diverse, largely-orthogonal underlying processes, these models question some of the tacit concepts underlying precision medicine. One of these is the notion that characterization of the specific defect contributing to an individual’s disease invites therapeutic approaches that are designed to specifically correct it. This model has proven effective in monogenic diabetes – where one molecular defect is largely responsible for the phenotype – but it is less clear this can be implemented in polygenic disease. In people in whom the disease is caused by multiple processes, it will be unlikely that modulating a single pathway will be sufficient to correct metabolic derangements; whereas in those in whom the 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: 10.1210/er.2019-00088 16 contributions of specific genetic defects are modest, equivalent reductions in disease risk and progression may be possible through interventions that boost the performance of other processes contributing to overall T2D risk, even those that are already performing at healthy levels. Indeed, because the effects of common variants on the hyperglycemic phenotype are modest, current T2D drugs that target specific pathways (e.g. sulfonylureas and thiazolidinediones) appear to be effective in both carriers and non-carriers of T2D-associated alleles in the respective target-encoding genes [98,99]. Nevertheless, it is possible that some individuals will