The MR’s protocol was based on three basic assumptions, as illustrated in Figure 1 . The association assumption required that the single nucleotide polymorphism (SNP) was closely related to the exposure. The independence assumption required that the SNP was not influenced by confounding variables. The exclusivity assumption required that the SNP only acted on the outcome through the exposure and not through other pathways.

The data for ASD was obtained from IEU open genome-wide association studies (GWAS) project (gwas.mrcieu.ac.uk/), and the data for T1DM, T2DM, and GDM were obtained from FinnGen (www.finngen.fi/fi), as shown in Table 1 . The data for ASD contained 46,351 Europeans and 933 closely related SNPs, dataset number: ieu-a-1185. The data for T1DM contained 282,809 Europeans and 32,038 closely related SNPs, dataset number: finngen_R8_T1D_STRICT1. Data for T2DM contained 365,950 Europeans and 16,143 closely related SNPs, dataset number: finngen_R9_T2D. Data for GDM contained 190,879 Europeans and 2,087 closely related SNPs, dataset number: finngen_R8_GEST_DIABETES. As the databases are open, ethical approval is no longer required.

Firstly, p < 5×10^-8 was restricted to search for SNPs strongly linked to T1DM and T2DM, while p < 5×10^-6 was restricted to search for SNPs strongly linked to ASD and GDM to fulfill the association assumption. Secondly, limiting kb = 10,000 and R² < 0.001 to exclude the interference of linkage disequilibrium. Thirdly, SNPs with strong correlation were searched for with a threshold of F > 10, where F = [R ²/(1-R ²)]*[(N-K-1)/k], R² representing the cumulative explained variance, N representing the sample size of the GWAS, and K representing the number of paired samples. Fourthly, SNPs with confounding variables were excluded by PhenoScanner and related literature to fulfill the independence assumption. Fifthly, the direction of exposure and outcome alleles were adjusted, and non-matching SNPs were excluded based on the effect allele frequency. Lastly, MR-Pleiotropy RESidual Sum and Outlier (MR-PRESSO) was used to eliminate outlier SNPs (p < 1) to ensure the correct causal inference.

The STROBE-MR was used as a guiding method (19). Two-sample bidirectional MR was used to assess the causal relationship of ASD with T1DM, T2DM, and GDM. The “TwoSampleMR (0.5.7)” package of R 4.3.1 was used to perform all operations of the MR analysis. IVW was set as the primary assessment tool, which achieved unbiased causal analysis in the absence of pleiotropy. Weighted median, which was sensitive to outliers, and MR-Egger, which analyzed data in the presence of pleiotropy, were set as secondary assessment tools. The MR-Egger’s intercept was also used to assess horizontal pleiotropy, which was required to satisfy the exclusivity assumption (p ≥ 0.05). Cochran’s Q and leave-one-out method were used to analyze heterogeneity and sensitivity, respectively. There was no heterogeneity in the results at p ≥ 0.05, and the results were robust when there were no significant changes in the combined effect sizes.

After association, independence and exclusivity tests, this study included 32 SNPs for ASD, 21 SNPs for T1DM, 148 SNPs for T2DM, and 38 SNPs for GDM ( Supplementary Table S1 ). Then, mismatched and outlier SNPs were excluded based on effect allele frequency and MR-PRESSO, respectively. Supplementary Table S2 illustrates the final included SNPs.

Two-sample bidirectional MR analysis assessed the causal effect between ASD and DM. Figure 2 presents the forest plot, while Figure 3 displays a scatter plot. The MR-Egger’s intercept assessed the horizontal pleiotropy of results, as shown in Supplementary Table S3 .

Cochran’s Q revealed heterogeneity only in the results of T1DM on the risk of ASD (p = 0.036), while the remaining results showed no significant heterogeneity (p ≥ 0.05), as depicted in Figure 4 and Supplementary Table S4 . The illustrated sensitivity analysis affirmed the robustness of all MR analysis results in Figure 5 .

As a growing public health problem, ASD increases the risk of suicide and self-harm in patients and brings a significant burden on families and society (20). Early detection and intervention are essential to prevent behavioral deficits and complications of ASD (21). Parental illness, gestational hypertension, and preterm delivery are considered major risk factors for ASD (22, 23), and controlling these risk factors helps to reduce the risk of ASD. As research has progressed, some researchers have found that maternal DM is associated with an increased risk of ASD in offspring (11, 12), and that ASD increases the risk of DM in the same individual (15). However, other studies have shown that DM has no significant effect on the incidence of ASD (16). The causal relationship between ASD and different types of DM is controversial. Given the controversial causal relationship between ASD and different types of DM, we used MR to evaluate the bidirectional relationship between them. This is the first MR study based on GWAS data that comprehensively analyzes the bidirectional causal relationship between three types of DM and ASD, ultimately revealing the risk cycling between T2DM and ASD.

This MR analysis showed that ASD and T2DM were associated with an increased risk of each other in European populations, whereas there was no causal relationship between ASD and T1DM or between ASD and GDM. Intercept analysis showed no horizontal pleiotropy for any of these outcomes. Cochran’s Q test showed no heterogeneity for any of the outcomes except for exposure (T1D)-outcome (ASD). Since the entire dataset was derived from Europeans, this study mainly explained the interaction of ASD with T1DM, T2DM, and GDM in Europeans.

Our study finds that T2DM is associated with an increased risk of ASD. A published clinical study supports this result, suggesting that T2DM is an independent risk factor for ASD in offspring (24). A cohort study that included 877,233 singleton infants born in Taiwan found that T2DM was associated with an increased risk of neurodevelopmental disorders (NDDs) such as ASD, attention deficit hyperactivity disorder (ADHD), developmental delays, and intellectual disabilities (25). A cohort study by the Swedish Psychiatric Association showed that pre-pregnancy T2DM was associated with an increased incidence of NDDs in offspring and that it increased the risk of ASD by 37% and ADHD by 43% (26). Yamamoto JM et al.’s (8) meta-analysis showed that T2DM exposure during pregnancy increased the risk of ASD in offspring by approximately 2-fold (OR 1.98, 95% CI 1.46 to 2.68). These pieces of evidence suggest that maternal T2DM is a potential risk factor for ASD in offspring, and a persistent hyperglycemic state may mediate this causal relationship (25). For one thing, hyperglycemia causes intrauterine hypoxia, resulting in impaired fetal neurodevelopment; for another thing, hyperglycemia induces free radical production and oxidative stress, which can affect fetal neurodevelopment (27). These suggest that hypoxia and oxidative stress mediated by hyperglycemia may be the mechanism by which T2DM increases the risk of ASD.

This MR analysis suggests that GDM is not associated with the risk of ASD, a result supported by several clinical studies. A US cohort study showed that the risk of ASD in offspring of patients with GDM diagnosed at 26 weeks of gestation and later was comparable to that of subjects without DM exposure (OR 0.99, 95% CI 0.88 to 1.12) (28). Another Spanish study that included 1,037 GDM pregnant women with singleton showed that GDM was associated with an increased risk of ADHD in offspring but not with the risk of ASD (29). Moussa HN et al.’s (30) systematic review pointed out that factors such as GDM, chronic hypertension, and autoimmune diseases were not associated with the progression of ASD in offspring. These pieces of evidence point to GDM not being a risk factor for ASD in offspring.

However, more studies support that GDM increases the risk of ASD in offspring. A US cohort study showed that pregnant women diagnosed with GDM at 26 weeks of gestation had a 63% increased incidence of ASD in their offspring (HR 1.63, 95% CI 1.35 to 1.97) (31). A meta-analysis that included 17 clinical studies showed that the offspring of patients with GDM had a 49% increased risk of developing ASD compared to non-diabetic patients (OR 1.49, 95% CI 1.18 to 1.89) (32). These studies support GDM as an independent risk factor for ASD in offspring, in contrast to this MR result.

Several reasons are used to explain the difference in results between MR analysis and clinical studies. First, the effect of GDM on ASD is related to the time of onset. Xiang AH et al.’s (28) cohort study showed that GDM diagnosed before 26 weeks of gestation resulted in a 30% additional risk of ASD, whereas GDM diagnosed after 26 weeks of gestation was not associated with the risk of ASD. In fact, the period before 24 weeks is the critical stage of fetal brain and spinal cord development (33). This means that the earlier the hyperglycemia occurs, the more lasting and significant the neurodevelopmental effects are (33). Unfortunately, this MR only analyzed the effect of the presence or absence of GDM on the risk of ASD and not the effect of GDM occurring at different gestational periods on the risk of ASD, as all the data in this study were derived from pooled GWAS data. Second, patients with GDM may include a subset of patients with impaired glucose tolerance or DM that was not diagnosed in a timely manner prior to pregnancy. In clinical trials, the additional risk of ASD in offspring of these patients was thought to be associated with GDM, but in fact, they were affected by T2DM rather than GDM. Third, considering that obesity is associated with lower insulin sensitivity and higher risk of GDM (34), the increased risk of ASD in offspring observed in clinical studies may be mediated by obesity rather than GDM. Kong L et al. (35) confirmed that the offspring of obese patients were at significantly increased risk of ASD, ADHD, and mixed disorder of conduct and emotion, but the risk of these disorders was not associated with GDM. In addition, three published MR studies support our speculation. Among them, the MR study by Ding H et al. (36) showed a correlation between obesity and increased genetic susceptibility to ASD, while the MR studies by Hu M et al. (37) and Song X et al. (38) supported a correlation between obesity and increased genetic susceptibility to GDM.

Literature on the causal relationship between T1DM and ASD is scarce, and they provide some controversial results. A Colorado case-control study found that the prevalence of ASD in the pediatric T1DM population was approximately 1.16%, which is comparable to the general population in Colorado (39). Another retrospective cohort study in Taiwan showed that T1DM exposure was associated with an increased risk of developmental delay, intellectual disability, and epilepsy in offspring but not with the risk of ASD (25). These pieces of evidence point to no significant effect of T1DM on either one’s own risk of ASD or the risk of ASD in offspring, consistent with the results of this MR analysis.

However, Chen MH et al. (40), in a Cox regression analysis of data from the Taiwan National Health Insurance Research Database, found a significant increase in ASD prevalence in patients with T1DM compared with controls. In a California retrospective cohort study, Xiang AH et al. (28) reported that T1DM was still associated with an increased risk of ASD in offspring after adjusting for smoking and pre-pregnancy body mass index (HR 2.33, 95% CI 1.29 to 4.21). Although these two studies support T1DM as a potential risk factor for ASD in offspring, they included different population than our MR analysis. The participants of the first study were Chinese belonging to Asian ethnicity, and the participants of the second study were Americans who included a multiracial population of European, Latino, African, and Asian ethnicity, whereas the subjects of our MR study were of European ethnicity. These racial disparities could potentially explain the variations in findings between the clinical study and our MR analysis. Additionally, Persson M et al. (23) found that maternal T1DM increased the risk of ASD by 40% in offspring in a Swedish population. According to our speculation, the effect of T1DM on the risk of ASD in offspring is mediated through preterm births. This is because T1DM leads to a higher risk of preterm births (28), and preterm births increase the risk of ASD in turn (41). The presence of an intermediate link, preterm births, may account for the difference between the MR analysis and clinical studies. In summary, considering the current lack of clinical evidence, more clinical studies are needed in the future to explore the potential role of T1DM in ASD.

In the reverse analysis, our MR shows that ASD is associated with an increased risk of T2DM, with no significant effect on the risk of T1DM and GDM. Although the effect of ASD on the risk of GDM has not been reported, current studies support the results of MR analysis of T1DM and T2DM. Zerbo O et al.’s (42) case-control study showed comparable rates of T1DM diagnosis in children with and without ASD (0.22% vs. 0.19%), supporting that ASD is not associated with the risk of T1DM. Chen MH and his colleagues (43) conducted a clinical study in Taiwan that showed both adolescents (HR 2.71, 95% CI 1.64 to 4.48) and young adults (HR 5.31, 95% CI 2.85 to 9.90) with ASD had a significantly increased risk of T2DM compared with age-matched controls. Even with short-term (HR 1.97, 95% CI 1.20 to 3.23) or long-term (HR 1.64, 95% CI 1.02 to 2.63) use of atypical antipsychotic medications, patients with ASD continued to have a higher risk of T2DM (43). Chieh AY et al.’s (15) systematic review still supports ASD as a potential risk factor for T2DM and metabolic disorders after excluding the interference of some confounding variables. These show that ASD is a potential risk factor for T2DM rather than T1DM.

In conclusion, ASD and T2DM are risk factors for each other, that T2DM increases the risk of ASD in offspring, and that ASD in turn increases its own risk of T2DM, that they may create an intergenerational risk cycling in female patients. Aggressive prevention and treatment of T2DM and ASD help to break the trap of this risk cycling.

To further elucidate the genetic mechanisms underlying the interaction between ASD and T2DM, we conducted a search in public databases for 20 SNPs for ASD acting on T2DM and 113 SNPs for T2DM acting on ASD. Among them, 13 SNPs have relevant literature reports. The rs3887925, rs11558471, rs6931514, rs34872471, rs6878122, rs7018475, rs348330, rs4862423, and rs878521 have been reported to be related to glucose metabolism or diabetes, while the rest of SNPs have not been reported to be related to T2DM. All SNPs have not been reported to be related to ASD. As shown in Supplementary Table S5 .

While this MR analysis enhances the genetic evidence, there are inevitably some limitations. First, as the data were derived from Europeans, our findings may not extrapolate to elucidate the relationship of ASD and DM in other racial groups. Second, since GWAS does not provide specific information on when GDM was diagnosed, this MR analysis can only explain changes in the risk of ASD at the time of GDM exposure and cannot analyze the effect of GDM occurring at different gestational periods on the risk of ASD. Third, obesity may be intrinsic to the increased risk of GDM in oneself and the risk of ASD in offspring, but no sufficient clinical evidence has yet supported this inference. Fourth, the effect of T1DM in increasing the risk of ASD may be mediated through preterm birth, but no high-quality clinical study has yet analyzed this scenario. Fifth, there may exist unrecognized confounding variables between exposure factors and outcome indicators, which increase the risk of bias in MR results.

Acknowledging these limitations, we anticipate that future researchers will continue to enrich the GWAS database, aiming for racially diverse MR research and promoting health equity. Second, stratified clinical studies should be conducted to explore the effect of GDM at different gestational periods on the risk of ASD in offspring. Third, researchers are expected to use large sample and multicenter clinical studies to analyze the role of obesity on the risk of GDM and ASD, as well as the role of T1DM on the risk of preterm birth and preterm birth on the risk of ASD, to validate the speculations of the study further.