GWAS-by-subtraction reveals new genetic architecture and health implications of type 2 diabetes-independent gestational diabetes mellitus

Abstract

Background

The etiology of gestational diabetes mellitus (GDM) is constituted by both type 2 diabetes (T2D)-dependent and T2D-independent mechanisms. However, existing studies have evaluated the impact of T2D-dependent loci only for GDM, which limits the power to assess to what extent genetic variants or biological pathways are specific to GDM.

Methods

We subtracted the genetic effects of a genome-wide association study (GWAS) data for T2D from a GWAS data for GDM to reveal loci linked with T2D-independent components of GDM using genomic structural equation model (SEM). In the discovery stage of GWAS-by-subtraction, we used GWAS summary statistics of GDM and T2D from the FinnGen study as input latent variables (16,802 GDM cases and 237,816 controls; 71,728 T2D cases and 369,007 controls). In the replication stage, we adopted the summary statistics from a previously reported GWAS for GDM and T2D as input (21,263 GDM cases and 301,918 controls; 50,409 T2D cases and 523,897 controls). We functionally annotated variants with genome-wide significance, and performed comprehensive analyses including transcriptome-wide and proteome-wide associations, summary-data-based mendelian randomization, linkage disequilibrium score regression, and Mendelian randomization, to explore genetic patterns for T2D-dependent and -independent components of GDM.

Results

We found 69 independent genome-wide significant loci associated with the T2D-independent components of GDM for discovery stage, of which 49 SNPs are not significant in the GWAS for GDM used as input. T2D-independent components of GDM showed only genetic correlation with birth weight dependent on maternal genetic effect (rg = 0.19; P = 0.01), whilst T2D-dependent components of GDM showed genetic correlation with birth weight dependent on fetal genetic effect (rg = -0.22; P = 4.76 × 10^− 7). Loci of T2D-independent components of GDM effects map to genes related to islets of Langerhans, neural stem cells, blood levels of mannose, abundance of Streptococcus thermophilus and Bacteroides vulgatus, glucose and carbohydrate homeostasis, and the MAPK cascade and adenylate cyclase-activating G protein-coupled receptor signaling pathway.

Conclusions

We identified independent genetic components, loci and genes of GDM that may have been previously masked by T2D, providing more accurate targets for early detection and management of GDM.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13073-026-01674-2.

Background

Gestational diabetes mellitus (GDM) is a prevalent disorder with a significant rise across various populations over the past 15 years [1]. GDM shows strong genetic correlation with type 2 diabetes (T2D), and is related to an increased risk of T2D, with about a third of females developing T2D within 15 years of their GDM diagnosis [2]. While GDM and T2D in part share a polygenic predisposition, there is a second category of GDM genetic risk factors that are predominantly gestational contributors. The largest existing genome-wide association study (GWAS) of GDM revealed that the genetics of GDM falls into two categories: conventional T2D genetic etiology and etiology predominantly influenced by pregnancy [3]. Yet, existing studies only evaluated the impact of T2D loci, and lack of consistent and reliable measurements of T2D-independent components of GDM in existing genetic datasets pose challenge to control glycemic disturbed by natural perturbation of pregnancy.

Although recent GWAS for GDM identified 13 loci with GDM-predominant effects using heterogeneity test, 11 loci were previously well-established T2D related loci [3]. Therefore, GWAS of GDM cases and healthy controls may not be the most effective way to delineate the genetic architecture of the non-conventional T2D-independent components of GDM. GWAS-by-subtraction, a genomic structural equation model (SEM), can facilitate a GWAS of a latent trait, i.e. a sub-level trait not measured in any of the genotyped subjects [4], and has been used to comprehend non-intraocular pressure-dependent of primary open angle glaucoma and non-cognitive skills influenced educational success [4, 5]. Hence, we determined the genetic impacts of T2D-independent components, and considered these effects as the genetic variations in GDM not previously explained by T2D (Fig. 1a).

Fig. 1.

Overall study design. a Flowchart of the analysis process. b Cholesky model as fitted in Genomic SEM, with path estimates for a single SNP included as illustration. SNP, T2D and GDM are observed variables based on GWAS summary statistics. The genetic covariance between T2D and GDM is estimated based on GWAS summary statistics for T2D and GDM. The model is fitted to a 3 × 3 observed variance-covariance matrix (i.e. SNP, T2D and GDM). T2D dependent and T2D independent are latent (unobserved) variables. The covariances between T2D dependent and between T2D dependent and T2D independent are fixed to 0. The variance of the SNP is fixed to the value of 2pq (p = reference allele frequency, q = alternative allele frequency, based on 1000 Genomes phase 3). The residual variances of T2D and GDM are fixed to 0, so that all variance is explained by the latent factors. The variances of the latent factors are fixed to 1. T2D, type 2 diabetes; GDM, gestational diabetes mellitus

Methods

GWAS-by-subtraction

The objective of our GWAS-by-subtraction analysis, conducted by genomic SEM [4, 6], was to estimate, for each SNP, the association with GDM that was independent of that SNP’s association with T2D (hereafter, the T2D-independent SNP effect). The model regressed the GDM and T2D GWAS summary statistics on two latent variables, T2D-dependent and T2D-independent, which were then regressed on each SNP in the genome. This analysis allowed for two paths of association with GDM for each SNP. One path was fully mediated by T2D. The other path was independent of T2D and measured the T2D-independent SNP effect. The principle of our GWAS-by-subtraction is based on previous theoretical basis [4] (Fig. 1b). And recently cross-trait genetic architectures analysis for GDM and T2D suggest no evidence of a sex-specific classification for T2D [3]. Therefore, we used T2D GWAS including female and male in the GWAS-by-subtraction to increase more comprehensive SNP effect for T2D. Table 1 shows information for contributing GWAS summary statistics.

Table 1.

A list of Contributing Genome-Wide Association Studies

Phenotype	Ancestry	Used in	Soured	Cases	Control
GDM	European	Discovery GWAS for GWAS-by-subtraction	FinnGen study [7]	16,802	237,816
T2D	European	Discovery GWAS for GWAS-by-subtraction	FinnGen study [7]	71,728	369,007
GDM	European	Replication GWAS for GWAS-by-subtraction	Elliott A et al. [3]	21,263	301,918
T2D	European	Replication GWAS for GWAS-by-subtraction	Mahajan A et al.[8]	50,409	523,897
GDM	East Asian	Replication GWAS for GWAS-by-subtraction	Gu Y et al. [9]	12,024	67,845
T2D	East Asian	Replication GWAS for GWAS-by-subtraction	Spracklen CN et al. [10]	54,481	224,231

T2D Type 2 diabetes, GDM Gestational diabetes mellitus

In the discovery stage of GWAS-by-subtraction, we used GWAS summary statistics of GDM and T2D from the FinnGen study as input latent variables (16,802 GDM cases and 237,816 controls; 71,728 T2D cases and 369,007 controls) [7]. In brief, FinnGen individuals have been genotyped with Illumina and Affymetrix chip arrays [7]. Quality control was performed to remove samples and variants of poor quality. Imputation was performed using a population-specific SISu v3 imputation reference panel [7]. FinnGen GWAS was performed using regenie [7]. Sex, age, 10 principal components, and genotyping batch were included as covariates in the analysis. In the replication stage of GWAS-by-subtraction, we used GWAS summary statistics of GDM reported by Elliott A et al. (21,263 cases and 301,918 controls) and T2D reported by Mahajan et al. (50,409 cases and 523,897 controls) as input latent variables [3, 8]. GWAS GDM included individuals from FinnGen and Estonian Biobank. Genotyping and GWAS for FinnGen individuals are in line with the discovery stage of GWAS-by-subtraction. All Estonian Biobank participants have been genotyped at the core genotyping lab of the institute of genomics, university of Tartu, using illumina global screening array [3]. Samples were genotyped, and plink format files were created using illumina genomestudio [3]. Association analysis was carried out using scalable and accurate implementation of generalized mixed model, using sex, age, and ten principal components as covariates. GWAS T2D were generated using additive logistic regression under an additive genetic model, conducted separately within each of 32 contributing studies of European ancestry [8]. Statistical models were adjusted for body mass index, age, sex, and study-specific genetic principal components to account for population stratification. There was no sample overlap between the discovery and replication GWAS for T2D.

To further validated our findings are in Asian populations, we used Asian populations GWAS summary statistics of GDM reported by Gu Y et al. (12,024 cases and 67,845), and T2D reported by Spracklen CN et al. (54,481 cases and 224,231 controls) as input latent variables in GWAS-by-subtraction analysis [9, 10]. In brief, GWAS for GDM in Asian individuals were generated using low-depth non-invasive prenatal testing sequencing data. Genotypes were imputed with genotype likelihood-based method for phasing and imputation using sparse estimation using a 10,000-sample Chinese reference panel [9]. Primary association testing used plink2 logistic or linear regression under an additive model, adjusted for maternal age, BMI, gestational week, and top 10 principal components and regenie and bayesian optimal linear mixed model were used for validation. GWAS for T2D in Asian individuals using an additive genetic model within a logistic regression framework. Models adjusted for body mass index, age, sex, study-specific covariates, and principal components to account for population stratification.

GWAS-by-subtraction analysis was performed in three stages. At stage 1, we prepared two univariate input GWASs and used multivariate extension of cross-trait linkage disequilibrium (LD) score regression to generate the empirical genetic covariance matrix between the two traits as inputs for the GWAS-by-subtraction model. We restricted four univariate input summary statistics to the HapMap3 variants for estimating genetic covariance and the sampling covariance matrix in LD score regression with a minor allele frequency (MAF) > 0.01 to remove rare SNPs. At stage 2, we removed SNPs with a MAF < 0.01 (to avoid error due to fewer samples within the genotype cluster), SNPs with effect estimates exactly equal to zero, SNPs not matching the 1000 Genomes Phase 3 European or East Asian reference panel, and SNPs with mismatched alleles. Subsequently, we performed genomic SEM GWAS-by-subtraction analysis, in which SNPs were simultaneously regressed on the T2D-dependent and T2D-independent latent variables. The genetic correlation (rg) between the T2D-dependent and T2D-independent latent factors was constrained to zero. In sensitivity analyses, rg was set as 0.1 or 0.2 to explore how robust the results were to this parameter.

From the output GWAS summary statistics, SNPs with P < 5 × 10^{− 8} were considered as genome-wide significant. We used P < 1 × 10⁻³ as a nominal threshold for directional consistency between discovery and replication, not for genome-wide significance. The lead SNP in each region was identified by clumping the genome-wide significant SNPs using a threshold r² < 0.1 for pairs of SNPs located within 250 kb using functional mapping and annotation (FUMA) [11]. The cross-trait effects of the T2D dependent and T2D independent were further investigated in the GWAS catalog by retrieving all association results for SNPs in high LD (r² ≥ 0.8; European ancestry) with the leading SNPs and were genome-wide significant at P < 5.0 × 10^{− 8}.

Genetic correlation

SNP heritability (the phenotype variance explained by a specific set of GWAS variants) and genetic correlations were estimated by LD score regression [12, 13]. To explore genetic difference for glycemic traits with T2D-dependent and T2D-independent components of GDM, we collected the GWAS summary statistics from the Meta-Analyses of Glucose and Insulin-related traits Consortium, including 2-h glucose after an oral glucose challenge [14], fasting glucose [14], fasting insulin [14], glycated hemoglobin [14], fasting proinsulin [15], indices of β-cell function [16], insulin resistance [16], insulin fold change [17], insulin sensitivity index [17], proinsulin [18] and random glucose [19]. To explore genetic difference for pregnancy outcomes with T2D-dependent and T2D-independent components of GDM, we collected the GWAS summary statistics from the Early Growth Genetics Consortium, including birth length [20], preterm delivery [21], gestational duration [21], post-term delivery [21], head circumference [22], fetal birth weight [23], fetal effect birth weight [23], maternal birth weight [23], maternal effect birth weight [23], the last gestational weight [24], fetal effect placental weight [25], and maternal effect placental weight [25]. To explore the network of laboratory biomarkers to which T2D-independent components of GDM was genetically correlated, and which diverged from genetic correlations with T2D-dependent components of GDM, we collected the GWAS summary statistics from a previous study [26]. GWAS summary statistic for GDM, or the summary statistic for the T2D-dependent or T2D-independent components of GDM generated by GWAS-by-subtraction was treated as input summary statistics. Genetic correlations between the GDM traits and a series of traits were calculated.

Collecting, processing and statistical analysis of clinical profiles

We collected clinical profiles from the information system of West China Second Hospital, Sichuan University. We extracted information on mother and their newborns who were in their first pregnancy and had a live birth, excluded those with unclear age, height, weight, and diagnostic information, leading to 2,881 profiles (374 with pre-pregnancy T2D and 1081 with GDM). We found that the data did not satisfy normality or chi-square, so the Kruskal-Wallis test was chosen to compare the between-group variability, and Dunn’s tests were used to make multiple group comparisons.

Biological annotation

The goal of biological annotation of GWAS generated by discovery stage is to elucidate molecular mechanisms mediating genetic influences on the phenotype of interest. Our biological annotation analysis proceeded in three steps. First, we conducted enrichment analysis to test if some tissues and cell-types were more likely to mediate T2D-dependent and T2D-independent components of GDM heritability. Second, we conducted transcriptome wide association study (TWAS) [27], proteome wide association study (PWAS) [28] and summary-data-based mendelian randomization (SMR) [29] analysis to explore how the T2D-dependent and T2D-independent components of GDM genetics related to different genes transcription and proteins expression. Third, we conducted metabolome wide association study (MWAS) [30] and mendelian randomization (MR) [31] analysis to explore how the T2D-dependent and T2D-independent components of GDM related to blood metabolites and gut microbiota.

Enrichment of tissue and cell-type specific expression

To confirm the role of specific tissues or cells in mediating the impacts of T2D-dependent and T2D-independent components on GDM, we used previously defined gene-sets to test for the enrichment of genes specifically expressed in one of 53 genotype-tissue expression (GTEx) tissues or cell types and 152 Franke tissues or cell types captured by aggregation of RNA-seq studies [32]. The tissue-specific and cell-type specific analysis for T2D-dependent and T2D-independent components of GDM was conducted by LD score regression [13, 32].

Enrichment of gene and protein specific expression

To elaborate the role of specific genes and pathways in mediating the influences of T2D-dependent and T2D-independent components on GDM, firstly, we annotated the closet gene to the lead SNP using annotation of variants annotations, and mapped genes to functional pathways using the Metascape [11, 33]. To further expand our scope for gene discovery, we performed a TWAS and PWAS for T2D-dependent and T2D-independent components of GDM using Functional Summary-based Imputation (FUSION) [27]. The FUSION method involves three steps: (1) identify gene or protein expression features that are cis-heritable (i.e., variants associated with gene expression or protein within or near the genomic locus); (2) construct a linear predictor for each cis-heritable gene or protein (i.e., a SNP-based prediction weight of the gene or protein feature); and (3) calculate both TWAS test statistics incorporating these SNP-based prediction weights and summary-level GWAS Z scores. We used pre-computed blood gene and protein expression weights available, generated from the GTEx (version 8) and the atherosclerosis risk in communities study [27, 28], respectively, and performed bonferroni correction.

We then considered a more conservative approach, SMR, to prioritize specific genes and proteins that could be causally linked to T2D-dependent and T2D-independent components of GDM through leveraging summary-level SNP data of blood from multi-omics studies [29]. We used pre-computed expression quantitative trait loci (QTL) from and protein abundance QTL, generated from the PsychENCODE cohort and fenland cohort [34], respectively, and performed bonferroni correction based on the number of genes tested.

Enrichment of blood metabolites and gut microbiota specific abundance

To explain the role of metabolites abundance in mediating the effects of T2D-dependent and T2D-independent components on GDM, we performed a MWAS for T2D-dependent and T2D-independent components of GDM using FUSION [27, 30]. We used available pre-computed blood metabolites prediction weights, generated the TwinsUK dataset [31], and performed bonferroni correction. Meanwhile, given the compositions of gut microbiota are influenced by GDM, we used MR to explore the role of specific composition of gut microbiota in mediating the influences of T2D-dependent and T2D-independent components on GDM. We collected available GWAS for the composition of gut bacterial and gut bacterial pathway abundance [35]. We implemented MR analyses via TwoSampleMR (v.0.6.2) packages [36]. For MR analysis, we utilized the inverse-variance weighted (IVW) as main method to estimate the effect of T2D-dependent and T2D-independent components of GDM on gut microbiota abundance [37]. We used weighted median [38], weighted mode [39], and the MR-Egger methods [40] as sensitivity analysis.

Results

T2D-independent components for GDM

In the discovery stage, GWAS-by-subtraction identified 69 lead SNPs (P < 5 × 10^{− 8}) from 51 separate regions that are associated with the T2D-independent components of GDM (Table 2; Fig. 2a). The SNP heritability for the T2D-independent components of GDM estimated by LDSC is 5.35%, with no evidence of inflation of test statistics due to population structure (LDSC intercept = 1.02). Notably, 49 of 69 lead SNPs are not significant in the GWAS for GDM used as input for the GWAS-by-subtraction analysis (Table 2). This demonstrated that our approach is beyond simply separating known GDM loci into T2D-dependent and T2D-independent categories. In the replication stage, GWAS-by-subtraction analysis identified 42 independent significant SNPs (P < 5.0 × 10^{− 8}) (Fig. 2b, Additional file 1: Table S1). There are 20 lead SNPs with nominal significance (P < 1 × 10^{− 3}) for discovery and replication stage (Table 2). An additional 45 SNPs were selected as proxies for those not present for discovery and replication stage (Additional file 1: Table S2), supporting 65 out of 69 discovery signals. In addition, to validated our findings in non-European populations, we performed GWAS-subtraction analysis on GDM and T2D in Asian populations, and found that 6 of 20 lead SNPs, identified in European populations, are associated with the T2D-independent components of GDM (P < 1 × 10^{− 5}), including rs10830963, rs10830964, rs12270363, rs4526739, rs636696, rs780094, rs9285019 (Table 2). However, only rs12270363 and rs780094 exhibited effect sizes consistent with the direction observed in European populations. The reasons for these may be that there are fundamental differences in the pathophysiological mechanisms and disease phenotypes of T2D between East Asian and European populations [41, 42]. East Asian populations develop the disease at a lower BMI with more significant visceral fat accumulation, and β-cell dysfunction is the core driving factor for its onset, which is markedly distinct from the pathological characteristics of European and American populations. Given that genetic association analyses of GDM and T2D are highly dependent on the pathophysiological basis of the diseases, such ethnic differences in phenotypes and mechanisms render the relevant results of GWAS subtraction analysis non-transferable across racial groups.

Table 2.

Genome-wide significant lead variants for the non-T2D-dependent component of GDM

SNP	EA	NEA	Locus	CHR	POS	MAFEUR	BETAEUR——EUR	SEEUR	P EUR	Function	Nearest Gene	P GDM	P replication	BETAEAS	SEEAS	P EAS
rs10154014	C	T	50	20	48,356,146	0.139	2.143	0.168	4.03 × 10− 37	intergenic	B4GALT5	0.089	-	-	-	-
rs10830963	C	G	36	11	92,708,710	0.288	-0.692	0.028	6.76 × 10− 133	intron	MTNR1B	5.87 × 10− 216	1.11 × 10− 129	0.657	0.039	1.67 × 10− 63
rs10830964	C	T	36	11	92,719,681	0.147	0.31	0.037	5.35 × 10− 17	downstream	MTNR1B	5.38 × 10− 25	7.80 × 10− 12	-0.324	0.064	3.59 × 10− 7
rs10831057	A	C	36	11	93,267,564	0.197	-0.264	0.038	6.53 × 10− 12	intron	SMCO4	1.35 × 10− 18	-	0.006	0.051	9.09 × 10− 1
rs11020131	A	G	36	11	92,713,376	0.02	-0.535	0.06	5.73 × 10− 19	intron	MTNR1B	4.57 × 10− 28	1.58 × 10− 11	-	-	-
rs11190875	A	G	32	10	102,999,459	0.157	1.807	0.189	1.05 × 10− 21	intron	LBX1	0.034	-	-	-	-
rs112432204	C	T	36	11	92,708,092	0.027	-0.553	0.074	1.1 × 10− 13	intron	MTNR1B	1.42 × 10− 21	1.64 × 10− 16	-	-	-
rs112569024	A	T	25	7	157,944,883	0.153	1.487	0.233	1.84 × 10− 10	-	PTPRN2	0.324	-	-	-	-
rs113150876	A	G	15	4	160,202,839	0.183	1.798	0.162	9.28 × 10− 29	intron	RAPGEF2	0.061	-	-	-	-
rs113777162	G	T	20	6	38,690,353	0.28	1.997	0.11	5.1 × 10− 73	intron	D-H8	0.148	-	-	-	-
rs113977410	C	T	5	1	226,826,772	0.08	2.875	0.172	8.47 × 10− 63	intron	STUM	0.107	-	-	-	-
rs114417557	A	C	38	13	76,326,228	0.13	2.279	0.155	8.97 × 10− 49	intron	LMO7	0.300	-	-	-	-
rs11578147	A	G	2	1	95,243,022	0.193	1.85	0.155	6.95 × 10− 33	intron	SLC44A3	0.471	-	-0.400	0.768	6.02 × 10− 1
rs12270363	G	T	36	11	92,751,333	0.475	-0.174	0.028	8.51 × 10− 10	regulatory	MTNR1B	1.58 × 10− 14	2.51 × 10− 9	-0.219	0.043	3.75 × 10− 7
rs12549902	A	G	28	8	41,509,259	0.43	0.174	0.029	1.3 × 10− 9	upstream	NKX6-3	0.002	1.92 × 10− 6	-0.054	0.030	7.04 × 10− 2
rs12625547	G	T	51	20	50,154,647	0.186	-0.215	0.038	1.05 × 10− 8	intron	NFATC2	3.19 × 10− 6	6.37 × 10− 4	0.060	0.056	2.84 × 10− 1
rs12795282	C	T	33	11	25,443,086	0.219	1.64	0.161	2.72 × 10− 24	intergenic	ANO3	0.262	-	-	-	-
rs12804291	C	T	36	11	92,705,307	0.091	0.325	0.059	3.83 × 10− 8	intron	MTNR1B	2.73 × 10− 10	2.09 × 10− 8	-	-	-
rs13167145	G	T	17	5	95,711,140	0.123	0.194	0.03	1.24 × 10− 10	intron	PCSK1	7.15 × 10− 13	-	-	-	-
rs142242199	A	T	47	18	14,679,099	0.106	2.07	0.217	1.53 × 10− 21	intergenic	ANKRD30B	0.033	-	-	-	-
rs1447349	C	G	36	11	92,676,440	0.065	-0.322	0.05	1.71 × 10− 10	intergenic	MTNR1B	1.15 × 10− 18	1.19 × 10− 11	-	-	-
rs146228963	A	G	36	11	92,737,847	0.077	0.317	0.053	2.51 × 10− 9	intergenic	MTNR1B	8.56 × 10− 11	3.44 × 10− 7	-	-	-
rs147247861	C	G	29	8	103,743,138	0.479	2.043	0.084	1.49 × 10− 131	-	AP003356.1	0.407	-	-	-	-
rs148124765	A	T	40	15	84,657,647	0.101	1.701	0.279	1.08 × 10− 9	intron	GOLGA6L4	0.331	-	-	-	-
rs150205522	A	G	36	11	93,064,637	0.052	-0.391	0.061	1.63 × 10− 10	intron	DEUP1	4.38 × 10− 14	5.72 × 10− 7	-	-	-
rs150807747	A	G	19	5	177,267,016	0.48	1.654	0.102	4.09 × 10− 59	intron	FAM153A	0.624	-	-	-	-
rs154453	C	T	17	5	95,211,426	0.229	-0.193	0.034	1.29 × 10− 8	downstream	GLRX	1.04 × 10− 11	1.32 × 10− 9	-0.150	0.079	5.84 × 10− 2
rs17400671	A	G	46	18	8,659,445	0.072	1.912	0.349	4.46 × 10− 8	intergenic	MTCL1	0.225	-	-	-	-
rs183495841	A	G	45	17	77,969,351	0.377	1.298	0.152	1.43 × 10− 17	intron	TBC1D16	0.053	-	-	-	-
rs199746255	A	T	7	2	108,085,738	0.151	1.539	0.237	7.84 × 10− 11	intron	RGPD4	0.108	-	-	-	-
rs2089304	C	T	43	16	32,457,793	0.395	2.43	0.068	6.3 × 10− 276	-	PABPC1P13	0.023	-	-	-	-
rs2405819	A	T	26	8	2,747,064	0.323	1.478	0.14	5.9 × 10− 26	intergenic	MYOM2	0.312	-	-	-	-
rs2845871	C	T	35	11	92,032,930	0.317	0.185	0.032	4.99 × 10− 9	intergenic	FAT3	3.86 × 10− 9	4.46 × 10− 9	-0.017	0.033	6.12 × 10− 1
rs28691250	A	C	13	4	128,947,455	0.096	2.284	0.193	3.51 × 10− 32	intron	LARP1B	0.121	-	-	-	-
rs35153269	A	G	39	14	103,412,652	0.366	1.596	0.123	2.35 × 10− 38	intron	AMN	0.059	-	0.035	0.581	9.52 × 10− 1
rs36049512	C	T	49	20	11,602,755	0.185	2.954	0.082	2.94 × 10− 282	downstream	BTBD3	0.030	-	-	-	-
rs367810486	C	T	11	4	35,607,479	0.125	2.122	0.184	1.23 × 1030	-	RNU6-573P	0.323	-	-	-	-
rs371504276	C	T	10	3	129,184,764	0.052	2.114	0.384	3.57 × 1008	-	IFT122	0.038	-	-	-	-
rs375336641	A	G	12	4	105,244,753	0.142	2.997	0.104	2.42 × 10181	intron	CXXC4	0.086	-	-	-	-
rs376750538	C	G	42	16	29,284,832	0.355	1.005	0.179	2 × 10− 08	intron	NPIPB11	0.266	-	-0.278	0.425	5.13 × 10− 1
rs4285019	C	T	9	3	110,166,875	0.115	2.353	0.174	1.22 × 10− 41	intergenic	NECTIN3	0.181	-	-	-	-
rs4526739	C	T	36	11	92,665,020	0.358	-0.201	0.029	3.95 × 10− 12	intergenic	MTNR1B	8.27 × 10− 20	1.47 × 10− 13	-0.386	0.058	3.94 × 10− 11
rs4639974	G	T	37	12	48,851,067	0.215	1.345	0.201	2.41 × 10− 11	intergenic	ANP32D	0.413	-	-	-	-
rs4863928	A	C	14	4	135,997,446	0.267	2.273	0.097	4.31 × 10− 121	intron	PABPC4L	0.328	-	-	-	-
rs5020059	A	T	3	1	155,507,544	0.085	2.008	0.273	1.91 × 10− 13	intron	ASH1L	0.260	-	-	-	-
rs55726352	G	T	24	7	148,740,846	0.294	1.334	0.165	7.52 × 10− 16	intergenic	PDIA4	0.468	-	-	-	-
rs56043976	C	G	4	1	158,593,909	0.414	1.071	0.181	3.05 × 10− 09	intron	OR10Z1	0.161	-	-	-	-
rs58470262	C	T	8	2	152,143,842	0.233	1.558	0.164	2.3 × 10− 21	-	NMI	0.186	-	-	-	-
rs61870173	A	C	31	10	91,823,836	0.104	2.447	0.185	5.14 × 10− 40	intergenic	KIF20B	0.005	-	0.002	0.074	9.82 × 10− 1
rs61919163	A	G	36	11	93,071,705	0.067	-0.302	0.048	3.13 × 10− 10	intron	DEUP1	5.07 × 10− 14	2.78 × 10− 8	0.011	0.032	7.22 × 10− 1
rs62381642	C	G	18	5	170,799,941	0.133	2.21	0.168	1.11 × 10− 39	intergenic	NPM1	0.464	-	-	-	-
rs636696	G	T	34	11	48,435,748	0.412	0.545	0.086	2.88 × 10− 10	intergenic	OR4C5	0.404	-	0.226	0.089	1.13 × 10− 2
rs66619306	C	G	41	15	97,521,162	0.3	1.607	0.13	5.92 × 10− 35	intergenic	NR2F2	0.583	-	-	-	-
rs66715133	C	G	36	11	92,668,132	0.12	0.332	0.051	6.23 × 10− 11	upstream	MTNR1B	8.21 × 10− 14	5.36 × 10− 8	-0.032	0.092	7.31 × 10− 1
rs6698457	A	T	1	1	46,741,639	0.016	3.463	0.395	1.84 × 10− 18	intron	LRRC41	0.418	-	-	-	-
rs67907831	A	C	27	8	3,471,259	0.205	1.76	0.156	1.69 × 10− 29	intron	CSMD1	0.103	-	-	-	-
rs7047201	C	T	30	9	108,307,717	0.103	2.376	0.186	1.62 × 10− 37	-	FSD1L	0.454	-	-	-	-
rs71208329	C	T	8	2	151,915,828	0.036	3.079	0.268	1.6 × 10− 30	intergenic	RBM43	0.035	-	0.322	0.409	4.31 × 10− 1
rs72843707	A	C	44	17	59,901,157	0.194	2.518	0.103	5.13 × 10− 131	intron	BRIP1	0.115	-	-	-	-
rs74584198	C	T	22	6	57,309,137	0.179	2.016	0.143	2.12 × 10− 45	-	PRIM2	0.643	-	-	-	-
rs75211051	A	T	23	6	127,668,660	0.187	1.891	0.155	4.08 × 10− 34	-	ECHDC1	0.010	-	-	-	-
rs77055503	A	C	16	5	86,304,226	0.077	2.83	0.191	8.53 × 10− 50	intron	RASA1	0.363	-	-	-	-
rs77515437	A	C	21	6	47,662,280	0.154	2.515	0.115	1.17 × 10− 105	intron	ADGRF4	0.302	-	-	-	-
rs77695435	C	G	48	18	18,512,213	0.389	1.28	0.15	1.19 × 10− 17	-	ROCK1	0.189	-	-	-	-
rs780094	C	T	6	2	27,741,237	0.411	-0.186	0.03	4.43 × 10− 10	intron	C2orf16	8.26 × 10− 19	1.25 × 10− 9	-0.142	0.030	2.54 × 10− 6
rs78020513	C	T	35	11	92,089,311	0.034	-0.462	0.084	4.07 × 10− 08	intron	FAT3	4.54 × 10− 09	1.39 × 10− 7	-	-	-
rs78265553	A	G	36	11	92,732,391	0.031	-0.464	0.068	1.01 × 10− 11	intergenic	MTNR1B	1.10 × 10− 18	1.08 × 10− 12	-	-	-
rs80067804	C	T	43	16	32,526,341	0.486	1.617	0.112	4.98 × 10− 47	-	ABCD1P3	0.421	-	-	-	-
rs9285019	C	T	17	5	95,719,294	0.284	0.324	0.031	5.55 × 10− 26	intron	PCSK1	4.46 × 10− 32	1.24 × 10− 16	-0.136	0.029	1.91 × 10− 6

EA Effect allele, NEA Non-effect allele, CHR Chromosome, POS Position, MAF Minor allele frequency, BETA_EUR, the effect value for GWAS of T2D-independent GDM in European populations, SE_EUR, the standard error value for GWAS of T2D-independent GDM in European populations, P_EUR, the P value for GWAS of T2D-independent GDM in European populations, P_GDM, the P value for GWAS of original GDM in European populations, Preplication, the P value for GWAS of T2D-independent GDM at replication stage in European populations, BETA_EAS, the effect value for GWAS of T2D-independent GDM in East Asian populations, SE_EUR, the standard error value for GWAS of T2D-independent GDM in East Asian populations, P_EUR, the P value for GWAS of T2D-independent GDM in East Asian populations

Fig. 2.

Manhattan plot for the GWAS-by-subtraction derived T2D-independent component of GDM. a Manhattan plot for the GWAS-by-subtraction derived T2D-independent component of GDM in the discovery stage. b Manhattan plot for the GWAS-by-subtraction derived T2D-independent component of GDM in the replication stage. Plot of the -log10(P-value) associated with Wald’s test (two-sided) of the beta coefficient for the T2D-independent component of GDM for all SNPs, ordered by chromosome and base position. The threshold for genome-wide significance after considering of multiple tests is indicated by the grey dashed line (P = 5 × 10^− 8). Red dots represent lead SNPs

Sensitivity analyses in which the genetic correlation between the T2D-dependent and T2D-independent latent factors for GDM was less tightly constrained (rg = 0.1 or 0.2) during the GWAS-by-subtraction analysis led to progressively more SNPs being associated with the T2D-independent components of GDM at P < 5 × 10^− 8 (Additional file 1: Table S3-4; Additional file 2: Fig. S1-S2). In addition, there are 279 independent genome-wide significant SNPs from 139 regions associated with the T2D-dependent components of GDM (Additional file 1: Table S5; Additional file 2: Fig. S3). Only 15 of 279 lead SNPs are significant in the GWAS for GDM used as input for the GWAS-by-subtraction analysis. And there are 47 lead loci for GDM in the original study, in which 7 loci were significant in the GWAS for T2D-independent GDM; 13 loci were significant in the GWAS for T2D-dependent GDM. In total, there are 29 GDM loci which were not among their T2D-independent or -dependent loci. There is minimal evidence of inflation due to population stratification (LDSC intercept = 0.99). The QQ plots for results of GWAS-by-subtraction analysis showed at Additional file 2: Fig. S4. The correlation between the initial GWAS for GDM and the GWAS-by-subtraction results for the T2D-dependent components of GDM is high (rg = 0.76, P < 0.001). The SNP heritability of GDM was estimated to be 13.0%. Using GWAS-by-subtraction, the SNP heritability of the T2D-dependent component of GDM was 13.5%, and that of the T2D-independent component was 5.35%. The overall GDM heritability reflects the total additive genetic variance of the GDM phenotype, whereas the component heritabilities represent the genetic variances of two orthogonal latent factors constructed using GWAS-by-subtraction. They are not simple additive partitions of the total GDM heritability, and direct numerical comparison between the overall GDM heritability and the component heritability is not appropriate.

Association of T2D-dependent and -independent components of GDM with other traits

Seven significant SNPs from four regions for the T2D-independent components of GDM directly are associated to blood glucose or diabetes-related traits in previous GWAS analyses, and 18 significant SNPs from six regions demonstrate pleiotropic effects on other traits (Additional file 1: Table S6). For example, the lead SNP at the MTNR1B locus, rs10830956, is in strong LD with a known T2D-associated variant, rs7112766, suggesting that MTNR1B or a nearby gene may influence susceptibility to GDM through pleiotropic mechanisms that are both T2D dependent and independent. Many of the significant T2D-independent SNPs are associated to blood glucose-related traits, such as fasting blood glucose, glycated hemoglobin levels, and insulin levels. In contrast, many of the significant T2D-dependent SNPs are associated to diabetes-related traits, such as T2D and medication use in diabetes (Additional file 1: Table S7). In addition, many T2D-dependent SNPs are associated to cognitive traits and reproductive behaviors, while T2D-independent SNPs do not demonstrate such associations (Additional file 1: Table S6-7).

The genetic correlation between GDM and various glycemic traits is moderate-to-high: fasting glucose (rg = 0.47; P = 2.21 × 10^− 26), glycated hemoglobin (rg = 0.39; P = 1.02 × 10^− 11), insulin resistance (rg = 0.56; P = 2.46 × 10^− 8), fasting insulin (rg = 0.21; P = 2.68 × 10^− 5), random glucose (rg = 0.47; P = 8.71 × 10^− 30) and 2-h glucose after an oral glucose challenge (rg = 0.33; P = 3.52 × 10^− 4). In agreement with our previous results, genetic correlations between the T2D-dependent components of GDM and these glycemic traits are moderate-to-high: fasting glucose (rg = 0.42; P = 4.85 × 10^− 21), glycated hemoglobin (rg = 0.49; P = 6.07 × 10^− 44), insulin resistance (rg = 0.60; P = 6.32 × 10^− 14), fasting insulin (rg = 0.28; P = 1.44 × 10^− 11), random glucose (rg = 0.42; P = 5.03 × 10^− 23) and 2-h glucose after an oral glucose challenge (rg = 0.43; P = 2.32 × 10^− 15) (Fig. 3a; Additional file 1: Table S8). In contrast, the genetic correlations between the T2D-independent components of GDM and these glycemic traits are notably lower, at about half the magnitude observed for the T2D-dependent components: fasting glucose (rg = 0.23; P = 9.00 × 10^− 4), and random glucose (rg = 0.24; P = 1.86 × 10^− 4) (Fig. 3a; Additional file 1: Table S8).

Fig. 3.

Genetic correlation of the T2D-dependent component of GDM, the T2D-independent component of GDM, and GDM itself with other traits. a Genetic correlation of the T2D-dependent component of GDM, the T2D-independent component of GDM, and GDM itself with glycemic traits. b Genetic correlation of the T2D-dependent component of GDM, the T2D-independent component of GDM, and GDM itself with pregnancy outcomes. blue is GDM; red is T2D-independent component of GDM; green is T2D-dependent component of GDM. T2D, type 2 diabetes; GDM, gestational diabetes mellitus

We found distinct association patterns for the T2D-dependent and T2D-independent components of GDM and pregnancy outcomes (Fig. 3b). We first analyzed clinical profiles of 2,881 Chinese pregnant mother aged 19–48 years. Pregnant mother with GDM showed significant difference at preterm birth rate and offspring birth length compared to those with pre-pregnancy T2D (Additional file 1: Table S9). In genetic correlation analysis, GDM demonstrates a low, while nominal statistically significant, genetic correlation with birth length (rg = -0.16, P = 0.047), birth weight dependent on fetal genetic effect (rg = -0.22, P = 4.74 × 10^− 4), birth weight dependent on maternal genetic effect (rg = 0.14, P = 9.06 × 10^− 3), preterm delivery (rg = -0.23, P = 0.018), and placental weight (rg = 0.134, P = 0.029). T2D-dependent components of GDM demonstrate genetic correlation with birth length (rg = -0.15, P = 5.55 × 10^− 3), birth weight dependent on fetal genetic effect (rg = -0.22, P = 4.76 × 10^− 7), gestational duration (rg = 0.10, P = 6.26 × 10^− 3), and placental weight (rg = 0.14, P = 2.24 × 10^− 3), whilst T2D-independent components of GDM demonstrate genetic correlation with birth weight dependent on maternal genetic effect (rg = 0.19, P = 0.014) (Additional file 1: Table S10).

We observed significant genetic correlations for 13 blood laboratory values phenotypically related to GDM (Additional file 1: Table S11; Additional file 2: Fig. S5). Consistent with the results described above, genetic correlations between the T2D-dependent components of GDM and these blood laboratory values were significant (Additional file 1: Table S12; Additional file 2: Fig. S6). In contrast, we only observed genetic correlations between the T2D-independent components of GDM and urea (rg = -0.18, P = 2.05 × 10^− 4) and glucose (rg = 0.26, P = 7.33 × 10^− 5) (Additional file 1: Table S13; Additional file 2: Fig. S7).

T2D-dependent and -independent components of GDM in tissues and cells

We tested whether common variants in genes specifically expressed in 205 tissues and cell types are enriched in the effects on T2D-dependent and T2D-independent components of GDM (Fig. 4; Additional file 1: Table S14-16). GDM demonstrates nominal enrichment in the Langerhans, entorhinal cortex, parietal lobe, intestinal mucosa, brain frontal cortex (BA9), cervix ectocervix, brain and artery tibial. T2D-dependent components of GDM demonstrate nominal enrichment in the pancreas, liver, stomach, adipocytes, uterus, limbic system, upper gastrointestinal tract, rectum and colon sigmoid, whilst T2D-independent components of GDM demonstrate nominal enrichment in the islets of Langerhans, lung, parietal lobe, serous membrane, glucagon secreting cells, cervix ectocervix, germ cells, brain cerebellum, fibroblasts and neural stem cells.

Fig. 4.

Enrichment of tissue and cell-type specific expression for the T2D-dependent component of GDM, the T2D-independent component of GDM, and GDM itself. The threshold for genome-wide significance is indicated by the grey dashed line. blue is GDM; red is T2D-independent component of GDM; green is T2D-dependent component of GDM. T2D, type 2 diabetes; GDM, gestational diabetes mellitus

Genes and proteins of T2D-dependent and -independent components of GDM

There are 60 closest genes annotated to the lead SNPs for T2D-independent components of GDM and 218 closest genes annotated to the lead SNPs for T2D-dependent components of GDM, where 15 genes are overlapped genes for T2D-dependent components and T2D-independent components of GDM, including GCKR, MTNR1B and SMCO4 (Additional file 1: Table S17-18). Function pathways enriched for closest genes of T2D-independent components of GDM are not only in glucose and carbohydrate homeostasis, consistent with T2D-dependent components of GDM, but also in mitogen-activated protein kinase cascade and adenylate cyclase-activating G protein-coupled receptor signaling (Additional file 1: Table S19-20; Additional file 2: Fig. S8-9).

We tested whether genetic effects on gene and protein expression in blood, are enriched in the effects on T2D-dependent and T2D-independent components of GDM using TWAS and PWAS. At gene level, GDM demonstrates six genes (SMCO4, NRBP1, TSHZ2, KRTCAP3, PPM1G and HLA-DMA) through TWAS (Fig. 5a; Additional file 1: Table S). We found only two genes (ANK1 and HLA-DMA) through TWAS for T2D-independent components of GDM after multiple testing corrections, whilst we identified 54 genes by TWAS for T2D-dependent components of GDM after multiple testing corrections (Fig. 5b-c; Additional file 1: Table S22-23). Genetically predicted mRNA expression of ANK1 is a shared gene for the T2D-dependent and T2D-independent components of GDM, with opposite effects on T2D-dependent and T2D-independent components of GDM. At protein level, GDM demonstrates two proteins (PCSK1 and CTSF) through PWAS (Fig. 5d; Additional file 1: Table S24). We found only one protein, PCSK1, through PWAS for T2D-independent components of GDM after multiple testing corrections, whilst we identified ten proteins by PWAS for T2D-dependent components of GDM after multiple testing corrections, and there is no shared protein for T2D-dependent and T2D-independent components of GDM (Fig. 5e-f; Additional file 1: Table S25-26).

Fig. 5.

Manhattan plot for the TWAS and PWAS of the T2D-dependent component of GDM, the T2D-independent component of GDM, and GDM itself. a Manhattan plot for TWAS for GDM. b Manhattan plot for TWAS for T2D-independent component of GDM. c Manhattan plot for TWAS for T2D-dependent component of GDM. d Manhattan plot for PWAS for GDM. e Manhattan plot for PWAS for T2D-independent component of GDM. f Manhattan plot for PWAS for T2D-dependent component of GDM. Plot of the Z values associated with Wald’s test (two-sided) of the beta coefficient for all genes or proteins, ordered by chromosome and base position. The threshold for genome-wide significance after considering of multiple tests is indicated by the grey dashed line. Red dots represent significant genes or proteins

In SMR analysis, at gene level, there are two SMR significant genes for T2D-independent components of GDM after applying multiple-testing correction (ANK1 and MRPL33), and only ANK1 passed HEIDI (P > 0.01), while there are 61 SMR significant genes for T2D-dependent components of GDM after applying multiple-testing correction, and 29 gene passed HEIDI (P > 0.01; Additional file 1: Table S27-28). Consistent with our TWAS results, genetically predicted mRNA expression of ANK1 demonstrate the opposite effects on T2D-dependent and T2D-independent components of GDM. At protein level, there two SMR significant proteins for T2D-independent components of GDM (PCSK1 and GCKR), which supports the results for PWAS, while there are ten proteins identified for T2D-dependent components of GDM after multiple testing corrections, and GCKR is shared significant protein for T2D-dependent and T2D-independent components of GDM (Additional file 1: Table S29-30).

Metabolites of T2D-dependent and -independent components of GDM

We tested whether genetic effects on metabolites in blood are enriched in the effects on T2D-dependent and T2D-independent components of GDM using MWAS. GDM demonstrates two metabolites (X-22822 and mannose) through MWAS (Fig. 6; Additional file 1: Table S31), also significant for T2D-independent components of GDM after multiple testing corrections. There are nine metabolites significant for T2D-dependent components of GDM after multiple testing corrections, and X-22,822 and mannose are shared metabolites for T2D-dependent and T2D-independent components of GDM (Fig. 6; Additional file 1: Table S32-33).

Fig. 6.

Bar chart of MWAS for the T2D-dependent component of GDM, the T2D-independent component of GDM, and GDM itself. The threshold for significance is indicated by the grey dashed line. blue is GDM; red is T2D-independent component of GDM; green is T2D-dependent component of GDM. T2D, type 2 diabetes; GDM, gestational diabetes mellitus

Gut bacterial of T2D-dependent and -independent components of GDM

We tested if genetic effects on gut bacterial abundance are enriched in the effects on T2D-dependent and T2D-independent components of GDM using MR. GDM demonstrates nominal causal effect on the abundance of nine gut bacterial (Fig. 7a; Additional file 1: Table S34). T2D-independent components of GDM demonstrate nominal causal effect on the abundance of two gut bacterial (Streptococcus thermophilus and Bacteroides vulgatus). There are ten gut bacterial identified for T2D-dependent components of GDM with nominally significant, and there is no shared gut bacterial for T2D-dependent and T2D-independent components of GDM (Fig. 7b-c; Additional file 1: Table S35-36). For gut bacterial pathway, GDM demonstrates nominal causal effect on the abundance of two gut bacterial pathways. T2D-independent components of GDM demonstrate nominal causal effect on the abundance of five gut bacterial pathways. There are 49 gut bacterial pathways identified for T2D-dependent components of GDM with nominally significant, and pyrimidine deoxyribonucleotides de novo biosynthesis II is a shared gut bacterial pathway for T2D-dependent and T2D-independent components of GDM (Additional file 1: Table S37-39; Additional file 2: Fig. S10-12). In addition, the direction of causal analysis was supported by our sensitivity analysis.

Fig. 7.

Forest plot of MR for the T2D-dependent component of GDM (a), the T2D-independent component of GDM (b), and GDM itself (c) on gut microbiota abundance. blue is GDM; red is T2D-independent component of GDM; green is T2D-dependent component of GDM. T2D, type 2 diabetes; GDM, gestational diabetes mellitus

Discussion

In our study, we employed GWAS-by-subtraction to dissect the genetic risk of GDM into the T2D-independent components and the T2D-dependent components. Unlike single-trait GWAS, GWAS-by-subtraction incorporates multiple traits in a single model, accounting for intricate interrelationships, thereby increasing the probability of uncovering associations specific to underlying mechanisms. We hypothesized that both the T2D-independent and T2D-dependent biological processes contribute to the development of GDM by isolating the T2D-dependent effects on GDM, the residual genetic effects represent distinct components of GDM that is not influenced by T2D. GWAS summary statistics for the T2D-independent component revealed 69 independent SNPs located in 51 distinct genomic regions, which exhibit genome-wide significantly association with GDM through mechanisms presumably unrelated to T2D. The GWAS summary statistics for T2D-dependent components yielded 279 independent variants from 139 unique regions associated with GDM, likely through mechanisms linked to T2D.

A wide range of glycemic traits have been discovered related to GDM, including fasting glucose, hemoglobin A1c and 2-h glucose results on oral glucose tolerance testing [3]. In our study, T2D-dependent components of GDM display similar patterns of association with these glycemic traits, while T2D-independent components of GDM are notably less associated with these glycemic traits, which means glycemic changes in GDM may be more attributable to T2D from genetic perspectives. In addition, T2D-dependent and T2D-independent components of GDM display different patterns of association with pregnancy outcomes. For instance, while T2D-independent components of GDM are predominantly associated to birth weight dependent on maternal genetic effect, rather than with birth weight dependent on fetal genetic effect, the T2D-dependent components share a greater genetic architecture with preterm delivery and placental weight. These findings further supported the idea that GDM may influence offspring growth and development through maternal genetic effects. Shared blood biomarker signatures between GDM and T2D-dependent components further underscore their mechanistic overlap, highlighting maternal genetic effects as a driver of GDM-associated offspring development.

Four novel GDM-associated variants (rs71208329, rs28691250, rs11190875, rs114417557) demonstrate divergent effects: while reducing T2D-dependent GDM risk, they paradoxically increase T2D-independent GDM risk. The T2D-associated variants rs71208329, rs28691250 and rs11190875 would be expected a priori to be associated with a decreased risk of GDM via their effect on T2D. However, our GWAS-by-subtraction analysis revealed that this is not the case and, instead, suggested that these variants are associated with an increased risk of GDM through a T2D-independent mechanism. Our work suggests facilitating effects of these loci; a decreased risk of GDM via an association with lower T2D balanced by an increased risk of GDM via a T2D independent pathway. rs71208329 is located within the RBM43 gene, which is expressed selectively in white adipose depots that have low thermogenic potential, and is induced by inflammatory cytokines [43, 44]. Adipocyte-selective RBM43 disruption increases PGC-1α translation, resulting in mitochondrial biogenesis and adipose thermogenesis, thus improving glucose homeostasis in mice [45]. rs28691250 is located within the LARP1B gene, as part of the protein complex regulating the 5’-TOP mRNA translation, which have higher expression in the β-cells from diabetic patients due to the high metabolic demand during the development of diabetes [46–48]. Nearest gene for rs11190875 is LBX1, required for the development of GABAergic interneurons in the dorsal horn of the spinal cord and migration and further development of hypaxial muscle precursor cells for limb muscles, diaphragm and hypoglossal cord [49]. LBX1 functions as a negative regulator of energy metabolism and overexpression of LBX1 decreases glucose uptake [50]. Therefore, these evidences suggest that mutations in rs71208329, rs28691250 and rs11190875 may affect glucose homeostasis by affecting the expression of nearest genes, which in turn increases the risk of developing GDM.

Genetic analyses revealed that while T2D-dependent and T2D-independent GDM components are genetically uncorrelated, both traits demonstrate enrichment in pancreas-specific gene sets. However, T2D-independent GDM variants show preferential enrichment in brain-expressed genes, while T2D-dependent variants are enriched in gastrointestinal tract-expressed genes, suggesting distinct neural versus metabolic regulatory mechanisms. At gene level, ANK1 expression, component of the ankyrin-1 complex involving in the stability and shape of the erythrocyte membrane [51], exhibits opposing effects—negatively associated with T2D-independent GDM but positively associated with the T2D-dependent component-indicating pathway-specific regulatory roles. Variants in the ANK1 gene result in alterations in the structure of the encoded anchor protein, thereby reducing its plasticity and stability on the erythrocyte membrane, which accelerates the dissolution and destruction of erythrocytes [52]. Associations of ANK1 variants and T2D were identified by GWAS studies [53–55]. For example, GWAS for Chinese patients found increasing ANK1 expression in skeletal muscle contributes to T2D susceptibility through regulating energy and lipid homeostasis or skeletal muscle differentiation [54]. However, to date, the role of ANK1 variants in GDM has not been adequately investigated. Only one study found that ANK1 expression is the cores down-regulated genes in the placenta of mothers with GDM exchange, thus regulating the occurrence and development of GDM through maintaining cytoskeleton function [56]. ANK1 expression may play different roles in T2D and GDM through multi-channel and multi-link regulation.

Functionally, T2D-independent GDM genes are enriched not only in glucose homeostasis but also in MAPK cascade and adenylate cyclase-activating G protein-coupled receptor signaling pathways. Inactivating MAPK cascade ameliorates insulin resistance and placental injury in GDM [57], and G protein-coupled receptors (e.g. GPR1 activating protein kinase B phosphorylation) represent therapeutic targets for GDM [56–61]. Metabolically, elevated mannose levels are a shared risk factor for both GDM components. Elevating mannose level, a bioactive monosaccharide, may be a promising GDM biomarker and intervention target [62]. Notably, elevated mannose levels strongly indicate future risk for various chronic diseases, such as polycystic ovarian syndrome, cardiovascular disease, and albuminuria, potentially contributing to their development rather than serving solely as a novel biomarker [63–65]. Although mechanistic studies on mannose and GDM are currently scarce, circulating mannose levels are positively associated with obesity-independent insulin resistance due to mannose’s interference with insulin receptor function or its role in glycation end-product [66], thereby further increasing GDM risk. In addition, another potential pathway is that mannose could impact the gut microbiome [67]. Moreover, we further investigated gut bacterial abundance effects on T2D-dependent and T2D-independent components of GDM. T2D-independent components of GDM demonstrate nominal causal effect on the abundance of streptococcus thermophilus and bacteroides vulgatus, which is not identified for GDM. The mannose is a sugar available to streptococcus thermophilus which is transported and utilized by streptococcus thermophilus through a specific glucose/mannose phosphotransferase system [68], which suggests association between increased abundance of streptococcus thermophilus and higher risk of GDM may be attributed to the higher circulating mannose levels.

We acknowledge several limitations. Firstly, GDM GWAS encompasses a relatively limited number of cases, resulting in constrained statistical precision when estimating the correlations between SNPs and GDM risk. Secondly, inclusion of abnormal blood glucose test results from the pregnancy registry in GDM GWAS may lead to discrepancies in SNP heritability. While GWAS-by-subtraction is theoretically anticipated to be resilient to such variations in heritability, we cannot exclude this potential source of inaccuracy. Thirdly, we utilized GWAS summary statistics as input for GWAS-by-subtraction analysis, while it would be more ideal to investigate the causal variants for each trait, given the possibility of subtle differences in tagging causal variants across two GWAS samples. Additionally, GWAS-by-subtraction posits a simplified framework of causal relationships, encompassing only two latent risk factors for GDM: one entirely dependent on T2D and the other fully independent. These relationships are likely far more intricate with potential pleiotropic effects, like common risk factor (body mass index) for both GDM and T2D. Lastly, differences in LD structure and allele frequency between Finnish and other European populations may contribute to lower replication rates for some discovery signals. Future studies should use Finnish-specific LD references and larger Finnish replication data.

Conclusions

Our study unraveled the genetic risk of GDM into T2D-dependent and T2D-independent components, underlying further insight into GDM mechanism. We identified and explored loci and genes that may have been previously been masked by the effects of T2D, providing more accurate targets for GDM early detection and management.

Abbreviations

GDM

Gestational diabetes mellitus

T2D

Type 2 diabetes

GWAS

Genome-wide association study

SEM

Structural equation model

LD

Linkage disequilibrium

FUMA

Functional mapping and annotation

rg

Genetic correlation

MAF

Minor allele frequency

TWAS

Transcriptome wide association study

PWAS

Proteome wide association study

SMR

Summary-data-based mendelian randomization

MWAS

Metabolome wide association study

MR

Mendelian randomization

GTEx

Genotype-tissue expression

FUSION

Functional Summary-based Imputation

QTL

Quantitative trait loci

IVW

Inverse-variance weighted

Authors’ contributions

Zhou S: data collection, analyses and drafting the manuscript. Ran Z, Li Y, He J, Yuan X, Hu Y, Wang X, Li R, Xu Y and Yan C: analyses and reviewing and revising the manuscript. Xiong J and Cheng G: idea conceiving, funding, reviewing and revising the manuscript. All authors read and approved the final manuscript.

Funding

This work was supported by the National Natural Science Foundation of China (U25A20154) and the Department of Science and Technology of Sichuan Province (2025ZNSFSC0641). The funders played no role in the study design, collection, analysis, or interpretation of data, in the writing of the report, or in the decision to submit the paper for publication.

Data availability

All analyses were based on publicly available data. The GWAS summary statistics produced by GWAS-by-subtraction are available from Zenodo (10.5281/zenodo.19532939). GWAS for GDM and T2D are available from the FinnGen study (https://www.finngen.fi/en/access_results). GWAS for glycemic traits are available from MAGIC (https://magicinvestigators.org/downloads/). GWAS for pregnancy outcomes from the Early Growth Genetics Consortium (http://egg-consortium.org/). Previous GWAS for T2D are available from the DIAGRAM Consortium (https://diagram-consortium.org/downloads.html). GWAS for biomarkers are available form Sinnott-Armstrong et al. on FigShare (https://nih.figshare.com/articles/dataset/The_meta-analyzed_GWAS_summary_statistics_for_35_lab_biomarkers_described_in_Genetics_of_35_blood_and_urine_biomarkers_in_the_UK_Biobank_/12355382). GTEx weights for FUSION analyses are available at https://gusevlab.org/projects/fusion/. Plasma protein weights for FUSION analyses are available at http://nilanjanchatterjeelab.org/pwas/. Metabolites prediction models for FUSION analyses are available at https://github.com/Arthur1021/Metabolites-prediction-models. GWAS for gut bacterial abundance and pathway are available at NHGRI-EBI GWAS Catalog under the study accession numbers GCST90027446-GCST90027857.

Declarations

Ethics approval and consent to participate

Ethics approval was granted by the Medical Research Ethics Committee of West China Second Hospital Sichuan University (Protocol 2024–174). Analyses on these de-identified, secondary data were approved by the Resource Access Board of West China Second Hospital.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.