Gestational diabetes mellitus (GDM) is glucose intolerance first recognized during pregnancy and carries long-term metabolic risks for both mother and offspring. Population-level genetic studies (GWAS) have uncovered GDM risk loci, but bulk-tissue analyses obscure cell-type-specific regulatory mechanisms. Single-cell expression quantitative trait loci (sc-eQTL) mapping now permits genotype–expression associations at cellular resolution, enabling refined gene prioritization for complex traits such as GDM.
The analysis combined FinnGen GDM GWAS summary statistics (12,332 cases and 131,109 female controls of Finnish ancestry) with sc-eQTL summary statistics derived from the OneK1K cohort, which profiled approximately 1.27 million peripheral blood mononuclear cells (PBMCs) from 982 healthy donors. Variant QC removed SNPs with missing effect alleles or standard errors and retained HapMap3 sites, yielding 1,266,324 high-quality SNPs for GWAS. OneK1K cis-eQTL mappings were converted from GRCh37 to GRCh38 and harmonized with GWAS alleles and directions, producing 1,009,861 shared SNPs for genetically regulated expression (GReX) modeling.
SNP-based heritability estimated by LD Score Regression (LDSC) for GDM was 0.0352 (SE = 0.0052, P < 0.001). Local heritability partitioning using LAVA identified 828 genomic regions with significant enrichment of GDM heritability after Bonferroni correction. The strongest regional signals included loci on chromosome 11 (chr11: 92.5–93.8 Mb; local h2 = 0.0236) and chromosome 2 (chr2: 168.4–169.3 Mb; local h2 = 0.00367). LAVA used permutation-based empirical P values and a Bayesian evidence framework appropriate for regional inference.
As a sensitivity analysis, the authors performed a transcriptome-wide association study (TWAS) using GTEx whole blood expression models via FUSION. Two genes showed nominal association with GDM (SMCO4: TWAS Z = 4.97, P = 6.82 × 10−7; RP11-443B20.1: TWAS Z = 4.14, P = 3.42 × 10−5) but neither survived FDR correction (q > 0.05). This contrast underscores the potential of single-cell approaches to reveal associations missed at bulk tissue level.
The OneK1K sc-eQTL resource encompassed 12 immune cell subtypes: CD4 ET, CD4 NC, CD4 SOX4, CD8 NC, CD8 S100B, CD8 ET, B Mem, B IN, Plasma, Mono C (classical monocytes), Mono NC (nonclassical monocytes), and dendritic cells (DC). Significant cis-sc-eQTL counts varied markedly by cell type. CD4 NC exhibited the highest number (795,929 significant eQTLs), followed by CD8 ET and CD8 NC. Plasma cells had the fewest (155,713). The variation correlated with cell abundance, indicating sample size is a major driver of eQTL discovery across cell types.
The study implemented the OTTERS framework for single-cell transcriptome-wide association studies (scTWAS). Stage I trained gene-level GReX imputation models independently for each of the 12 cell types. For each gene, four modeling approaches were used: lassosum, SDPR, PRS-CS, and a frequentist P+T baseline. Training used ancestry-matched linkage disequilibrium from 503 European-ancestry individuals (1000 Genomes Phase 3) and gene cis-regions defined as ±1 Mb around gene start and end based on GENCODE v41. Stage II applied the pre-trained weights to the FinnGen GDM GWAS summary statistics to impute GReX and compute method-specific TWAS Z-scores. The four per-gene P values were combined using the ACAT-O omnibus test to produce robust gene–trait association statistics. The authors released the eQTL weight files and models via Figshare to enable reproducibility and downstream use.
After multiple testing correction (FDR < 0.05), the scTWAS identified 63 significant gene–cell type associations corresponding to 14 unique genes. The distribution of significant associations varied by cell type: CD4 ET and B IN had the most (8 associations each), followed by CD4 NC and Mono C (7 each). CD8 ET showed no significant associations. The strongest aggregate signals were observed in CD4+ T cells and monocyte subsets.
Two aminopeptidase genes, ERAP1 and ERAP2, were associated across 10 of the 12 cell types (ACAT P = 2.17 × 10−5). RIOK2 also emerged as a shared regulator across multiple cell types. These multi-cell-type associations suggest loci with widespread regulatory effects in peripheral immune cells, though the authors note these may reflect complex linkage disequilibrium or tissue-specific mechanisms rather than single-gene causality.
Gene Ontology (GO) enrichment consistently highlighted antigen processing and presentation via MHC class I as the top enriched pathway (FDR < 10−5), implicating dysregulated antigen presentation in peripheral immune cells in GDM. Colocalization analyses prioritized monocytic LNPEP as a primary causal candidate at a locus of interest, whereas the ERAP1, ERAP2, and RIOK2 associations were interpreted as potentially influenced by local LD structure or distinct tissue-specific regulatory effects.
The single-cell TWAS approach uncovered immune cell–specific genes linked to GDM that were not detected in bulk blood TWAS. The repeated implication of antigen processing and MHC class I pathways supports immune-related mechanisms in GDM pathogenesis in peripheral blood immune cells. Prioritization of LNPEP in monocytes suggests a candidate causal gene for follow-up, while widely associated genes such as ERAP1 and ERAP2 may mark loci with complex regulation across cell types.
The analysis depends on sc-eQTL discovery power that is influenced by cell abundance; cell types with larger sample counts yielded more eQTLs. OTTERS requires at least one significant cis-eQTL per gene in the reference data, so genes lacking detectable genetic regulation were excluded. The study used publicly available, de-identified summary data; FinnGen and OneK1K data were collected under appropriate ethics approvals and informed consent. Data resources and the trained eQTL weight files are publicly available (FinnGen GWAS, OneK1K sc-eQTL, Figshare deposit) as reported by the authors.
Integrating FinnGen GDM GWAS with OneK1K sc-eQTLs via OTTERS and ACAT-O revealed 14 unique cell-type-specific genes associated with GDM and implicated antigen processing and MHC class I pathways in peripheral immune cells. Monocytic LNPEP was prioritized by colocalization as a primary causal candidate, while ERAP1, ERAP2, and RIOK2 showed broad multi–cell-type associations that may reflect complex LD or tissue-specific regulation. These results demonstrate that scTWAS can identify regulatory mechanisms missed by bulk analyses and provide candidate genes and pathways for mechanistic follow-up in GDM.