2023 Volume 70 Issue 12 Pages 1159-1168
Our objective was to construct a polygenic risk score (PRS) and assess its utility and effectiveness in predicting the risk of gestational diabetes mellitus (GDM) in a Chinese population. We performed a case-control study involving 638 patients with GDM and 1,062 healthy controls. Genotyping was conducted utilizing a genome-wide association study (GWAS), and a PRS was constructed. We identified 12 susceptibility loci that exhibited significant associations with the risk of GDM at a p-value threshold of ≤5.0 × 10–8, of which four loci were newly discovered. A higher PRS was associated with an increased risk of GDM (OR: 1.44; 95% CI: 1.03, 2.01 for the highest quartile compared to the lowest quartile). The PRS demonstrated a clear linear relationship with the fasting plasma glucose (FPG), 1-hour postprandial glucose (1hPG), and 2-hour postprandial glucose (2hPG) levels. The maximally adjusted β coefficients and their corresponding 95% CIs were 0.181 (0.041, 0.320) for FPG, 0.225 (0.103, 0.346) for 1hPG, and 0.172 (0.036, 0.307) for 2hPG. Among the genetic variants examined, TCF7L2 rs7903146 displayed the strongest association with GDM risk (logOR = 0.18, p = 2.37 × 10–19), followed by ADAMTSL1 rs10963767 (logOR = 0.14, p = 3.58 × 10–15). The areas under the curve (AUCs) was significantly increased from 0.703 (0.678, 0.728) in the traditional risk factor model to 0.765 (0.741, 0.788) by including PRS. These findings indicate that pregnant women with a higher PRS could potentially derive considerable advantages from the implementation of a feasible PRS-based GDM screening program aimed at delivering precision prevention strategies within Chinese populations.
GESTATIONAL DIABETES MELLITUS (GDM) is a metabolic disorder characterized by the onset of hyperglycemia during pregnancy in women who do not have preexisting diabetes. Typically, GDM is diagnosed between the 24th and 28th weeks of gestation, and it manifests as elevated blood glucose levels [1]. The global incidence of hyperglycemia during pregnancy is estimated to be around 15.8%, with GDM accounting for more than 80% of these cases [2]. In China, the prevalence of GDM has substantially risen from 4% in 2010 to 21% in 2020 [3]. Effective risk prediction plays a pivotal role in GDM prevention, and the continual refinement of prediction strategies remains crucial for targeted treatment recommendations.
Like type 2 diabetes mellitus (T2DM), GDM exhibits close associations with genetic and environmental factors, with each factor having the individual potential to elevate the risk of GDM onset [4, 5]. The emergence of genome-wide association studies (GWAS) has held great promise in revolutionizing personalized medicine through the provision of individualized risk predictions, prevention strategies, and treatment options [6, 7]. The impact of a single or a limited number of single nucleotide polymorphisms (SNPs) is insufficient to accurately predict disease risk. Additionally, studies have unveiled a genetic architecture of greater intricacy than initially foreseen for most prevalent diseases—a complexity that constrains the immediate application of these findings [6]. Hence, it is imperative to ascertain the cumulative effect of multiple SNPs. To address this and to better comprehend the intricate interplay between genetics and diabetes, including GDM, the application of a polygenic risk score (PRS) has been proposed [8]. The use of a PRS represents a valuable genetic analysis strategy that combines information from multiple genetic loci to predict an individual’s susceptibility to developing a disease (or a specific clinical trait) and to evaluate the predictive capacity of associated genetic variations for the disease [9].
Extensive research has consistently shown that the utilization of a PRS enhances the accuracy of predicting the risk of T2DM [10, 11], but there is a paucity of studies investigating the applicability of PRS in predicting GDM [9, 12]. The pathogenesis of GDM is notably more intricate, as it involves not only the mother’s susceptibility to diabetes but also the secretion of various hormones from the placenta, which play roles in maintaining blood glucose levels [13]. Consequently, the specific SNPs implicated in the two conditions and the resulting PRS constructions are likely to differ significantly, leading to a more diverse PRS for GDM.
Several epidemiological studies have explored the development of a PRS in the context of GDM. However, there is significant variability in the number and selection of SNPs used in these studies. The selection of SNPs is primarily based on previous reports, with an emphasis on genetic risk loci identified through GWAS, ensuring alignment with the characteristics of the specific population under investigation. Some studies have highlighted the substantial utility of PRS in GDM prediction [8, 14, 15], while others have indicated that PRS offers limited value in identifying GDM cases [16]. Furthermore, there is a dearth of research specifically conducted among Chinese women, whose genetic background differs significantly from that of other populations, emphasizing the need for further investigation.
Therefore, we developed a GWAS-based PRS by carefully selecting genetic markers associated with GDM. We evaluated the predictive capacity of these scores through a case-control study. We then examined the predictive performance of genetic information alone and in combination with established GDM risk factors.
In this study, pregnant women with available medical records who underwent regular prenatal examinations were recruited from the Department of Obstetrics at the First People’s Hospital of Shuangliu District, Chengdu, between March 2020 and December 2021. The inclusion criteria were: (1) age ≥18 years; (2) gestational age of 8–12 weeks at the time of enrollment to the study; and (3) having a 75 g oral glucose tolerance test (OGTT) conducted at 24–28 weeks of gestation. The exclusion criteria were: (1) multiple pregnancies; (2) induction of labor for therapeutic reasons related to fetal malformation, stillbirth, or other conditions; (3) presence of a malignant tumor, essential hypertension, type 1 diabetes mellitus or T2DM, etc.; and (4) incomplete information in the medical records. All participants underwent a 75 g OGTT between 24 and 28 weeks of gestation. The diagnosis of GDM was based on the criteria established by the International Association of Diabetes and Pregnancy Study Groups, which suggests a diagnosis of GDM when any of the plasma glucose values in the OGTT are greater than or equal to the following: fasting plasma glucose (FPG) of 92 mg/dL (5.1 mmol/L), 1-hour plasma glucose (1hPG) of 180 mg/dL (10.0 mmol/L), or 2-hour plasma glucose (2hPG) of 153 mg/dL (8.5 mmol/L).
This investigation was carried out in accordance with the principles outlined in the Declaration of Helsinki by the World Medical Association, and the study received ethical approval from the Ethics Committee of The First People’s Hospital of Shuangliu District (2022-KS-03). Prior to participation, written informed consent was obtained from all enrolled individuals.
Data collectionExpert nurses received specialized training to administer the “gestational diabetes risk questionnaire” to pregnant women who visited our hospital for the first time. Each questionnaire was meticulously administered, with individual questions being posed and recorded to ensure the authenticity and reliability of the data. The questionnaire was used to gather comprehensive information about the mothers, which included sociodemographic characteristics (age, marital status, education, and family history of diabetes), pregnancy history (number of pregnancies, births, abortions, spontaneous abortions, induced abortions, and history of polycystic ovary syndrome [PCOS]), and fundamental details of the current pregnancy (pre-pregnancy weight, gestational age at delivery, gestational weight gain, self-reported hypertensive disorders of pregnancy, treatment of GDM, and lifestyle factors like smoking and alcohol consumption). Additionally, measurements of height, weight, and blood pressure were taken. In cases where data were missing, a telephone follow-up was conducted to ensure the collection of complete data.
Peripheral blood samples were collected in EDTA tubes after the interview. Biochemical markers such as blood lipid, FPG, postprandial blood glucose, blood urea nitrogen, creatinine, and liver and kidney function markers were measured and encoded into an electronic medical record system (Laboratory Information System).
Sequencing and genotypingFor each participant, 5 mL of peripheral blood was obtained. Genomic DNA (gDNA) was extracted utilizing a Genomic DP319 DNA Mini Kit (DP319, TIANGEN, Beijing, China) following the manufacturer’s recommended protocol. Roughly 2 μg of gDNA was prepared to construct the DNA libraries. The quality and quantity of the gDNA were assessed using agarose gel electrophoresis and the Nanodrop-2000 method (Thermo Scientific). The concentration of gDNA was adjusted to 50 ng/μL, after which all samples underwent whole genome amplification, followed by incubation at –37°C for a period of 20–24 hours. Next, the amplified DNA was fragmented, precipitated, and resuspended in a hybridization buffer. The resuspended DNA fragments were then added to the chip and subjected to hybridization, with incubation at –48°C for 16–24 hours. Post-hybridization, the non-specifically bound DNA was eliminated through washing procedures, and a single-base extension was performed on the remaining specific binding sites. Finally, the stained DNA fragments were scanned using the Illumina Iscan Reader.
The raw sequence data (FASTQ file reads) underwent quality control assessment using FastQC (v.0.11.7). The clean paired-end reads were then aligned to the human reference genome (GRCh37/hg19) utilizing Burrow-Wheeler Aligner (v.0.7.15). The variants, comprising SNPs and insertion/deletion (indel) polymorphisms, were identified using the Genome Analysis Toolkit (GATK). The identified SNPs/indels were then annotated against the National Center for Biotechnology Information dbSNP database.
The samples were subjected to the application of the following exclusion criteria: (1) average sequencing depth <10×; (2) coverage of at least 10× <90%; (3) identification of outliers in GC content; (4) assessment of relative duplication; (5) determination of an absolute inbreeding coefficient exceeding 1; (6) identification of outliers through principal component analysis; and (7) detection of any instances of sex mismatch.
The quality control of variants was implemented using the recommended GATK filters, which included variant quality score recalibration, assessment of the largest contiguous homopolymer run of the variant allele (HomopolymerRun), a binomial test (GetHetCoverage), the root mean square of mapping quality (RMSMappingQuality), and evaluation of strand bias (FisherStrand). Furthermore, to mitigate bias, the following exclusion criteria were applied: (1) minor allele average depth <4×; (2) average depth in either the case or control group <8; (3) eightfold rate for the case or control group <0.9; (4) and a p-value for the Hardy–Weinberg equilibrium test <10–4. Additionally, variants lacking dbSNP IDs (also with a minor allele frequency <0.005) were excluded.
Construction of the PRSWe constructed a PRS according to previously utilized methods [17]. The estimation of individual genetic scores was conducted using a weighted genetic risk score. This approach accounts for the individual effects of genetic loci when combining various risk factors. The genetic score for each variable was obtained by assigning weights to the odds ratio (OR) values calculated through logistic regression, and these scores were incorporated into the model. A participant’s PRS was generated by summing the number of risk alleles (i.e., 0, 1, or 2) at each SNP.
Statistical analysisThe study findings were reported as the mean ± standard error (SE), and the normality of the data was assessed using the Kolmogorov–Smirnov test. Group differences were analyzed using either Student’s t-test or the Mann–Whitney U test. For comparisons involving more than two independent groups, either one-way ANOVA or the Kruskal–Wallis test was used. Within-group comparisons were conducted using either repeated measures ANOVA or Friedman’s test. In instances where the assumption of sphericity was violated in repeated measures ANOVA, the Greenhouse–Geisser adjustment was applied.
The per-allele ORs and corresponding SEs were calculated using logistic regression assuming an additive genetic model with PLINK software. The threshold for genome-wide significance was set at p < 5.0 × 10–8 [18]. Categories of the PRS were defined based on quartiles in the control group. Logistic regression models were used to estimate the odds ratios and 95% confidence intervals (95% CI) for the risk of GDM across the PRS quartiles. Two models were applied. In model 1, age was adjusted for; in model 2, further adjustments were included for body mass index (BMI) before pregnancy, age of menarche, education level, family history of diabetes, number of births and spontaneous abortions, and history of PCOS. All of the covariates were introduced using the forward stepwise method. To test for linear trends, the median values of the quartiles were used and treated as continuous variables. Linear regression was performed to examine the associations between PRS and plasma glucose levels (i.e., FPG, 1hPG, and 2hPG). The area under the receiver operating curves (AUC) for the receiver operating characteristic (ROC) curve was used to estimate the prediction performance of PRS on GDM, with an AUC of 1 indicating perfect prediction whereas an AUC of 0.5 suggesting poor performance. Sensitivity analyses were conducted by excluding participants with a family history of diabetes from the described analyses.
All statistical analyses were conducted using STATA version 11.0 software (Stata Corp., College Station, TX, USA). Statistical significance was defined as a two-sided p-value less than 0.05.
The flowchart of participant recruitment in this study is presented in Fig. 1. A total of 2,134 gravid individuals in their third trimester were considered, of whom 434 were subsequently excluded due to incomplete data on pre-pregnancy BMI or 75 g OGTT results. Additionally, participants were excluded for having preexisting medical conditions such as pregestational diabetes, hypertension, preeclampsia, thyroid gland disorder, chronic renal disease, and collagen disorder, or for carrying multiple fetuses or a fetus with chromosomal abnormalities. Ultimately, the study included 1,700 expectant women, who were subsequently classified into two groups based on their glucose tolerance status: the normal glucose tolerance (NGT) group (n = 1,062) and the GDM group (n = 638).
Flowchart of participant recruitment in the case-control study.
The demographic and other characteristics are presented in Table 1. Compared with the controls, the GDM group exhibited a higher likelihood of being older, having a higher BMI before pregnancy or at antepartum, a longer gestational age at delivery, and having a family history of diabetes. While these differences between the two groups were observed, they were appropriately adjusted for in the regression models. The distribution of number of different risk loci between two groups was shown in Fig. 2.
Characteristics among women with GDM and controls
| Cases (n = 638) | Controls (n = 1,062) | p value | |
|---|---|---|---|
| Age, years | 29.75 ± 4.19 | 28.17 ± 3.98 | <0.001 |
| BMI before pregnancy, kg/m2 | 22.47 ± 3.25 | 21.4 ± 2.98 | <0.001 |
| BMI at ante partum, kg/m2 | 27.12 ± 3.38 | 26.81 ± 3.27 | |
| Age of menarche, years | 13.05 ± 1.16 | 13.07 ± 1.16 | 0.774 |
| Gestational age at delivery, weeks | 38.37 ± 1.30 | 38.62 ± 1.39 | <0.001 |
| Education level, n (%) | 0.674 | ||
| Junior high school or below | 120 (18.87) | 192 (18.10) | |
| High school | 313 (49.21) | 553 (52.12) | |
| Junior college | 187 (29.40) | 294 (27.71) | |
| Master degree or above | 16 (2.52) | 22 (2.07) | |
| Family history of diabetes, n (%) | <0.001 | ||
| No | 590 (92.77) | 1,046 (98.68) | |
| Yes | 46 (7.23) | 14 (1.32) | |
| Number of births | 0.051 | ||
| 0 | 358 (56.11) | 659 (62.05) | |
| 1 | 253 (39.66) | 361 (33.99) | |
| ≥2 | 27 (4.23) | 42 (3.95) | |
| Number of spontaneous abortions | 0.233 | ||
| 0 | 562 (88.09) | 963 (90.68) | |
| 1 | 65 (10.19) | 84 (7.91) | |
| ≥2 | 11 (1.72) | 15 (1.41) | |
| History of PCOS | 0.161 | ||
| No | 633 (99.53) | 1,050 (98.87) | |
| Yes | 3 (0.47) | 12 (1.13) | |
| HbA1c | 5.24 ± 0.29 | 5.04 ± 0.27 | <0.001 |
| FPG | 4.91 ± 0.41 | 4.69 ± 0.35 | <0.001 |
| 1hPG | 9.77 ± 1.63 | 7.58 ± 1.31 | <0.001 |
| 2hPG | 8.30 ± 1.40 | 6.53 ± 0.99 | <0.001 |
Continuous variables were described by means ± SDs and categorical variables were described by n (%).
Abbreviations: GDM: gestational diabetes; PCOS: polycystic ovary syndrome; HbA1c: glycated haemoglobin; FPG: fasting plasma glucose; 1hPG: 1-hour postprandial glucose; 2hPG: 2-hour postprandial glucose; SD: standard deviation.
The distribution of number of different risk loci between GDM cases and controls.
Following adjustment for age, unconditional logistic regression analyses revealed a dose-dependent positive association between PRS and the risk of GDM (p for trend <0.001) (Table 2). Even after additional adjustments were made for factors such as BMI before pregnancy, age of menarche, education level, family history of diabetes, number of births and spontaneous abortions, and history of PCOS, a significant association remained. The multivariate-adjusted ORs and 95% CIs comparing with the lowest quartile of PRS were 1.18 (0.85, 1.65) for quartile 2, 1.30 (1.01, 1.67) for quartile 3, and 1.44 (1.03, 2.01) for quartile 4, respectively. The association attenuated but persisted after excluding participants with a family history of diabetes from the analyses.
Risk of GDM according to polygenic risk scores (PRS)
| Model 1 | Model 2 | |||
|---|---|---|---|---|
| OR1 (95% CI) | p value | OR2 (95% CI) | p value | |
| All participants | ||||
| Q1 | 1.00 (ref.) | 1.00 (ref.) | ||
| Q2 | 1.17 (0.91, 1.51) | 0.230 | 1.18 (0.85, 1.65) | 0.320 |
| Q3 | 1.31 (1.05, 1.63) | 0.016 | 1.30 (1.01, 1.67) | 0.045 |
| Q4 | 1.45 (1.11, 1.91) | 0.007 | 1.44 (1.03, 2.01) | 0.031 |
| Excluding participants with family history of diabetes | ||||
| Q1 | 1.00 (ref.) | 1.00 (ref.) | ||
| Q2 | 1.18 (0.90, 1.56) | 0.281 | 1.18 (0.82, 1.64) | 0.361 |
| Q3 | 1.28 (1.03, 1.63) | 0.021 | 1.29 (1.01, 1.68) | 0.048 |
| Q4 | 1.39 (1.08, 1.90) | 0.009 | 1.40 (1.02, 1.98) | 0.038 |
OR1, OR2: Odds ratios (95% confidence interval) from unconditional logistic model. OR1: covariates adjusted for age; OR2: covariates further adjusted for BMI before pregnancy, age of menarche, education level, family history of diabetes, number of births and spontaneous abortions, and history of PCOS by stepwise forward method.
Abbreviations: GDM: gestational diabetes; PCOS: polycystic ovary syndrome; PRS: polygenic risk scores; OR: odds ratio; CI: confidential interval; BMI: body mass index.
The associations between PRS and glycemic traits were examined and are reported in Table 3. In both minimally adjusted and maximally adjusted models, PRS exhibited significant linear associations with FPG, 1hPG, and 2hPG. In the maximally adjusted model, the β coefficients and their corresponding 95% CIs were as follows: 0.181 (0.041, 0.320) for FPG, 0.225 (0.103, 0.346) for 1hPG, and 0.172 (0.036, 0.307) for 2hPG. These associations remained robust even after excluding participants with a family history of diabetes from the analyses.
Linear regression analysis of associations between polygenic risk scores (PRS) with glycemic traits
| Trait | Model 1 | Model 2 | ||
|---|---|---|---|---|
| Beta (95% CI) | p value | Beta (95% CI) | p value | |
| All participants | ||||
| FPG | 0.187 (0.059, 0.314) | 0.004 | 0.181 (0.041, 0.320) | 0.011 |
| 1hPG | 0.221 (0.101, 0.340) | <0.001 | 0.225 (0.103, 0.346) | <0.001 |
| 2hPG | 0.176 (0.054, 0.297) | 0.005 | 0.172 (0.036, 0.307) | 0.013 |
| Excluding participants with family history of diabetes | ||||
| FPG | 0.179 (0.043, 0.298) | 0.017 | 0.181 (0.041, 0.305) | 0.011 |
| 1hPG | 0.201 (0.086, 0.327) | 0.001 | 0.203 (0.088, 0.337) | 0.001 |
| 2hPG | 0.165 (0.038, 0.217) | 0.035 | 0.166 (0.036, 0.224) | 0.039 |
OR (95% CI): from linear regression models. Covariates adjusted for: see OR2 in Table 2.
Abbreviations: PRS: polygenic risk scores; OR: odds ratio; CI: confidential interval.
Table 4 presents a summary of the SNPs included in the PRS. Among them, 12 SNPs were identified to be significantly associated with GDM. Eight of these SNPs were previously reported to be associated with diabetes or its complications, while four SNPs (LOC105376481 rs7095095, ROBO3 rs7925879, CMIP rs29259794, and APOO rs12009114) are newly identified in relation to GDM. The analysis revealed that SNPs such as PAX7 rs2236835, GLIS3 rs10758593, ADAMTSL1 rs10963767, TCF7L2 rs7903146, ROBO3 rs7925879, CMIP rs29259794, PEPD rs3786897, RAE1 rs4811839, PIM3 rs28691713, and APOO rs12009114 exhibited positive associations with GDM risk. On the other hand, LINC02030 rs358806 and LOC105376481 rs7095095 were found to be negatively associated with GDM risk. Among these SNPs, TCF7L2 rs7903146 demonstrated the strongest association with GDM risk (logOR = 0.18, p = 2.37 × 10–19), followed by ADAMTSL1 rs10963767 (logOR = 0.14, p = 3.58 × 10–15).
SNPs used in the polygenic risk scores (PRS)
| SNPs used in the PRS | Chr | Positiona | Baseline | Effect | MAF | p-HWE | logOR | p value | Closest Gene | Variant type |
|---|---|---|---|---|---|---|---|---|---|---|
| 1. rs2236835 | 1 | 18990251 | G | A | 0.03 | 0.258 | 0.1 | 7.67E-10 | PAX7 | Intron |
| 2. rs358806 | 3 | 55313400 | C | A | 0.15 | 0.741 | –0.09 | 8.04E-08 | LINC02030 | Intergenic |
| 3. rs10758593 | 9 | 4292083 | G | A | 0.27 | 0.096 | 0.09 | 7.72E-08 | GLIS3 | Intron |
| 4. rs10963767 | 9 | 18797922 | T | C | 0.29 | 0.369 | 0.14 | 3.58E-15 | ADAMTSL1 | Intron |
| 5. rs7095095 | 10 | 31072535 | A | G | 0.12 | 0.214 | –0.09 | 9.12E-08 | LOC105376481 | Intron |
| 6. rs7903146 | 10 | 112998590 | C | T | 0.27 | 0.338 | 0.18 | 2.37E-19 | TCF7L2 | Intron |
| 7. rs7925879 | 11 | 124870795 | A | G | 0.29 | 0.364 | 0.11 | 3.08E-11 | ROBO3 | Intron |
| 8. rs29259794 | 16 | 81534790 | T | A | 0.23 | 0.229 | 0.12 | 6.00E-12 | CMIP | Intron |
| 9. rs3786897 | 19 | 33402102 | A | G | 0.41 | 0.192 | 0.11 | 6.72E-11 | PEPD | Intron |
| 10. rs4811839 | 20 | 57357551 | T | G | 0.39 | 0.159 | 0.1 | 5.82E-10 | RAE1 | Intron |
| 11. rs28691713 | 22 | 50356302 | C | T | 0.3 | 0.173 | 0.08 | 6.58E-08 | PIM3 | Intron |
| 12. rs12009114 | 23 | 23900803 | T | C | 0.01 | 0.988 | 0.08 | 5.81E-08 | APOO | Intron |
aData for SNP are based on data from https://wwwncbi.nlm.nih.gov/snp/.
Abbreviations: PRS: polygenic risk scores; OR: odds ratio; SNPs: single nucleotide polymorphisms; MAF: minor allele frequency; HWE: Hardy-Weinberg equilibrium.
The corresponding AUC values for the traditional risk factors and PRS + traditional risk factors were 0.703 (95%CI: 0.678, 0.728) and 0.765 (95%CI: 0.741, 0.788), respectively. The test between two AUCs reached significance (p < 0.001) (Fig. 3).
Receiver operating characteristic curves of traditional risk factors and PRS + traditional risk factors for GDM.
In this study, we conducted a GWAS on GDM and identified four novel variants and confirmed eight previously reported variants that showed associations with GDM risk. The PRS derived from the GWAS demonstrated a significant dose-response relationship with GDM risk and glycemic traits. One methodology for quantifying genetic factors is the PRS, and this approach relies on the inclusion of diverse low penetrance variants with statistical significance derived from extensive GWAS analyses [19]. These findings provide evidence that the PRS can serve as a valuable tool for predicting GDM risk and can potentially be utilized in personalized GDM prevention strategies.
To date, research exploring the utility of PRS for improving GDM risk prediction has yielded inconsistent findings. One PRS score developed by Kawai et al. [14] was based on a case-control study using the Vanderbilt Medical Center biobank, which includes data from 458 GDM cases and 1,538 pregnant controls with normal glucose tolerance. Their analysis utilized a straightforward count-based PRS comprising 34 variants previously associated with T2DM, FPG, GDM, or glucose intolerance during pregnancy. Although the PRS demonstrated an association with increased GDM risk, its utility in identifying GDM cases was limited. Lamri et al. [8] found that the predictive capacity of the PRS could be enhanced by integrating GWAS data and aggregate statistics extracted from a large-scale multi-ethnic meta-analysis. Among GDM patients in South Asia, those with the highest PRS exhibited greater risk than other groups [8]. Another study used two types of PRS, namely the simple-count PRS and the deep learning PRS, to explore the relationship between SNPs and identified risk factors such as age and BMI. This investigation revealed that the PRS of pregnant women with GDM was significantly higher than that of women in the control group (p < 0.001) [15]. Moen et al. [20] demonstrated that SNPs previously associated with hyperglycemia in Norway’s non-pregnant population could also predict the risk of hyperglycemia during pregnancy by analyzing known and theoretically related variations within the PRS. Their study, albeit having low statistical power, also found that SNPs implicated in glucose metabolism parameters among non-pregnant women exhibited associations with the same glucose metabolism parameters in pregnant women. Our results are partly consistent with previous studies and find that a PRS using 12 susceptibility loci show significant associations that reveal a 44% increased risk of GDM for the highest quartile compared with the lowest quartile of PRS. In addition, the PRS was positively associated with glucose traits such as the FPG, 1hPG, and 2hPG levels.
Among the SNPs, the risk alleles of transcription factor 7-like 2 (TCF7L2) rs7903146 have emerged as the most robust genetic indicator for GDM. Women who carry the TCF7L2 rs7903146 risk allele were found to demonstrate impaired insulin secretion and deficient proinsulin conversion compared with individuals without this allele [21]. Moreover, even after accounting for confounding variables such as BMI, the presence of the risk allele T in TCF7L2 rs7903146 is associated with early postprandial glucose control failure and a higher requirement for insulin treatment among women with GDM [22]. The genetic variant rs7903146 (C > T) within TCF7L2 exhibits a robust association with the risk of GDM [23]. One study unveiled a noteworthy association between the rs7903146 variant of TCF7L2 and GDM, with the TT genotype exhibiting a risk exceeding fivefold [24]. TCF7L2 serves as a vital gene involved in insulin secretion, islet beta-cell proliferation/apoptosis, and the maintenance of glucose homeostasis [25]. Furthermore, TCF7L2 expressed in the pancreatic beta cells plays a crucial role in glucose metabolism through regulation of the beta cell mass [26].
The genetic variant rs10963767 (T > C) located within ADAMTSL1 exhibited an association with the risk of GDM in the current investigation. ADAMTSL1 is a secreted molecule that resembles the proteases of the ADAMTS family. Nevertheless, it is devoid of the pro-metalloprotease and disintegrin-like domains typically observed in members of this family. ADAMTSL1 encompasses other ADAMTS domains that are organized in a distinct sequence, comprising four thrombospondin type I repeats [27]. In contrast to the widely distributed ADAMTSL2 and ADAMTSL3, ADAMTSL1 is expressed predominantly in human skeletal muscle. Within the scope of the present investigation, ADAMTSL1 rs10963767 displayed a robust association with an elevated risk of GDM. The ADAMTSL proteins, as dynamic non-structural proteins, are present in the extracellular matrix (ECM) and possess the ability to regulate ECM assembly [28]. Not only does the ECM participate in cellular repair and tissue remodeling, but its degradation is also intricately linked to embryonic development and angiogenesis. Anomalous ECM degradation instigates the onset of various diabetic complications. Further exploration in subsequent studies will have the potential to unravel additional intricacies, leading to a more comprehensive understanding of the mechanisms and potential therapeutic targets associated with ADAMTSL1 rs10963767 (T > C) in the context of GDM development [29].
This study identified four novel genetic loci (LOC105376481 rs7095095, ROBO3 rs7925879, CMIP rs29259794, and APOO rs12009114) associated with GDM. Among these loci, the presence of the CMIP rs2925979_T allele was previously demonstrated to elevate the risk of T2DM specifically in women [30]. Insulin resistance and defects in pancreatic β-cell function are two well-known pathophysiologic abnormalities related to the development of GDM [31]. CMIP was found to be related to mechanisms of insulin resistance or obesity. The expression of CMIP is not only negatively associated with insulin-stimulated lipogenesis through the conversion of glucose into lipids in abdominal subcutaneous adipocytes [32] but also modulates adipocyte lipolysis by regulating the signaling of NFκB [33], which increases the risk of T2DM. In addition, apolipoprotein O (APOO) enhances mitochondrial uncoupling, thereby leading to greater cellular lipid accumulation. Consequently, the accumulation of lipotoxic byproducts contributes to an augmented susceptibility to diabetes [34]. Regarding the LOC105376481 rs7095095 and ROBO3 rs7925879 variants, their associations with GDM risk have been less investigated, and the precise underlying causes for the associations remain elusive.
Timely identification and proactive measures are paramount in the management of GDM. The PRS offers a potential solution to address this concern. However, assessing the efficacy of the PRS necessitates the evaluation of its capacity to stratify the population based on discrete levels of absolute risk, thereby guiding clinical and personal decision-making processes [6]. The studies referenced in this article used PRS to predict the development of GDM and examined its associations with disease diagnosis. However, the development of PRS carries certain risks, including potential inaccuracies in individual risk estimation, inadequate representation of uncertainty in assessments, and the potential for genetic discrimination [28]. Although genetic methodologies such as GWAS and PRS have identified genes associated with GDM risk, the predictive value of these models remains limited [35]. It may be imperative to incorporate additional clinical factors and indicators to attain precise prediction.
Our study has several limitations. Firstly, our current predictive model was derived from the case-control study and incorporated age, BMI before pregnancy, age of menarche, education level, family history of diabetes, numbers of births and spontaneous abortions, and history of PCOS. An adequately designed cohort study controlling for additional identified risk factors relevant to GDM, such as dietary patterns and physical activity [33], may further enhance the present model. Secondly, this study solely examined the correlations between PRS and GDM risk. Future investigations could incorporate additional potentially functional SNPs that influence gene expression, with the aim of exploring the potential causal effects on GDM using the framework of Mendelian randomization analysis [5]. Thirdly, we exclusively evaluated risk loci identified within Chinese populations. International collaboration is necessary to explore genetic variants and elucidate the underlying biological mechanisms by which they influence GDM risk across diverse ethnic populations globally. Finally, the IADPSG criteria for the diagnosis of GDM is based on a statistical method using the odds ratio of 1.75 of macrosomia, based on the expert opinions, which indicates that the diagnosis of GDM does not imply a pathological condition. Further studies focusing on relatively severe hyperglycemia would be more meaningful from pathological and clinical perspectives. In summary, while we have presented evidence regarding the potential application of the PRS in GDM screening, our study serves as an initial investigation, and substantial efforts are still required to establish its discriminatory capacity within the general population. Moving forward, it is crucial to develop and meticulously evaluate a more precise PRS model.
The risk loci identified in this study elucidate the genetic foundation of GDM, thereby enhancing our comprehension of GDM risk. Our study substantiates the efficacy and utility of the PRS for GDM risk prediction and underscores its potential application in tailored preventive measures within GDM screening programs.
The authors would like to thank Chengdu Medical Research Project (Grant/Award Number: 2022248).
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
