---
title: "Genomic SEM identifies shared genetic architecture and candidate genes for insulin resistance–rela"
id: "plos-one-18-integrative-genomic-sem-analysis-of-shared-genetic-architecture-and-candidate"
canonical_url: "https://medichelpline.com/clinical-feed/plos-one-18-integrative-genomic-sem-analysis-of-shared-genetic-architecture-and-candidate"
content_type: "clinical_feed_article"
specialty: "Endocrinology"
source_name: "PLOS ONE (Medicine)"
source_url: "https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0361601"
published_at: "2026-10-29T14:00:00.000Z"
evidence_level: "Journal Feed"
license: "CC-BY-NC-4.0 / Informational Use"
---
# Genomic SEM identifies shared genetic architecture and candidate genes for insulin resistance–rela
## Provenance & Clinical Metadata
- **Canonical URL:** https://medichelpline.com/clinical-feed/plos-one-18-integrative-genomic-sem-analysis-of-shared-genetic-architecture-and-candidate
- **Specialty:** [Endocrinology](https://medichelpline.com/clinical-feed/endocrinology.md)
- **Primary Source:** PLOS ONE (Medicine)
- **Source URL:** [Original Journal Publication](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0361601)
- **Published At:** 2026-10-29T14:00:00.000Z
- **Evidence Rating:** Journal Feed
## Executive GIST (TL;DR)
- Investigators used **Genomic SEM** to integrate GWAS summary statistics for fasting insulin, insulin sensitivity index, triglycerides, and waist-to-hip ratio to define a single latent genetic factor representing shared insulin resistance-related biology. - The single-factor model delivered acceptable but not optimal approximate fit; loadings were positive for fasting insulin, triglycerides, and waist-to-hip ratio, and negative for insulin sensitivity index. - SNP-level GWAS of the latent factor produced detectable SNP heritability and yielded 3,067 genome-wide significant candidate SNPs; Q_SNP was retained as a heterogeneity diagnostic. - Functional annotation by FUMA defined 100 genomic risk loci and MAGMA identified 1,530 FDR-significant genes; Bonferroni-significant genes were used for pathway analyses. - Fine-mapping using SuSiE and FINEMAP prioritized regions including loci near **PPARG**, RSPO3, SLC38A11, SLC39A8, BCL2, GDF5, and SPATA5; prioritized variants were cross-validated between methods. - TWAS (FUSION, GTEx v8 sCCA weights) and FOCUS gene-level fine-mapping emphasized genes implicated in adipocyte function, glucose–lipid metabolism, and insulin secretion, such as **PPARG**, MLXIPL, TCF7L2, FAM13A, CCDC92, and SLC22A3. - Enrichment analyses linked the latent-factor genetic signal to fat distribution, cardiometabolic traits, adipose tissue and mesenchymal lineages, and active regulatory genomic regions. - Separate exploratory structural work used AlphaFold3, ThermoMPNN, and molecular dynamics to model three ultra-rare missense substitutions (BCL2-W188G, GDF5-C400Y, RSPO3-C129Y); these substitutions were not lead GWAS or fine-mapped causal variants and do not establish mechanisms. - Authors emphasized that the latent factor is a shared genetic phenotype rather than a clinical diagnostic, and they called for further experimental validation of prioritized genes and structural predictions.
## Clinical Analysis & Structured Key Points
Integrative genomic SEM analysis of shared genetic architecture and candidate genes for insulin resistance–related metabolic traits | PLOS One Browse Subject Areas ? Click through the PLOS taxonomy to find articles in your field. For more information about PLOS Subject Areas, click here . Article Authors Metrics Comments Media Coverage Reader Comments Figures Figures Abstract Insulin resistance is a complex metabolic state involving impaired insulin sensitivity, compensatory insulin secretion, abnormal lipid metabolism, and altered fat distribution. However, the shared genetic architecture underlying these correlated phenotypes remains incompletely defined. In this study, we applied Genomic Structural Equation Modeling to genome-wide association summary statistics of fasting insulin, insulin sensitivity index, triglycerides, and waist-to-hip ratio to construct an insulin resistance-related latent genetic factor. SNP-level association analysis was then performed for the latent factor, followed by risk locus annotation, fine-mapping, transcriptome-wide association analysis, gene-level prioritization, enrichment analysis, and exploratory structural biology assessment of selected missense substitutions. The single-factor model showed acceptable but not optimal approximate fit and captured a biologically consistent genetic pattern, with positive loadings for fasting insulin, triglycerides, and waist-to-hip ratio and a negative loading for insulin sensitivity index. The latent factor showed detectable SNP heritability and revealed 3,067 genome-wide significant candidate SNPs, with Q_SNP retained as a complementary SNP-level heterogeneity diagnostic.. Functional annotation identified 100 genomic risk loci and 1,530 FDR-significant genes in MAGMA. Fine-mapping prioritized multiple candidate regions, including loci near PPARG, RSPO3, SLC38A11, SLC39A8, BCL2, GDF5, and SPATA5. TWAS and FOCUS further highlighted candidate genes involved in adipocyte biology, glucose–lipid metabolism, insulin secretion, and expression regulation, including PPARG, MLXIPL, TCF7L2, FAM13A, CCDC92, and SLC22A3. Enrichment analyses suggested that genetic signals associated with the insulin resistance-related latent factor were mainly linked to fat distribution, cardiometabolic traits, adipose tissue, mesenchymal cell lineages, and active regulatory genomic regions. In a separate exploratory analysis, AlphaFold3 modeling, ThermoMPNN prediction, and molecular dynamics simulations suggested structural perturbations for three ultra-rare missense substitutions in nominated candidate proteins: BCL2-W188G, GDF5-C400Y, and RSPO3-C129Y. These substitutions were selected using functional-impact annotations and modeling feasibility, were not GWAS lead or fine-mapped causal variants, and do not establish the mechanisms underlying the GWAS associations. These findings provide an integrated genetic and structural framework for understanding shared mechanisms underlying insulin resistance-related metabolic abnormalities, while emphasizing the need for further experimental validation. Citation: Li S, Liang Y, Zhu G, Wu L, Li H, Li Y, et al. (2026) Integrative genomic SEM analysis of shared genetic architecture and candidate genes for insulin resistance–related metabolic traits. PLoS One 21(10): e0361601. https://doi.org/10.1371/journal.pone.0361601 Editor: Hoh Boon-Peng, International Medical University, MALAYSIA Received: June 28, 2026; Accepted: October 12, 2026; Published: October 29, 2026 Copyright: © 2026 Li et al. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability: All study-derived data are available through the Science Data Bank repository: https://www.scidb.cn/anonymous/ZW1NQlZ6 . The shared files comprise Genomic SEM-based GWAS summary results for the insulin resistance-related latent genetic factor, together with AlphaFold3-predicted PDBx/mmCIF structures of the selected candidate proteins and their corresponding mutants, namely BCL2, GDF5, and RSPO3. These files were generated for downstream analyses including protein stability prediction and molecular dynamics simulation. The original GWAS datasets were publicly accessible, and the relevant source information is provided in S1 Table . No individual participant-level genetic, phenotypic, or clinical records were included in the repository. Funding: This study was supported by the National Natural Science Foundation of China [grant number 82574805]. The funder had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. 1. Introduction Insulin resistance (IR) refers to a reduced responsiveness of target tissues to the biological actions of insulin [ 1 , 2 ]. It is characterized primarily by impaired glucose uptake in skeletal muscle, increased hepatic glucose production, dysregulated metabolic control in adipose tissue, and compensatory hyperinsulinemia [ 2 , 3 ]. Its development and progression are closely related to abnormal fat distribution, lipid metabolic disturbance, chronic low-grade inflammation, oxidative stress, and impaired insulin signaling [ 4 , 5 ]. As a common pathological basis of multiple metabolic disorders, insulin resistance is strongly associated with type 2 diabetes, obesity, metabolic syndrome, nonalcoholic fatty liver disease, and an increased risk of cardiovascular disease, often involving multiple organs and systems [ 2 ]. With the increasing prevalence of obesity, population aging, and lifestyle changes, the disease burden attributable to insulin resistance-related metabolic abnormalities has continued to rise, making insulin resistance a major challenge for cardiometabolic health [ 6 ]. Although progress has been made in the clinical assessment, mechanistic investigation, and therapeutic management of insulin resistance, its genetic basis and the biological mechanisms shared across related phenotypes remain incompletely understood [ 7 , 8 ]. Previous studies have suggested that adipose tissue dysfunction, reduced insulin sensitivity in skeletal muscle, abnormal hepatic glucose and lipid metabolism, and impaired vascular endothelial function may contribute to the onset and progression of insulin resistance [ 1 , 9 ]. However, these mechanisms do not fully explain interindividual differences in insulin sensitivity, metabolic phenotypes, or susceptibility to related diseases. To address the limitation that a single clinical indicator cannot fully capture insulin resistance, this study applied Genomic Structural Equation Modeling (Genomic SEM) to integrate published GWAS summary statistics of insulin resistance-related phenotypes. Based on fasting insulin, the insulin sensitivity index, triglycerides, and waist-to-hip ratio, we constructed an insulin resistance-related latent factor to represent the shared metabolic genetic component underlying these phenotypes [ 7 , 10 – 12 ]. By modeling the shared genetic structure across different phenotypes, we further estimated the genetic associations between single-nucleotide polymorphisms (SNPs) and this latent factor. We combined genetic locus identification, candidate gene prioritization, tissue expression enrichment, and cell-type enrichment to characterize the shared genetic component. A separate computational structural analysis explored selected amino-acid substitutions in nominated candidate proteins. It should be noted that this latent variable should not be regarded as a direct diagnostic indicator of insulin resistance. Rather, it is more appropriately interpreted as an insulin resistance-related shared metabolic genetic phenotype. In addition, this study focused on missense variants in core candidate genes that may affect protein structure and function. By integrating AlphaFold-predicted structures, confidence assessment of mutation sites, thermodynamic stability prediction, and molecular dynamics simulations, we explored the effects of representative variants on protein conformation, local stability, and dynamic properties. Overall, this study aimed to establish an analytical framework linking genetic statistical discovery to structural biological interpretation, thereby providing a theoretical basis for subsequent functional validation and the mechanistic investigation of prioritized candidate genes. 2. Methods 2.1 GWAS data sources and quality control This study included four publicly available GWAS summary statistics closely related to insulin resistance as input phenotypes, including fasting insulin (FI), the insulin sensitivity index (ISI), triglycerides (TG), and waist-to-hip ratio (WHR). These phenotypes characterize insulin resistance-related metabolic status from the perspectives of compensatory insulin secretion, peripheral insulin sensitivity, lipid metabolic disturbance, and fat distribution, respectively. All input data were derived from published or publicly accessible GWAS summary statistics based on European-ancestry populations. The original studies had obtained the required ethical approval and informed consent from participants. Details on the sample sources, phenotype definitions, sample sizes, and download information for each dataset are provided in S1 Table . Before multivariable genetic modeling, all input GWAS datasets were uniformly formatted and subjected to quality control. Variants located within the major histocompatibility complex region (MHC; chromosome 6: 25–35 Mb, GRCh37) were first removed to reduce the influence of the complex linkage disequilibrium structure in this region on genetic covariance estimation. Using the European reference panel from the 1000 Genomes Project Phase 3 as the LD reference, autosomal SNPs were then standardized and filtered [ 13 ]. Variants with a minor allele frequency (MAF) below 0.01, missing or zero effect estimates, inconsistent variant information relative to the reference panel, or allele orientations that could not be reliably harmonized were excluded. The quality-controlled summary statistics were used for subsequent genetic covariance estimation, Genomic SEM modeling, and SNP-level association analyses. Ethics statement. This study analyzed publicly available GWAS summary statistics. Ethical approval and informed consent had been obtained in the source studies. No new individual-level data were collected or analyzed; therefore, additional ethical approval and consent for publication were not applicable. 2.2 Genomic structural equation modeling and latent factor GWAS To identify the shared genetic structure across multiple insulin resistance-related phenotypes, this study constructed a genomic structural equation model using the GenomicSEM R package (version 0.0.5). This approach estimates genetic covariance among multiple complex phenotypes from GWAS summary statistics and identifies cross-phenotype shared genetic susceptibility components within a latent variable framework. It can also account for the potential influence of sample overlap and differences in sample size on model estimates [ 14 ]. The Genomic SEM analysis was performed in two steps. First, multivariable LD score regression was used to estimate the empirical genetic covariance matrix and the corresponding sampling covariance matrix among FI, ISI, TG, and WHR. Within the same LDSC framework, SNP-based heritability, the number of retained SNPs, genomic inflation factor (λGC), mean chi-square statistic, LDSC intercept, and attenuation ratio were extracted for each input GWAS dataset to evaluate polygenic signal and potential inflation of association statistics. Second, a single-factor structural equation model was fitted based on the genetic covariance matrix to evaluate whether the four metabolic phenotypes could be explained by a common insulin resistance-related latent genetic factor. Model fit was interpreted using pragmatic benchmarks of CFI ≥ 0.90 and SRMR ≤ 0.08 for acceptable approximate fit, with CFI around 0.95 treated as a stricter benchmark for stronger fit; the chi-square test was reported alongside these indices. Attenuation ratios were treated as quality-control diagnostics and were not interpreted quantitatively when the LDSC intercept was below 1. After fitting the model, SNP-level Genomic SEM GWAS was performed to estimate the association between each autosomal SNP and the latent genetic factor. 2.3 SNP heterogeneity test and statistical thresholds Q_SNP was calculated as a complementary SNP-level diagnostic of departures from the effect pattern expected under the common-factor model. Non-significant Q_SNP values were not interpreted as formal evidence of effect homogeneity. Downstream analyses were conducted using the complete post-QC common-factor GWAS summary statistics, with Q_SNP retained to aid the interpretation of individual loci. For statistical significance, P < 5 × 10 −8 was used as the genome-wide association threshold. More stringent thresholds of P < 5 × 10 −12 and P < 5 × 10 −16 were used for descriptive stratification of association strength. 2.4 Risk locus definition, functional annotation, and novel locus identification Based on the Genomic SEM GWAS summary statistics, this study used the FUMA platform to define genomic risk loci and perform functional annotation [ 15 ]. Independent significant SNPs reaching genome-wide significance and showing low linkage disequilibrium with one another were first identified, and lead SNPs were then determined accordingly. SNP independence was defined using an r² threshold of < 0.1. Adjacent significant signals were subsequently merged into risk loci according to physical distance and local LD clustering. To determine whether the multivariable model identified genetic signals not captured by single-phenotype GWAS, lead SNPs detected by Genomic SEM were systematically compared with known association loci reaching the same significance threshold in the univariate GWAS of FI, ISI, TG, and WHR. A lead SNP was defined as a potentially novel locus if it was located more than 1 Mb away from any significant locus reported in the corresponding univariate GWAS. In addition, a “GWAS-minus-locus” comparison strategy was applied, in which Genomic SEM lead loci were sequentially compared against significant loci from the four input GWAS in an exclusion-based manner to identify independent candidate variants that emerged only in the shared genetic structure [ 16 ]. To evaluate the potential pleiotropic features of the lead loci, this study further queried the GWAS Catalog for genome-wide significant associations between lead SNPs and other published complex traits [ 17 ]. Based on the risk loci defined by FUMA, MAGMA was used to perform gene-level association analysis [ 18 ]. By integrating the genetic effects of multiple SNPs within gene regions, MAGMA estimated the overall association strength between candidate genes and the insulin resistance-related latent factor. Genes with an FDR < 0.05 were defined as candidate genes. Genes surviving Bonferroni correction were further defined as high-confidence genes and used as the input for the main gene-set enrichment analysis. 2.5 Fine-mapping analysis To narrow the candidate set of potential causal variants, this study performed statistical fine-mapping analyses for the major risk regions identified by Genomic SEM GWAS. For each lead SNP, local association variants within a ± 250-kb window were extracted. Using LD reference data from the European population in the 1000 Genomes Project Phase 3, two Bayesian fine-mapping methods, SuSiE and FINEMAP, were applied separately to estimate the posterior inclusion probability (PIP) within each SNP within the region [ 19 , 20 ]. Both methods were used to construct 95% credible sets, and the resulting candidate variants were cross-validated between methods. Variants supported by both SuSiE and FINEMAP and showing relatively high posterior inclusion probabilities were prioritized for locus-level interpretation. Candidate genes for structural exploration were nominated from prioritized loci using locus-to-gene annotation. The missense substitutions subsequently retrieved within these genes were selected separately and were neither GWAS lead variants nor variants prioritized as putatively causal by these fine-mapping analyses. The fine-mapping evidence at these loci therefore does not constitute variant-level genetic evidence for the modeled substitutions. 2.6 Transcriptome integration analysis and candidate gene prioritization To identify candidate genes that may contribute to insulin resistance-related metabolic abnormalities through gene expression regulation, this study performed a transcriptome-wide association study (TWAS) based on the Genomic SEM GWAS results. TWAS was conducted using FUSION, with GTEx v8 cross-tissue sCCA expression weight models used as the reference [ 21 – 23 ]. Multiple testing was controlled by Bonferroni correction across 37,917 gene–panel models, giving a significance threshold of 1.3187 × 10 −6 (0.05/37,917). By integrating GWAS association statistics with expression quantitative trait locus information, this analysis evaluated the association between genetically predicted gene expression levels and the insulin resistance-related latent genetic factor. For statistically significant TWAS signals, FOCUS was further applied for Bayesian gene-level fine-mapping to calculate the posterior inclusion probability (PIP) of candidate genes and construct credible gene sets [ 24 ]. Finally, high-confidence candidate functional genes were prioritized by integrating TWAS significance, FOCUS prioritization, FUMA functional mapping, and MAGMA gene-level association results. 2.7 Gene-set, disease ontology, and phenotype ontology enrichment To interpret the genetic association results at the pathway and disease annotation levels, this study performed gene-set and pathway enrichment analyses using MAGMA and FUMA. These analyses were used to evaluate whether the candidate genes showed aggregation in specific biological processes or functional pathways. MendelVar was further used to perform Disease Ontology and Human Phenotype Ontology enrichment analyses for candidate genes related to significant risk regions, in order to assess whether these shared genetic signals were preferentially enriched in disease and phenotype categories related to metabolic abnormalities, fat distribution, cardiovascular changes, inflammatory responses, or endocrine regulation [ 25 ]. 2.8 Cell-type-specific enrichment analysis To explore the cellular sources through which shared genetic signals for insulin resistance may act, this study performed cell-type-specific enrichment analysis using the CELLECT framework [ 26 ]. First, based on the Tabula Muris single-cell RNA sequencing reference dataset, CELLEX was used to calculate expression specificity scores for each gene across different cell types [ 27 ]. These scores were then integrated with the Genomic SEM GWAS summary statistics, and stratified LD score regression was applied to evaluate the degree of heritability enrichment of genetic variants in cell-type-specific expressed gene sets [ 28 ]. Multiple testing was controlled using FDR correction, with an FDR-adjusted P < 0.05 defined as significant cell-type enrichment. Nominally significant results that did not survive FDR correction were interpreted only as suggestive findings. 2.9 Partitioned heritability analysis Beyond tissue and cell-type annotations, this study further applied partitioned LD score regression to evaluate the contribution of different functional genomic annotations to the SNP heritability of the insulin resistance-related latent factor [ 29 ]. This analysis integrated functional annotation
## Related Clinical Research

- [Amylyx Presents Avexitide Results; MiniMed Launches Tubeless Insulin Pump](https://medichelpline.com/clinical-feed/endocrine-news-0-pharma-friday-november-21-2026.md)
- [Later bedtimes and late eating tied to higher metabolic syndrome risk](https://medichelpline.com/clinical-feed/medical-news-today-0-eating-and-sleeping-later-associated-with-increased-health-risks.md)
- [Preconception, Gestational, and Postpartum Care for Type 2 Diabetes](https://medichelpline.com/clinical-feed/endocrine-news-0-preconception-gestational-and-postpartum-care.md)
- [Resistance exercise preserves lean mass during tirzepatide therapy for obesity](https://medichelpline.com/clinical-feed/nature-1-resistance-exercise-to-attenuate-lean-mass-loss-during-tirzepatide-therapy-for.md)
- [Anticipatory Insulin Secretion Is Disrupted in Obesity and Hyperglycemia](https://medichelpline.com/clinical-feed/medrxiv-7-disruption-of-anticipatory-insulin-secretion-in-patients-with-hyperglycemia-and.md)

## Navigation
- [← Back to Endocrinology Feed](https://medichelpline.com/clinical-feed/endocrinology.md)
- [← All Clinical Specialties](https://medichelpline.com/clinical-feed.md)
## Medical & Regulatory Disclaimer

> [!CAUTION]
> MedicHelpline content is structured for research, educational, and professional discovery purposes. It does not constitute individual medical advice, clinical diagnosis, or treatment recommendations.
> Always verify dosing, contraindications, and regulatory alerts against official product labeling and primary regulatory sources before clinical decision-making.