Background: Insulin resistance (IR) is a major risk factor for Alzheimer’s disease (AD) and is primarily driven by obesity, another AD risk factor. The biological mechanism by which IR predisposes to AD is unknown. Genetic clustering analyses of type 2 diabetes (T2D) loci can help characterize which mechanism of impaired insulin action may be involved in AD and related traits.
Method: We constructed five genetic clusters related to IR (Obesity, Lipodystrophy, Liver/Lipid, ALP [alkaline phosphatase] negative, and Hyper-Insulin Secretion) based on a recent clustering analysis of T2D loci (PMID: 36538063). We evaluated the association of each cluster with the Homeostatic Model Assessment for Insulin Resistance (HOMA-IR) and neurological traits (AD, dementia, general cognitive function, and four brain MRI volumes) in >38k participants (36% men, mean age 55yrs (14.5)) with diverse race or ethnicity, from the Trans-Omics for Precision Medicine (TOPMed) Program. We conducted both pooled and race/ethnicity-stratified association analyses (European-50%, African-American-22.5%, and Hispanic/Latino-21%). We used logistic or linear mixed-effect models, adjusted for age, sex, study, and the first 11 genetic principal components. We accounted for relatedness and allowed for heterogeneous variances across studies.
Result: We confirmed the association of each IR genetic cluster with HOMA-IR in the pooled analysis (2.6E-66≤P≤2E-4) as well as in group-stratified analyses (6.3E-39≤P≤0.05), except for the Hyper-Insulin Secretion cluster in African-Americans (P=0.30). The genetic clusters were not significantly associated with neurological outcomes after adjusting for multiple testing (P=0.05/Ngroups/Nclusters/Noutcomes=0.05/4/5/7=3.6×10-4). We observed nominal associations (P<0.05) for two genetic clusters. The Lipodystrophy cluster was negatively associated with intracranial volume in the pooled analysis (beta=-0.51, P=0.003) as well as in European (beta=-0.47, P=0.02) and Hispanic/Latino (beta=-1.23, P=0.01) participants. The Hyper-Insulin Secretion cluster was negatively associated with hippocampal volume in Hispanic/Latino participants (beta=-0.005, P=0.03), and positively associated with AD and dementia (OR=1.02 [1.001-1.038], P=0.03) in European participants.
Conclusion: Our findings are consistent with the literature suggesting that IR may be associated with lower brain volumes and increased AD risk. Our analyses shed light on two biological mechanisms of impaired insulin action potentially involved in AD and related traits, with genetic associations that may differ by race or ethnicity.
Funding: R00AG066849

Type 2 diabetes (T2D) has been well characterized by genetic studies with over 550 risk loci identified in mostly European populations. To better understand the genetic basis of T2D in diverse ancestries, we performed a whole genome sequence association analysis with T2D and related traits of fasting insulin (FI), glucose (FG) and hemoglobin A1c (HbA1c). We then used the colocalization method COLOC to evaluate if the risk locus and expression quantitative trait loci (eQTL) region shared a variant. Four relevant eQTL specific to tissues or cell types were obtained from the NHLBI Trans Omics for Precision Medicine (TOPMed) program, including 6,602 samples for whole blood, 1,265 for peripheral blood mononuclear cells (PBMC), 368 for T cells, and 352 for monocytes. A signal was considered colocalized if the posterior probability of colocalization (PP.coloc) was >0.5.

Our analysis included 21,913 prevalent T2D cases, 61,036 controls and up to 50,201 non-T2D individuals with FG, FI or HbA1c from African, Atlantic Islander, East Asian, European, Hispanic, and Samoan groups. Single variant analysis identified 35 variants in known regions that were significantly associated with T2D or related traits (P < 4.0×10^-9). Among these regions, we detected six genes (SLC2A2, NKX6-3, KCNQ1, HBB, FN3K, and G6PD) having colocalizing signals in at least one related tissue or cell type. For example, variant rs1050828 near G6PD on chromosome X was strongly associated with HbA1c (P <1.0×10^-200). Colocalization analysis revealed that rs1050828 also affected the gene expression in various tissues or cell types, including BCAP31 in whole blood, TKTL1 in PBMC, MPP1 in T cell, and RENBP in monocytes. This novel finding, obtained using TOPMed eQTL data from multiple cohorts, suggests a link between risk-associated variation and effector genes, providing insight into potential causal genes and regulatory variants at T2D risk loci in diverse populations.

Background: Type 2 diabetes (T2D) is a major public health concern for the nation of Samoa in the South Pacific. There have been no genome-wide studies of T2D or associated phenotypes in individuals of Polynesian ancestry, whose population history might result in allele frequencies allowing for the observation of associations undetectable in other populations. Here, we performed genome-wide association studies (GWAS) of six metabolic traits: T2D, fasting glucose, fasting insulin, and HOMA‑IR, as well as T2D among individuals with obesity and T2D among individuals without obesity.
Methods: Genotypes were measured from two sources. First 659,492 variants were genotyped on an Affymetrix 6.0 array in 2,890 Samoan individuals recruited in 2010. We then imputed an additional 9 million variants using a Samoan-specific haplotype reference panel derived from 1,285 Samoan individuals whole-genome sequenced by the TOPMed Program. We performed association testing using linear or logistic mixed models adjusting for fixed effects of age, sex, and principal components of ancestry derived through PC‑AiR and for random effects of kinship derived from PC-Relate.
Results: There were seven unique loci associated with metabolic phenotypes at p < 5 × 10⁻⁸, and an additional seventeen unique loci were associated at p < 1 × 10⁻⁶. The most significant signal is in intron 2 of PARD3B: an association between rs76755625 and T2D among individuals without obesity (p = 9.42 × 10⁻¹¹). This locus has been associated with T2D in an earlier multiancestry GWAS and has been associated body mass index in several GWASs. Notable among the other associating loci, variants in PPARGC1A were associated with T2D (peak p = 4.22 × 10⁻⁹). This locus has not been associated with T2D in genome-wide association studies to date. It encodes PPARγ coactivator 1α, which is a transcriptional coactivator for energy metabolism genes and regulates liver gluconeogenesis. Ppargc1a-knockout mice exhibit disrupted adipose morphology and abnormal glucose homeostasis.
Conclusion: We observed several known and novel genetic loci associated with metabolic phenotypes in Samoans, suggesting that while some of the genetic architecture of metabolic phenotypes is shared across ancestries, there may be unique associations among Polynesians. We are currently investigating potential associations between these loci and metabolic phenotypes in an independent cohort of Samoan adults. Additional studies will be necessary to validate these associations and the determine the biological underpinnings of these associations, which may point to previously unknown biological mechanisms or social determinants of metabolic health.

Objective:
To better characterize genetic variations underlying neurological traits by leveraging whole-genome sequencing (WGS) data in participants of diverse ancestry from the Trans-Omics for Precision Medicine (TOPMed) program.

Background:
Genome-wide association studies (GWAS), conducted mainly in European (EA) participants, have identified common genetic variants with modest effect sizes for brain volumes, accepted endophenotypes of vascular brain injury, and few genetic loci for insulin resistance (IR), a major risk factor for stroke and dementia.

Design/Methods:
1. We performed WGS association analyses of hippocampal (HV), total brain, lateral ventricular (LVV), and intracranial (ICV) volumes in ~8k participants (62% EA, 21% African-American (AA), 16% Hispanic/Latino (HA)).
2. We derived three ancestry-specific polygenic scores (PSEA, PSAA, PSHA) based on UK Biobank reference panels and fasting insulin GWAS summary statistics, adjusted for body mass index. We generated a multi-ancestry PS (Multi-PS) by fitting a linear combination of the standardized PSEA, PSAA, and PSHA that most accurately predicted IR in a validation set of ~17k participants without diabetes (34% EA, 28% AA, 38% HA). We evaluated the association of the PSs with neurological traits in a testing set of ~14k participants (66% EA, 22% AA, 11% HA).
Mixed-effect models were used for all analyses, adjusted for age, sex, study, and principal components, and accounting for relatedness and trait variance variability. Brain volume analyses were adjusted for ICV and excluded participants with dementia or stroke.

Results:
We identified novel significant hits (P<5×10-8) at 2q22 & 5q14 for LVV and at 1q32 & 13q14 for HV. The top 13q14 variant was common in AA (13%), less frequent in HA (1.4%), and rare in EA (0.1%), and had a consistent effect size across population groups (-0.27 to -0.34). The Multi-PS was strongly associated with IR (proportion of variance explained: 12%). The PSEA was significantly (P<0.002) associated with ICV (P=7×10-7), and suggestively with LVV (P=0.004).

Conclusions:
By leveraging TOPMed multi-ancestry and WGS data, we identified new loci underlying brain volumes, including ancestry-specific associations, and confirmed the association of IR with brain volumes.

Disclosure and Study Support:
None.
NINDS F30NS124136, NIA T32AG57461, K99AG066849, U01AG058589, NHLBI R01s HL131136, HL105756

Insulin resistance (IR) is a major risk factor for Alzheimer’s Disease (AD) and has been associated with cognitive impairment, dementia, and neurodegeneration. Few genetic loci have been uncovered by genome-wide association studies (GWAS) of IR, limiting the proportion of variance explained (PVE) of IR genetic instruments. We constructed polygenic scores (PSs) for IR based on whole-genome sequencing (WGS) data from the Trans-Omics for Precision Medicine (TOPMed). We derived ancestry-specific PSs using PRS-CSx based on ancestry-specific UK Biobank reference panels and fasting insulin (FI) GWAS summary statistics adjusted for body mass index (BMI). We generated a multi-ancestry PSFI (Multi-PSFI) by fitting a linear combination of the standardized ancestry-specific PSFIs that most accurately predicted HOMA-IR in Nvalidation~17k participants without diabetes (34% European-EA, 28% African-AA, 38% Hispanic-HA), and evaluated its association with AD, dementia, general cognitive function (GENCOG), and four brain volumes in Ntesting~14k participants (66% EA, 22% AA, 11% HA). Logistic or linear mixed-effect models were adjusted for age, sex, study, 11 genetic principal components, and accounted for relatedness using a genetic relationship matrix. IR, GENCOG, and brain volumes analyses were additionally adjusted for BMI, education, and intracranial volume respectively. Statistically significant associations were defined by P<0.05/Ntraits/NPSs=0.05/7/4=0.002. Multi-PSFI was strongly associated with HOMA-IR (PVEJoint=12%). No significant or suggestive association was detected for the Multi-PSFI, the AA-PSFI, or the HA-PSFI with the neurological outcomes. EA-PSFI was significantly associated with intracranial volume (P=7E-07), and suggestively with lateral ventricular (P=0.004) and total brain volumes (P=0.05). By leveraging multi-ancestry and WGS data, we increased the IR PSs PVE and confirmed the association of IR with brain volumes. Funded by NIA K99AG066849.

Background: Insulin resistance (IR) is a major risk factor for Alzheimer’s Disease (AD) and has been associated with cognitive impairment, dementia, and neurodegeneration. Though the association between IR and AD has been explored genetically, the proportion of variance explained (PVE) by the IR genetic instruments has been limited, due to few genetic loci identified in genome-wide association studies (GWAS) of IR.
Methods: We constructed polygenic scores (PSs) for fasting insulin (FI) based on the Trans-Omics for Precision Medicine (TOPMed) Freeze 9 whole-genome sequencing (WGS) data. We used PRS-CSx (Ruan et al., Nat Genet. 2022) to generate ancestry-specific PSs (European, African, and Hispanic/Latino), with weights derived based on ancestry-specific UK Biobank reference panels and FI GWAS summary statistics adjusted for body mass index (BMI) (Chen et al., Nat Genet., 2021). We used MetaSubtract (Nolte et al., Eur J Hum Genet. 2017) to remove the effect of TOPMed studies from the FI meta-analyses. We generated a multi-ancestry PSFI by fitting a linear combination of the standardized ancestry-specific PSs that most accurately predicted HOMA-IR in five TOPMed cohorts (validation set, N~17k participants [34% European, 28% African, 38% Hispanic] without diabetes), adjusting for BMI. We then evaluated the association of the multi-ancestry PSFI with HOMA-IR, AD, dementia, general cognitive function, and four brain volumes in eight TOPMed cohorts (testing set, ~14k participants [66% European, 22% African, 11% Hispanic]). Association analyses were performed using logistic or linear mixed-effect models adjusted for age, sex, study, 11 genetic principal components, and accounting for relatedness using a genetic relationship matrix. General cognitive function and brain volumes analyses were additionally adjusted for education and intracranial volume (ICV) respectively. We used a threshold of P<0.05/Ntraits/Ntests=0.05/7/4=0.002 to define an association as significant.
Results: The multi-ancestry PSFI was strongly associated with HOMA-IR (PVEJoint=12%). No significant or suggestive association was detected for the multi-ancestry, the African, or the Hispanic PSFI with any of the neurological outcomes. The European PSFI was significantly associated with ICV (P=7E-07), and suggestively associated with lateral ventricular (P=0.004) and total brain volumes (P=0.05).
Conclusion: By leveraging multi-ancestry and WGS data, we increased the PVE of the genetic instruments for IR and confirmed the association of IR with brain volumes. The identified European-specific associations require further investigation.
Funding: NIH K99AG066849-02