Large language models (LLMs) have shown impressive capabilities across diverse settings, but still struggle as the length and complexity of the context increases. To address this challenge, we propose Thinking Recursively and Dynamically (ThReaD). THREAD frames model generation as a thread of execution that, based on the context, can run to completion or dynamically spawn new threads. By spawning, threads can offload work (e.g., thinking, retrieving information) to child threads, which only return tokens needed for the parent thread to do its work. We apply THREAD in the settings of LLM task solving and question answering, where the dynamic threading allows the model to recursively decompose the given task or question into progressively simpler sub-problems that can be solved by separate child threads. We test THREAD, implemented using a few-shot learning approach, on diverse benchmarks for agent tasks and data-grounded question answering. THREAD achieves state-of-the-art performance with GPT-4 and GPT3.5 on these benchmarks, including ALFWorld, TextCraft, and WebShop, along with two new benchmarks, DataCommons QA and MIMIC-III ICU QA. In addition, THREAD outperforms existing frameworks by 10% to 50% absolute points with smaller models, including Llama-3-8b and CodeLlama-7b.

Background: Diabetes is a multifactorial disease with significant genetic predisposition. Polygenic risk scores (PRS) have been developed to estimate an individual's genetic risk of a disease. Traditionally, PRS utilize sex-combined genome-wide association studies (GWAS) due to the limited availability of sex-stratified summary statistics. This study explores sex-dimorphic genetic effects and evaluates the potential benefits of incorporating sex-stratified effects in PRS for type 2 diabetes mellitus (T2DM) and glycemic traits by comparing PRS performance derived from sex-combined versus sex-stratified GWAS. Methods: We performed a sex-heterogeneity test across sex-specific GWAS and identified nine signals with sex-dimorphic effects for T2DM. PRS[sex-combined] and PRS[sex-stratified] were developed using sex-combined and sex-stratified GWAS results for T2DM (41,444 cases and 354,539 controls), fasting glucose (n= 120,595) and fasting insulin (n= 98,210). We evaluated these PRS models in 8,379 participants (1,303 cases and 7,076 controls) from the Framingham Heart Study not included in the PRS derivation. Results: Our findings suggest that sex-combined PRS currently offer better predictive performance for T2DM and glycemic traits. Conclusion: These results highlight the need for larger sex-stratified studies and the optimization of sex-stratified risk models for clinical practice.

Diabetes complications occur at higher rates in individuals of African ancestry. Glucose-6-phosphate dehydrogenase deficiency (G6PDdef), common in some African populations, confers malaria resistance, and reduces hemoglobin A1c (HbA1c) levels by shortening erythrocyte lifespan. In a combined-ancestry genome-wide association study of diabetic retinopathy, we identified nine loci including a G6PDdef causal variant, rs1050828 -T (Val98Met), which was also associated with increased risk of other diabetes complications. The effect of rs1050828 -T on retinopathy was fully mediated by glucose levels. In the years preceding diabetes diagnosis and insulin prescription, glucose levels were significantly higher and HbA1c significantly lower in those with versus without G6PDdef. In the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial, participants with G6PDdef had significantly higher hazards of incident retinopathy and neuropathy. At the same HbA1c levels, G6PDdef participants in both ACCORD and the Million Veteran Program had significantly increased risk of retinopathy. We estimate that 12% and 9% of diabetic retinopathy and neuropathy cases, respectively, in participants of African ancestry are due to this exposure. Across continentally defined ancestral populations, the differences in frequency of rs1050828 -T and other G6PDdef alleles contribute to disparities in diabetes complications. Diabetes management guided by glucose or potentially genotype-adjusted HbA1c levels could lead to more timely diagnoses and appropriate intensification of therapy, decreasing the risk of diabetes complications in patients with G6PDdef alleles.

OBJECTIVE To identify genetic risk factors for incident cardiovascular disease (CVD) among people with type 2 diabetes (T2D). RESEARCH DESIGN AND METHODS We conducted a multi-ancestry time-to-event genome-wide association study for incident CVD among people with T2D. We also tested 204 known coronary artery disease (CAD) variants for association with incident CVD. RESULTS Among 49,230 participants with T2D, 8,956 had incident CVD events (event rate 18.2%). We identified three novel genetic loci for incident CVD: rs147138607 (near CACNA1E/ZNF648, HR 1.23, P=3.6×10-9), rs11444867 (near HS3ST1, HR 1.89, P=9.9×10-9), and rs335407 (near TFB1M/NOX3, HR 1.25, P=1.5×10-8). Among 204 known CAD loci, 5 were associated with incident CVD in T2D (multiple comparison-adjusted P < 0.00024, 0.05/204). A standardized polygenic score of these 204 variants was associated with incident CVD with HR 1.14 (P=1.0×10-16). CONCLUSIONS The data point to novel and known genomic regions associated with incident CVD among individuals with T2D.

Objective: Individuals with diabetes who carry genetic variants that lower hemoglobin A1c (HbA1c) independently of glycemia may have higher real, but undetected, hyperglycemia compared to those without these variants despite achieving similar HbA1c targets, potentially placing them at greater risk for diabetes-related complications. We sought to determine whether these genetic variants, aggregated in a polygenic score, and the large-effect African-ancestry specific missense variant in G6PD (rs1050828) that lower HbA1c were associated with higher retinopathy risk. Research Design and Methods: Using data from 29,828 type 2 diabetes cases of genetically inferred African-American/African-British and European ancestries, we calculated ancestry-specific nonglycemic HbA1c polygenic scores (ngA1cPS) composed of 122 variants associated with HbA1c at genome-wide significance, but not with glucose. We tested the association of the ngA1cPS and the G6PD variant with retinopathy, adjusting for measured HbA1c and retinopathy risk factors. Results: Participants in the bottom quintile of the ngA1cPS showed between 20 to 50% higher retinopathy prevalence, compared to those above this quintile, despite similar levels of measured HbA1c. The adjusted meta-analytic odds ratio for the bottom quintile was 1.31 (95% CI 1.0, 1.73; p=0.05) in African ancestry and 1.31 (95% CI 1.15, 1.50; p=6.5x10-5) in European ancestry. Among individuals of African ancestry with HbA1c below 7%-units, retinopathy prevalence was higher in individuals below, compared to above, the 50th percentile of the ngA1cPS regardless of sex or G6PD carrier status Conclusions: Genetic effects need to be considered to personalize HbA1c targets and improve outcomes of people with diabetes from diverse ancestries.

The All of Us Research Program (AoU) is an initiative designed to gather a comprehensive and diverse dataset from at least one million individuals across the USA. This longitudinal cohort study aims to advance research by providing a rich resource of genetic and phenotypic information, enabling powerful studies on the epidemiology and genetics of human diseases. One critical challenge to maximizing its use is the development of accurate algorithms that can efficiently and accurately identify well-defined disease and disease-free participants for case-control studies. This study aimed to develop and validate type 1 (T1D) and type 2 diabetes (T2D) algorithms in the AoU cohort, using electronic health record (EHR) and survey data. Building on existing algorithms and using diagnosis codes, medications, laboratory results, and survey data, we developed and implemented algorithms for identifying prevalent cases of type 1 and type 2 diabetes. The first set of algorithms used only EHR data (EHR-only), and the second set used a combination of EHR and survey data (EHR+). A universal algorithm was also developed to identify individuals without diabetes. The performance of each algorithm was evaluated by testing its association with polygenic scores (PSs) for type 1 and type 2 diabetes. We demonstrated the feasibility and utility of using AoU EHR and survey data to employ diabetes algorithms. For T1D, the EHR-only algorithm showed a stronger association with T1D-PS compared to the EHR + algorithm (DeLong p-value = 3 x 10-5). For T2D, the EHR + algorithm outperformed both the EHR-only and the existing T2D definition provided in the AoU Phenotyping Library (DeLong p-values = 0.03 and 1 x 10-4, respectively), identifying 25.79% and 22.57% more cases, respectively, and providing an improved association with T2D PS. We provide a new validated type 1 diabetes definition and an improved type 2 diabetes definition in AoU, which are freely available for diabetes research in the AoU. These algorithms ensure consistency of diabetes definitions in the cohort, facilitating high-quality diabetes research.

OBJECTIVE:The clinical utility of genetic information for type 2 diabetes (T2D) prediction with polygenic score (PGS) in ancestrally diverse, real-world US healthcare systems is unclear, especially for those at low clinical phenotypic risk for T2D.RESEARCH DESIGN AND METHODS:We tested the association of PGS with T2D incidence in patients followed within a primary care practice network over 16 years in four hypothetical scenarios that varied by clinical data availability (N = 14,712): 1) age and sex, 2) age, sex, BMI, systolic blood pressure, and family history of diabetes; 3) all variables in (2) and random glucose; 4) all variables in (3), HDL, total cholesterol, and triglycerides, combined in a clinical risk score (CRS). To determine whether genetic effects differed by baseline clinical risk, we tested for interaction with the CRS.RESULTS:PGS was associated with incident diabetes in all models. Adjusting for age and sex only, the Hazard Ratio (HR) per PGS standard deviation (SD) was 1.76 (95% CI 1.68, 1.84) and the HR of top 5% of PGS vs interquartile range (IQR) was 2.80 (2.39, 3.28). Adjusting for the CRS, the HR per SD was 1.48 (1.40, 1.57) and HR of top 5% of PGS vs IQR was 2.09 (1.72, 2.55). Genetic effects differed by baseline clinical risk [(PGS-CRS interaction p =0.05; CRS below the median: HR 1.60 (1.43, 1.79); CRS above the median: HR 1.45 (1.35, 1.55)].CONCLUSIONS:Genetic information can help identify high-risk patients even among those perceived to be low risk in a clinical evaluation.

Type 2 diabetes (T2D) genome-wide association studies (GWASs) often overlook rare variants as a result of previous imputation panels' limitations and scarce whole-genome sequencing (WGS) data. We used TOPMed imputation and WGS to conduct the largest T2D GWAS meta-analysis involving 51,256 cases of T2D and 370,487 controls, targeting variants with a minor allele frequency as low as 5 x 10-5. We identified 12 new variants, including a rare African/African American-enriched enhancer variant near the LEP gene (rs147287548), associated with fourfold increased T2D risk. We also identified a rare missense variant in HNF4A (p.Arg114Trp), associated with eightfold increased T2D risk, previously reported in maturity-onset diabetes of the young with reduced penetrance, but observed here in a T2D GWAS. We further leveraged these data to analyze 1,634 ClinVar variants in 22 genes related to monogenic diabetes, identifying two additional rare variants in HNF1A and GCK associated with fivefold and eightfold increased T2D risk, respectively, the effects of which were modified by the individual's polygenic risk score. For 21% of the variants with conflicting interpretations or uncertain significance in ClinVar, we provided support of being benign based on their lack of association with T2D. Our work provides a framework for using rare variant GWASs to identify large-effect variants and assess variant pathogenicity in monogenic diabetes genes. Rare variant analyses identify a new type 2 diabetes risk allele near the LEP gene, which encodes leptin, and other risk alleles of intermediate penetrance in genes previously implicated in monogenic forms of diabetes.

Introduction: People with type 2 diabetes (T2D) who carry genetic variants that lower hemoglobin A1c (A1c) independently of glycemia may have higher real, but undetected, hyperglycemia compared to those who do not despite achieving similar A1c targets, placing them at greater risk for complications. Methods: Using data from 39,199 T2D cases of European ancestry (EA) and African ancestry (AA) in UK Biobank and All of Us, we calculated a nonglycemic A1c polygenic score (ngA1cPS) composed of 123 variants previously associated with A1c at genome-wide significance, but not with glucose. We tested the association of the ngA1cPS with DR (n = 3,859), adjusting for A1c and known DR risk factors (T2D duration, kidney disease, hypertension, lipids, smoking). Results: Participants at the bottom ngA1cPS quintile showed 25-50% higher DR prevalence vs. the rest despite having lower measured A1c (6.6 vs 6.7%). Odds ratios were 1.28 in EA and 1.37 in AA (Fig. 1). Among carriers of the African-specific G6PD variant rs1050828 known to lower A1c, DR prevalence was 27% higher among those with ngA1cPS below the median vs. above the median. Conclusions: The aggregate effect of variants that lower A1c independently of glycemia is associated with a higher DR risk in both EA and AA. Genetic effects need to be considered to define personalized A1c targets to improve the outcomes of patients with T2D. Disclosure P.H.Schroeder: None. R.Mandla: None. J.C.Florez: Consultant; AstraZeneca, Novo Nordisk, Other Relationship; AstraZeneca, Merck & Co., Inc. J.M.Mercader: None. A.Leong: Other Relationship; Merck & Co., Inc. Funding Doris Duke Charitable Foundation (2020096)

OBJECTIVEQuantify the impact of genetic and socioeconomic factors on risk of type 2 diabetes (T2D) and obesity. RESEARCH DESIGN AND METHODSAmong participants in the Mass General Brigham Biobank (MGBB) and UK Biobank (UKB), we used logistic regression models to calculate cross-sectional odds of T2D and obesity using 1) polygenic risk scores for T2D and BMI and 2) area-level socioeconomic risk (educational attainment) measures. The primary analysis included 26,737 participants of European genetic ancestry in MGBB with replication in UKB (N = 223,843), as well as in participants of non-European ancestry (MGBB N = 3,468; UKB N = 7,459). RESULTSThe area-level socioeconomic measure most strongly associated with both T2D and obesity was percent without a college degree, and associations with disease prevalence were independent of genetic risk (P < 0.001 for each). Moving from lowest to highest quintiles of combined genetic and socioeconomic burden more than tripled T2D (3.1% to 22.2%) and obesity (20.9% to 69.0%) prevalence. Favorable socioeconomic risk was associated with lower disease prevalence, even in those with highest genetic risk (T2D 13.0% vs. 22.2%, obesity 53.6% vs. 69.0% in lowest vs. highest socioeconomic risk quintiles). Additive effects of genetic and socioeconomic factors accounted for 13.2% and 16.7% of T2D and obesity prevalence, respectively, explained by these models. Findings were replicated in independent European and non-European ancestral populations. CONCLUSIONSGenetic and socioeconomic factors significantly interact to increase risk of T2D and obesity. Favorable area-level socioeconomic status was associated with an almost 50% lower T2D prevalence in those with high genetic risk.

OBJECTIVE The study aimed to develop and validate algorithms for identifying people with type 1 and type 2 diabetes in the All of Us Research Program (AoU) cohort, using electronic health record (EHR) and survey data. RESEARCH DESIGN AND METHODS Two sets of algorithms were developed, one using only EHR data (EHR), and the other using a combination of EHR and survey data (EHR+). Their performance was evaluated by testing their association with polygenic scores for both type 1 and type 2 diabetes. RESULTS For type 1 diabetes, the EHR-only algorithm showed a stronger association with T1D polygenic score (p=3x10-5) than the EHR+. For type 2 diabetes, the EHR+ algorithm outperformed both the EHR-only and the existing AoU definition, identifying additional cases (25.79% and 22.57% more, respectively) and showing stronger association with T2D polygenic score (DeLong p=0.03 and 1x10-4, respectively). CONCLUSIONS We provide new validated definitions of type 1 and type 2 diabetes in AoU, and make them available for researchers. These algorithms, by ensuring consistent diabetes definitions, pave the way for high-quality diabetes research and future clinical discoveries.

Introduction: Glucocorticoids are commonly prescribed medications with a known side effect of hyperglycemia due to their effect on glucose metabolism. However, only 10-50% of patients receiving glucocorticoids develop overt hyperglycemia. It is unknown whether patients with increased genetic risk for type 2 diabetes (T2D) are predisposed to develop hyperglycemia after glucocorticoid treatment. Methods: We accessed electronic health records and genetic data in the Mass General Brigham Biobank. We examined patients with no diagnosis of diabetes or prediabetes who received a glucocorticoid dose equivalent to 10 mg of prednisone or more, and who had glucose levels checked within 7 days of glucocorticoid administration. Hyperglycemia was defined as fasting blood glucose ≥126 mg/dL or random blood glucose ≥200 mg/dL. A T2D global extended polygenic score was constructed through a meta-analysis of T2D genome-wide association studies in the Million Veteran Program, DIAMANTE, and FinnGen. We performed logistic regression to analyze the association between the polygenic score and hyperglycemia, controlling for age, gender, BMI, baseline creatinine, glucocorticoid dose and duration, and the first 10 principal components of genetic ancestry. Results: Out of 552 patients who received glucocorticoids, 216 developed hyperglycemia and 336 did not. The T2D polygenic score was significantly associated with glucocorticoid-induced hyperglycemia (p = 0.032, OR = 1.5 per standard deviation of the polygenic score). Other significant covariates included baseline creatinine (p = 8.0 × 10−4, OR = 2.2 per mg/dL of creatinine) and glucocorticoid dose (p = 9.5 × 10−5, OR = 1.01 per 10 mg of prednisone). Conclusions: Patients carrying a higher burden of genetic variants that confer risk for T2D have an increased risk of hyperglycemia after receiving glucocorticoids. This finding offers a mechanism for risk stratification as part of a precision approach to medical treatment. Disclosure A.J.Deutsch: None. P.H.Schroeder: None. R.Mandla: None. F.Erenler: None. J.M.Mercader: None. M.Udler: None. J.C.Florez: Consultant; AstraZeneca, Novo Nordisk, Other Relationship; AstraZeneca, Merck & Co., Inc. L.Brenner: None. Funding National Institutes of Health (T32DK007028)

ABSTRACTHypothesisThe prevalence of type 2 diabetes is higher in Latino populations compared with other major ancestry groups. Not only has the Latino population been systematically underrepresented in large-scale genetic analyses, but previous studies relied on the imputation of ungenotyped variants based on the 1000 Genomes (1000G) imputation reference panel, which results in suboptimal capture of low-frequency or Latino-enriched variants. The NHLBI Trans-Omics for Precision Medicine (TOPMed) reference panel represents a unique opportunity to analyze rare genetic variations in the Latino population.MethodsWe evaluate the TOPMed imputation performance using genotyping array and whole-exome sequence data in 6 Latino cohorts. To evaluate the ability of TOPMed imputation of increasing the identified loci, we performed a Latino type 2 diabetes GWAS meta-analysis in 8,150 type 2 diabetes cases and 10,735 controls and replicated the results in 6 additional cohorts including whole-genome sequence data from the All of Us cohort.ResultsWe show that, compared to imputation with 1000G, the TOPMed panel improves the identification of rare and low-frequency variants. We identified 26 distinct signals including a novel genome-wide significant variant (minor allele frequency 1.6%, OR=2.0, P=3.4×10−9) near ORC5. A Latino-tailored polygenic score constructed from our data and GWAS data from East Asian and European populations improves the prediction accuracy in a Latino target dataset, explaining up to 7.6% of the type 2 diabetes risk variance.ConclusionsOur results demonstrate the utility of TOPMed imputation for identifying low-frequency variation in understudied populations, leading to the discovery of novel disease associations and the improvement of polygenic scores.

To identify potential subtypes of type 2 diabetes (T2D) anchored in genetics but informed by physiology, we previously used a soft clustering method to cluster T2D single nucleotide variants (SNVs) by their associated metabolic traits. The resulting European-based clusters represent likely disease mechanistic pathways. However, ancestry-specific SNVs, phenotype characteristics and prevalence rates suggest that a portion of the T2D’s heterogeneity is population-based, requiring expansion of this work to non-European populations. We created a semi-automated Bayesian non-negative matrix factorization (bNMF) pipeline to generate new ancestry-specific clusters in European (EUR, 390 SNVs), East Asian (EAS, 326 SNVs) and African (AFR, 172 SNVs), as well as trans-ancestry (TA, 498 SNVs), using up to 89 T2D-related traits. We validated the clusters by replicating their trait associations in the Mass General Brigham Biobank cohort (MGBB, N=62,252). The new ancestry-specific and TA clusters captured previously identified clusters, as well as novel clusters related to possible mechanisms of insulin resistance. In the 11 TA clusters, 127/498 SNVs were from non-EUR T2D studies, with 87 SNVs not represented in the EUR clusters. In the non-European subset of MGBB (N=8,990), 8 of 10 TA cluster pPS were more strongly associated with T2D compared to the corresponding EUR cluster pPS. We assessed in MGBB whether the proportion of cumulative genetic risk attributed to each cluster differed between sub populations. A significantly higher proportion was attributed to the Lipodystrophy cluster in EAS, to the Obesity cluster in EUR, and to the Liver/Lipid cluster in AFR (all t-test P<10-15). By expanding our previous clusters to include non-European SNVs, we were able to identify new genetic clusters and improve the predictive ability of cluster pPS. Our results suggest that polygenic processes contribute in different proportions across populations. Disclosure K.Smith: None. A.Manning: None. J.M.Mercader: None. M.Udler: None. H.Kim: None. K.E.Westerman: None. S.Hsu: None. R.Mandla: None. P.H.Schroeder: None. T.Majarian: Employee; Vertex Pharmaceuticals Incorporated. V.Kaur: None. J.C.Florez: Consultant; AstraZeneca, Novo Nordisk, Other Relationship; AstraZeneca, Merck & Co., Inc. Funding National Institute of Diabetes and Digestive and Kidney Diseases (R03DK131249)

Introduction: The aggregate effect of genetic variants reported to lower A1c independently of glycemia by ancestry remains unclear. Methods: We recalled 177 patients from the Mass General Brigham biobank with genetic data, enriching recruitment for carriers of the African G6PD variant (rs1050828), α thalassemia 3.7kb deletion, and people in the tails of a European polygenic score (PS) composed of variants associated with A1c but not glucose in GWAS. Subjects underwent 14 days of continuous glucose monitoring (CGM). We defined genetic ancestry using principal component analysis, and calculated A1c-glycemia discordance (measured A1c minus estimated A1c by CGM; ADAG study equation). Results: Measured A1c and mean glucose were correlated in all groups (r2>0.9). Mean A1c-glycemia discordance was 0.4%-unit higher in African vs. European ancestry. In African ancestry, discordance was 0.5%-unit lower in carriers of the G6PD T allele vs. noncarriers. The bottom decile of PS had 0.3%-unit lower discordance vs. top decile (Figure; p<0.01; all comparisons). Discordance was similar regardless of α thalassemia deletion. Conclusions: While A1c is an excellent proxy for glycemia in all groups, the A1c-glucose relationship differed by genotype and ancestry. Genetic variation should be considered to promote care equity in diverse populations and avoid misestimating glycemia when using A1c in clinical practice. Disclosure N. Thangthaeng: None. M. N. Facibene: None. S. Kartik: None. R. Mandla: None. P. H. Schroeder: None. N. R. Norgil: None. J. M. Mercader: None. A. Leong: Other Relationship; Merck & Co., Inc. Funding Doris Duke Charitable Foundation (2020096)

SUGAR-MGH is a pharmacogenetic resource for characterizing genetic influences on pharmacological perturbations relevant to type 2 diabetes (T2D). 1,000 participants who were naïve to T2D treatment received 5 mg glipizide, followed by 4 doses of 500 mg metformin and a 75-g oral glucose tolerance test (OGTT) a week later. Glucose and insulin were measured at pre-specified endpoints after glipizide and during the OGTT. Incretin levels were measured in a subset. We calculated physiologic endpoints reflecting insulin secretion, insulin sensitivity, and incretin secretion. The mean age was 47 years, 54% were women, >35% were non-white, and the mean BMI was 30.2 kg/m2. 890 participants underwent genome-wide genotyping using the Illumina Multi-Ethnic Genotyping Array, and high-quality imputation was performed with the TOPMed reference panel. Multiple linear regression using an additive model tested for association between genetic variants and endpoints. We identified 21 genome-wide significant variants, of which 3 met experiment-wide significance (P<8.3×10-9 for 2 drugs × 3 physiological hypotheses tested). The most significant association was between an African ancestry-specific variant (minor allele frequency=0.026) at rs149403252 and lower fasting glucose following metformin, adjusted for baseline glucose (P=1.9×10-9). We tested the influence of 429 known T2D-associated variants and found that the protective C allele of rs703972 near ZMIZ1 was associated with increased secretion of active GLP-1 in response to metformin (P=1.6×10-5), suggesting that an enhanced incretin response may be a mechanism by which this variant decreases T2D risk. We illustrate that a genome-wide approach in a multi-ethnic human perturbation study can uncover new variation associated with drug response and provide insight into mechanisms of action of known T2D genetic variation. Disclosure J. H. Li: None. V. Kaur: None. L. Brenner: Research Support; Self; Apple. J. M. Mercader: None. J. C. Florez: Consultant; Self; Goldfinch Bio, Inc., Other Relationship; Self; Novo Nordisk. Funding National Institutes of Health (R01DK088214, R03DK077675, P30DK036836, M01RR01066, 1UL1RR025758-04, 8UL1TR000170-05, T32DK007028)

We identified genetic subtypes of type 2 diabetes (T2D) by analyzing genetic data from diverse groups, including non-European populations. We implemented soft clustering with 650 T2D-associated genetic variants, capturing known and novel T2D subtypes with distinct cardiometabolic trait associations. The twelve genetic clusters were distinctively enriched for single-cell regulatory regions. Polygenic scores derived from the clusters differed in distribution between ancestry groups, including a significantly higher proportion of lipodystrophy-related polygenic risk in East Asian ancestry. T2D risk was equivalent at a BMI of 30 kg/m2 in the European subpopulation and 24.2 (22.9-25.5) kg/m2 in the East Asian subpopulation; after adjusting for cluster-specific genetic risk, the equivalent BMI threshold increased to 28.5 (27.1-30.0) kg/m2 in the East Asian group, explaining about 75% of the difference in BMI thresholds. Thus, these multi-ancestry T2D genetic subtypes encompass a broader range of biological mechanisms and help explain ancestry-associated differences in T2D risk profiles.

OBJECTIVE:To assess whether increased genetic risk of type 2 diabetes (T2D) is associated with the development of hyperglycemia after glucocorticoid treatment.RESEARCH DESIGN AND METHODS:We performed a retrospective analysis of individuals with no diagnosis of diabetes who received a glucocorticoid dose of ≥10 mg prednisone. We analyzed the association between hyperglycemia and a T2D global extended polygenic score, which was constructed through a meta-analysis of two published genome-wide association studies.RESULTS:Of 546 individuals who received glucocorticoids, 210 developed hyperglycemia and 336 did not. T2D polygenic score was significantly associated with glucocorticoid-induced hyperglycemia (odds ratio 1.4 per SD of polygenic score; P = 0.038).CONCLUSIONS:Individuals with increased genetic risk of T2D have a higher risk of glucocorticoid-induced hyperglycemia. This finding offers a mechanism for risk stratification as part of a precision approach to medical treatment.

Introduction: Most genome-wide association studies (GWAS) of type 2 diabetes (T2D) assume an additive mode of inheritance and equal effects in men and women. These assumptions can preclude the discovery of variants with effects that are non-additive and/or sex-specific. Focused exploration of these effects may reveal novel genetic variation associated with T2D and improve the predictive performance of polygenic risk scores (PRS) . Methods: We performed a sex-stratified GWAS, using additive and recessive models for T2D, in individuals of European ancestry in the UK Biobank (UKBB) and Genetic Epidemiology Research on Aging cohort. We also generated sex-specific and non-sex-specific T2D PRSs, using PRS-CS, and assessed their performance in Mass General Brigham Biobank (MGBB) . Results: As the largest sex-stratified additive and recessive GWAS of T2D performed to date, this study included 30,625 cases and 223,442 controls. In the recessive analysis, we identified 7 novel variants, of which 1 was female-specific and 1 was male-specific. Among these variants, 4 were associated with over 10-fold increase in risk for T2D. The male-specific variant, rs35725476 (OR = 1.49, p = 3×10- 9) , is associated with higher expression of a long non-coding RNA in pancreatic islets (p = 3×10- 14) . In the additive analysis, we identified a novel female-specific protective variant, rs12109272 (OR = 0.91, p = 3×10-8) , which is associated with lower PCSK1 expression, lower risk of gestational diabetes (p = 1×10- 6) and lower fasting glucose (p = 6×10- 32) in independent cohorts. Finally, the sex-specific PRS outperformed the non-sex-specific PRS at predicting T2D in MGBB (AUC of 0.653 versus 0.643, p [of difference] = 2×10-4 in males; AUC of 0.674 versus 0.656, p [of difference] = 3×10-4 in females) . Conclusions: These findings demonstrate the value of non-additive and sex-stratified analyses for both variant discovery and improving polygenic prediction for T2D. Disclosure P.H.Schroeder: None. J.B.Cole: None. A.Leong: None. J.C.Florez: Consultant; AstraZeneca, Goldfinch Bio, Inc., Other Relationship; AstraZeneca, Merck & Co., Inc., Novo Nordisk. J.M.Mercader: None. Funding American Diabetes Association (1-19-ICTS-068) ; National Human Genome Research Institute (U01HG011723)