way is important both for providing a cost-effective strategy to increase diagnosis rates for known forms of monogenic diabetes and for facilitating new gene discovery by reducing study subject heterogeneity. A related application of polygenic scores may be to explain some of the variable presentation of monogenic forms of diabetes, with respect to age of diagnosis for example [69]. The same variant within the HNF1A gene may segregate with early onset diabetes in some pedigrees, but also be observed in individuals who retain normal glucose tolerance into late adulthood and beyond [70]. Studying 410 individuals from 203 HNF1AMODY families, Lango Allen and colleagues found that a 15-SNP T2D rsPS was significantly associated with earlier age of diabetes diagnosis, with each additional risk allele accelerating diagnosis by around four months [71]. 3.4 Clinical application of predictive scores These data provide a sound basis for the use of polygenic scores to support discrimination of major diabetes subtypes and lend credence to their wider clinical value. Given analogous applications of the polygenic score approach for other multifactorial disease traits [13,72,73], these findings have collectively bolstered excitement about their potential to deliver clinical benefit across a wide range of common diseases. One major focus of current research activity lies in exploring the value of polygenic scores to predict individuals at the highest risk of T2D so as to enable early targeting of intervention strategies. If the estimates of relative risk seen in UK Biobank participants in recent studies generalize to the population level (and the current data indicate that performance seems to be sustained throughout the age ranges studied [9,43]), then there are likely to be in excess of one million individuals in the UK, who, on the basis of their polygenic score alone, have a ~50% lifetime risk of T2D [9]. With the price of wholegenome sequencing falling, and the potential to achieve near-perfect imputation by harnessing the combination of large-scale whole-genome sequencing (in a subset of the population) and dense GWAS arrays (in the rest), several countries are starting to plan for a future of universal genetic screening. The rationale is that “one-time” measurement of genome-wide genetic variation (achievable for the cost similar to that of a single outpatient appointment or a chest X-ray), would support a wide range of clinical applications throughout a person’s lifetime, including, but not limited to, the optimization of therapies (based on pharmacogenetic insights) and the prediction of future illness using polygenic scores for a range of diseases. However, there are clearly multiple obstacles to be overcome before this becomes the standard of care. Firstly, there are technical issues. The most critical amongst these involves ensuring that polygenic scores are appropriately calibrated to the ethnicity of the individual being tested. An rsPS or gePS generated using data solely derived from Europeans will have suboptimal ability to capture risk in individuals of non-European origin. The T2D gePS recently released by 23andMe demonstrates a marked fall-off in predictive performance in individuals of Asian and African-American origin [43]. In some settings, these issues with the transethnic portability of polygenic scores go beyond a simple dilution of performance: unpredictable biases and the consequences of genetic drift can result in entirely misleading results [74,75]. Recent studies have also emphasized the impact of residual population stratification effects 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 11 on the performance of these scores [76,77]. RsPS are likely to be more robust to these biases than gePS. The second question to be addressed concerns whether a given polygenic score adds clinical value to the predictions that are possible using existing risk factors. In the case of coronary artery disease, there is evidence that a substantial proportion of those at highest polygenic risk would not have been detected using classical risk factors [13]. In contrast, and as described earlier, the incremental benefit of a polygenic score over easily-accessible clinical parameters seems more limited for T2D, at least when applied at older ages. In fact, the non-genetic risk factors we already collect in clinic (family history, ethnicity, BMI, fat distribution) perform quite well in predicting T2D, particularly in the near term, especially when supported by direct biomarkers of the underlying disease process such as measures of glycemia [30-32,41]. There is an intrinsic limitation to the added value of a polygenic score arising from the fact that trait heritability provides a ceiling for the performance of any purely-genetic measure. Third, there is the issue as to whether early diagnosis can be shown to result in beneficial outcomes, for example by motivating improvements in lifestyle or treatment that reduce the risk of disease. In the case of T2D, the potential for lifestyle modification and/or pharmaceutical intervention (for example with metformin) to reduce diabetes progression is clear [39,40], and these benefits seem to accrue irrespective of genetic risk. In the