Novel Insights into the Causal Relationship between Antidiabetic Drugs and Adverse Perinatal Outcomes: A Mendelian Randomization Study
Article information
Abstract
Background
Hyperglycemia during pregnancy increases the risk of adverse perinatal outcomes and birth defects. Evidence regarding the long-term safety of antidiabetic drugs during pregnancy is still lacking.
Methods
A two-sample Mendelian randomization (MR) study was performed to assess the causal association between six antidiabetic drug targets (ABCC8, DPP4, INSR, GLP1R, PPARG, and SLC5A2) and seven adverse perinatal outcomes and five congenital malformation outcomes. Inverse variance weighted (IVW) was adopted as the main MR method, and sensitivity analysis using traditional MR methods was performed to evaluate the robustness of the results.
Results
We observed strong evidence that sodium-glucose cotransporter 2 (SGLT2) inhibitors (odds ratio [OR], 0.084; 95% confidence interval [CI], 0.009 to 0.834; P=0.034) reduces the risk of preterm birth; genetic variation in sulfonylurea drug targets (OR, 0.015; 95% CI, 2.50E-04 to 0.919; P=0.045) and genetic variation in thiazolidinedione drug targets (OR, 0.007; 95% CI, 4.16E-04 to 0.121; P=0.001) reduced the risk of eclampsia/preeclampsia; glucagon-like peptide 1 (GLP-1) analogues target (β=–0.549; 95% CI, –0.958 to –0.140; P=0.009) was inversely associated with fetal birth weight; thiazolidinedione target was inversely associated with gestational age (β=–0.952; 95% CI, –1.785 to –0.118; P=0.025); SGLT2 inhibitors reduced the risk of cardiocirculatory malformations (OR, 0.001; 95% CI, 8.75E-06 to 0.126; P=0.005).
Conclusion
Most antidiabetic drugs are safe when used during the perinatal period. Of note, GLP-1 analogues may lead to a risk of low birth weight, while thiazolidinediones may lead to a reduction in fetal gestational age.
Highlights
• This study assessed the safety of antidiabetic drugs during the perinatal period.
• Most antidiabetic drugs seem genetically safe, but perinatal adverse risks remain.
• Our findings guide medication use in pregnancy complicated by hyperglycemia.
INTRODUCTION
Hyperglycemia has become one of the most common diseases in pregnancy [1]. According to the survey, 21.1 million (16.7%) women suffered from hyperglycemia during pregnancy in 2021 [2]. Hyperglycemia during pregnancy increases the risk of adverse perinatal outcomes such as low birth weight, preeclampsia, and preterm birth and birth defects [3-5]. Adequate glycemic control can reduce the risk of these adverse perinatal outcomes [6]. Although some patients can maintain expected glycemic control through lifestyle intervention, some patients still fail to reach proper blood glucose and need drug intervention. So far, insulin and metformin are the only two approved glucose-lowering drugs for pregnancy due to the lack of evidence on the long-term safety of oral glucose-lowering drugs during pregnancy [7,8]. Studies have shown that metformin can cross the blood-placental barrier, and no animal studies or clinical trials have found metformin to have teratogenic potential [9].
Mendelian randomization (MR) is a method for causal inference based on genetic variation. Because single nucleotide polymorphism (SNP) alleles segregate randomly following Mendelian laws, MR methods have the advantage of mitigating confounding factors compared to other research methods [10]. Drug target MR analysis has been used in clinical trial design. It has become an effective technique for evaluating the potential risk of drugs, agonists, inhibitors, and antagonists targeting protein-coding genes in related diseases [11]. Therefore, this study aims to comprehensively discuss the relationship between commonly used antidiabetic drug targets and perinatal outcomes (including premature birth, eclampsia/preeclampsia, birth weight, pregnancy duration, postpartum hemorrhage, premature rupture of membranes, spontaneous abortion, and various types of congenital malformations) to evaluate further the safety of the use of antidiabetic drugs in the perinatal period.
METHODS
Study design
This study adopted a two-sample MR design (Fig. 1). Exposure and outcome data for this design were obtained from two independent, nonoverlapping populations. Proxies for antidiabetic drug use were composed by identifying genetic variants (cis-variants) within genes encoding protein targets for antidiabetic drugs in the genome-wide association study (GWAS) Glycated Hemoglobin Summary Dataset (https://www.ebi.ac.uk/gwas/studies/GCST90002244). To ensure the validity of potential causal effects, MR analyses needed to fulfill three core assumptions: (1) genetic variants are robustly associated with the exposure (relevance); (2) genetic variants are independent of confounders (exchangeability); and (3) genetic variants influence the outcome only through the exposure (exclusion restriction) [12]. This study is reported based on the guidelines outlined in the Strengthening the Reporting of Observational Studies in Epidemiology Using Mendelian Randomization (STROBE-MR) protocol [13].
Overview of study design and analysis strategy. SNP, single nucleotide polymorphism; PPARG, peroxisome proliferator-activated receptor gamma; SLC5A2, solute carrier family 5 member 2; GLP1R, glucagon-like peptide 1 receptor; DPP4, dipeptidyl peptidase 4; INSR, insulin receptor; ABCC8, ATP binding cassette subfamily C member 8; LD, linkage disequilibrium; IV, instrumental variable; KCNJ11, potassium inwardly rectifying channel subfamily J member 11; HbA1c, glycosylated hemoglobin; MR, Mendelian randomization; T2DM, type 2 diabetes mellitus.
Selection of genetic instruments
Seven antidiabetic drug targets were identified from Drugbank (https://go.drugbank.com/), including dipeptidyl peptidase 4 (DPP4, DPP4 inhibitors target), insulin receptor (INSR, insulin/insulin analogues target), solute carrier family 5 member 2 (SLC5A2) (sodium-glucose cotransporter 2 [SGLT2] inhibitors target), glucagon-like peptide 1 receptor (GLP1R, GLP-1 analogues target), ATP binding cassette subfamily C member 8 (ABCC8) and potassium inwardly rectifying channel subfamily J member 11 (KCNJ11) (sulfonylureas targets), and peroxisome proliferator-activated receptor gamma (PPARG, thiazolidinediones target). Metformin was excluded from further analysis because its target could not be identified. Since the chromosomal locations of ABCC8 and KCNJ11 are similar, we chose the midpoint between them as the dividing line to distinguish them (Table 1).
Summary information of antidiabetic drugs classes, targets, and encoding genes
We identified cis-expression Quantitative Trait Loci (cis-eQTL) within each coding gene (±1 Mb base pairs of the gene position) and selected those that retained significant correlations with glycosylated hemoglobin levels. Because the retrieved SNPs significantly associated with cis-variants were fewer (less than 3) at the genome-wide significance level P<5×10−8, we gradually extended the P value to 5×10−6, or even 0.05, thereby identifying the SNPs used for analysis [14]. We set the linkage disequilibrium (LD) parameter to r2<0.2 within 100 kb, and SNPs with palindromic structures were deleted to ensure the reliability of the results [11].
The influence of weak instrument bias on the results was eliminated by calculating the F statistic of the selected SNPs and removing SNPs with F<10. If an SNP was not present in the resulting GWAS, it was replaced with an alternative SNP in high LD (r2>0.80) using SNiPA (https://snipa.helmholtzmuenchen.de/snipa3/index.php). If there is no suitable alternative SNP, it is discarded.
Outcome data
To determine the reliability of the selected instrumental variables (IVs), MR analysis was performed using type 2 diabetes mellitus (T2DM) as a positive control. GWAS data for T2DM were obtained from the GWAS Catalog database. Eclampsia/preeclampsia, preterm birth, premature rupture of membranes, postpartum hemorrhage, spontaneous abortion, birth weight, pregnancy duration, and congenital malformations were selected as the primary outcomes of our research. In our study, eclampsia/preeclampsia, premature rupture of membranes, preterm birth, postpartum hemorrhage, and spontaneous abortion GWAS data from the FinnGen database (released 10) [15]. GWAS data for birth weight and gestational age were obtained from the Early Growth Genetics (EGG) Consortium. Considering the widespread harm caused by drug-induced diseases, we included data on congenital malformations on a large scale. Data on various congenital malformations come from the UK Biobank (UKB), including skin malformations, heart and circulatory system malformations, digestive system malformations, genitourinary system malformations, and coagulation malformations. The detailed data information is shown in Supplementary Table 1.
Statistical analysis
We evaluated the effects of antidiabetic drugs on perinatal outcomes and congenital malformations based on a two-sample MR analysis. The inverse variance weighted (IVW) method was mainly used to evaluate the causal relationship between exposure factors and outcome factors [16], and the remaining four methods were used as supplementary analysis, including MR-Egger, weighted median, simple model, and weighted model. The choice of fixed-effects model or random-effects model was determined by heterogeneity test. Specifically, Cochran’s Q test was used to determine whether there was heterogeneity. When the heterogeneity index showed that P<0.05 of the Q test and I2>50%, the random-effects model should be selected because it can provide more precise estimates and confidence intervals (CI) than the fixed-effects IVW method. If no significant heterogeneity was found, the fixed-effects model was used. Furthermore, the MR-Egger intercept test and MR pleiotropy residuals and outliers (MR-PRESSO) were used to assess relative pleiotropy. Outlier SNPs will be detected by MR-PRESSO outlier test and P index level of 0.05.
Sensitivity analysis
We used the intercept term of MR-Egger regression to represent horizontal pleiotropy and performed Cochrane’s Q value to assess the heterogeneity of IVW estimators, with P<0.05 indicating evidence of heterogeneity. In addition, we used MR-PRESSO as a supplement to assess horizontal pleiotropy [17]. MR-PRESSO is designed to detect outlier genetic variants and correct for horizontal multivariate validity by removing outliers. MR-PRESSO can also determine whether the causal effect changes substantially before and after removing outliers in MR analysis.
Ethical statement
Our analysis used publicly available GWAS summary statistics. No new data were collected, and no new ethical approval was required.
RESULTS
Selection and validation of genetic instruments
By applying the thresholds we set in our method analysis, a total of 28 significant cis-eQTL SNPs were included as genetic instruments, including four SNPs representing ABCC8, four SNPs representing DPP4, eight SNPs representing GLP1R, five SNPs representing INSR, and three SNPs representing PPARG and four SNPs representing SLC5A2. Afterward, the influence of weak IVs on the results was eliminated by calculating the F statistic. In this process, the F statistics of the IVs of KCNJ11 were all <10 and therefore were excluded from further analysis. The F statistics of the remaining studied IVs range from 10.016 to 49.904, indicating that our MR analysis is hardly affected by weak instrument bias. The detailed information of all included SNPs are listed in Supplementary Table 2. Then, we performed a two-sample MR analysis using T2DM as a positive control, and the results are shown in Supplementary Table 3. The results showed that all antidiabetic drug target genes involved in this study were significantly associated with a reduced risk of T2DM. No significant multiple effects and heterogeneity were observed in the results.
Impact of genetic variation in antidiabetic drug targets on perinatal outcomes and risk of congenital malformations
In our preliminary analysis using the IVW approach, we observed strong evidence that SGLT2 inhibitors (odds ratio [OR], 0.084; 95% CI, 0.009 to 0.834; P=0.034) reduced the risk of preterm birth. Genetic variation in sulfonylurea drug targets (OR, 0.015; 95% CI, 2.50E-04 to 0.919; P=0.045) and genetic variation in thiazolidinedione drug targets (OR, 0.007; 95% CI, 4.16E-04 to 0.121; P=0.001) reduced the risk of eclampsia/preeclampsia. GLP-1 analogues target (β=–0.549; 95% CI, –0.958 to –0.140; P=0.009) was inversely associated with fetal birth weight, indicating that GLP-1 analogues may cause reduced fetal birth weight. Thiazolidinedione target was inversely associated with gestational age (β=–0.952; 95% CI, –1.785 to –0.118; P=0.025) (Table 2, Figs. 2 and 3).
Results of two-sample MR analysis of antidiabetic drugs and adverse pregnancy outcomes
The forest plot of the effects of antidiabetic drugs on adverse perinatal outcomes. Odds ratio (OR), 95% confidence interval (CI), and P values were calculated for the respective method of Mendelian randomization (MR) analysis. P<0.05 was considered significant. nSNP, number of single nucleotide polymorphisms; IVW, inverse variance weighted; DPP4, dipeptidyl peptidase 4; GLP-1, glucagon-like peptide 1; SGLT2, sodium-glucose cotransporter 2.
The forest plot of the effects of antidiabetic drugs on birthweight and gestational age. Odds ratio, 95% confidence interval (CI), and P values were calculated for the respective method of Mendelian randomization (MR) analysis. P<0.05 was considered significant. nSNP, number of single nucleotide polymorphisms; IVW, inverse variance weighted; DPP4, dipeptidyl peptidase 4; GLP-1, glucagon-like peptide 1; SGLT2, sodium-glucose cotransporter 2.
In the MR analysis of congenital malformations, we found that SGLT2 inhibitors had a positive effect on cardiocirculatory malformations (OR, 0.001; 95% CI, 8.75E-06 to 0.126; P=0.005) (Table 2, Fig. 4). We did not observe significant heterogeneity or multiplicity of effects in our findings.
The forest plot of the effects of antidiabetic drugs on congenital malformation. Odds ratio (OR), 95% confidence interval (CI), and P values were calculated for the respective method of Mendelian randomization (MR) analysis. P<0.05 was considered significant. nSNP, number of single nucleotide polymorphisms; SGLT2, sodium-glucose cotransporter 2; IVW, inverse variance weighted; GLP-1, glucagon-like peptide 1; DPP4, dipeptidyl peptidase 4.
DISCUSSION
This study systematically evaluated the relationship between six antidiabetic drugs and multiple perinatal outcomes through drug-targeted MR analysis. We found that SGLT2 inhibitors were associated with a reduced risk of preterm birth and congenital malformations of the cardiocirculatory system. GLP-1 analogues were associated with reduced fetal birth weight. Both sulfonylureas and thiazolidinedione drug targets were associated with reduced risk of eclampsia/preeclampsia. We also found a negative correlation between thiazolidinedione drug targets and gestational age.
GLP-1 receptor agonists (GLP-1RA) is a new antidiabetic drug that maintains blood glucose homeostasis by stimulating insulin secretion from β-cells and inhibiting glucagon release from α-cells [18-20]. The current study found that GLP-1 analogues are generally safe and have no significant impact on perinatal outcomes and fetal congenital malformations. Consistently, a prospective multicenter observational study evaluated pregnancy outcomes after first-trimester exposure to GLP-1RA and found that these drugs did not increase the risk of adverse pregnancy outcomes and birth defects [21]. However, we found that GLP-1 analogues are associated with reduced fetal birth weight. An animal study suggests that maternal exposure to GLP-1RA during pregnancy or lactation is associated with fetal growth retardation and weight loss [22]. This may be because GLP-1 analogues influence appetite control and feelings of fullness by acting on the hypothalamus [23]. They reduce food intake and slow gastric emptying, leading to weight loss [24-26]. Therefore, the application of GLP-1RA should alert the risk of low-birth-weight infants. In addition, SGLT2 inhibitors, another common antidiabetic drug, can effectively reduce blood glucose levels in patients with T2DM by reducing renal tubular glucose reabsorption and have a certain weight loss effect [27]. However, we did not find that SGLT2 inhibitors are associated with low-birth-weight infants. Interestingly, we found that SGLT2 inhibitors were associated with a reduced risk of preterm birth and congenital malformations of the cardiocirculatory system. So far, a series of studies have proven that SGLT2 inhibitors have anti-inflammatory and anti-fibrotic effects [28-30]. Research shows that chronic inflammation throughout the body is a risk factor for premature birth in women [31]. Sulfonylureas are oral antidiabetic drugs whose main mechanism of action is to inhibit adenosine triphosphate-sensitive potassium channels (Katp channels) and increase insulin secretion from pancreatic β-cells [32-34]. We found that sulfonylureas reduced the risk of eclampsia. Previous studies have shown that hyperglycemia increases the production of various metabolic injuries, leading to insulin resistance in the placenta, stimulating the transcription and expression of proinflammatory cytokines, promoting placental cell apoptosis and dysfunction, and finally leading to eclampsia [35,36]. Therefore, effective blood glucose control will reduce the risk of eclampsia. Interestingly, we found that thiazolidinediones can also significantly reduce the risk of eclampsia, which is related to the effect of the drugs themselves on improving insulin resistance. Previous studies have demonstrated that PPAR is an important transcription factor involved in various placental pathways and plays a role in placental development and function [37]. Thiazolidinediones can have beneficial effects on stressed preeclamptic placentas by activating cell growth and anti-oxidative stress pathways, including the heme oxygenase 1 (Hmox1) gene, and restoring pathways disrupted by PPARG in preeclampsia [38]. However, our study also found that thiazolidinediones may lead to a reduction in fetal gestational age. A previous study found that women who carry homozygous variants of the PPARG gene have a 43% chance of being born small for gestational age due to perinatal brain injury [39].
There are few human clinical data on the effects of maternal use of antidiabetic drugs during pregnancy or lactation on the fetus, making it impossible to assess the safety of perinatal use of antidiabetic drugs reliably. GWAS is a hypothesis-free study design used to test the association between thousands or millions of genetic variants and phenotypes [40]. The principal aim of a GWAS is to identify variants associated with phenotypes, whose alleles have been randomly assigned to individuals before any exposure or outcome [40,41]. Compared with traditional regression analysis, MR analysis can avoid the influence of potential confounding factors. Due to the randomness of genetic variation, the distribution of exposure variables (such as a disease, biomarker, or drug effect) is usually independent of the pregnancy status or other similar subgroup divisions of the participants [42]. Therefore, even in different subgroups (such as pregnant and non-pregnant populations), MR methods can still provide robust causal inferences [43,44]. We present the first large-scale study of the association between multiple antidiabetic drugs and adverse perinatal outcomes and fetal congenital malformations. Considering the interference of the mother’s genetic background on the results, we chose data on fetal congenital malformations as the outcome, in which the effects of these SNPs are mainly manifested in fetal physiology and focused more precisely on the genetic background of the fetus [45,46]. We provide evidence that using GLP-1 analogues and thiazolidinediones may contribute to adverse perinatal outcomes. With the increased availability of GWAS data, drug-target MR can effectively identify targets with an efficiency of up to 70% [11]. The results of positive controls found that the genetically predicted drug effects were consistent with clinical evidence and mechanisms, indicating the robustness of the selected IVs [22,47,48]. This concept will give a new direction for choosing medication regimens for patients with hyperglycemia during pregnancy. However, our study also has some limitations. First, the data we selected for congenital malformations of the heart and large vessels include a small amount of non-European population data, which may have confounded the results. However, their proportion in the overall population is minimal, and their impact on the results is negligible. Second, during our investigation of the link between antidiabetic drugs and congenital malformations, we excluded ABCC8 and INSR from the analysis due to the limited number of SNPs and the low reliability of the findings. Finally, MR is limited to predicting adverse events related to the drug mechanism. In contrast, off-target effects that are not related to the drug mechanism of action cannot be predicted by MR [49]. However, drug-target MR was performed using cis-variants, reducing the possibility of horizontal pleiotropy, which are selected in and around drug-target encoding loci [46]. Moreover, we also performed sensitivity analyses using robust methods, including weighted median and MR Egger methods, which further supported the absence of horizontal pleiotropy in these MR analyses [11]. DPP4 inhibitors, insulin/insulin analogues, and thiazolidinediones were found to be possibly associated with an increased risk of spontaneous abortion, although their P were not significant (P>0.05). A more rigorous approach should be to expand the sample size and population groups and conduct an in-depth analysis. Meanwhile, more efforts should be made to combine both laboratory studies and clinical trials to clarify the effect of antidiabetic drugs on adverse perinatal outcomes.
In conclusion, our findings provide new evidence for the perinatal use of antidiabetic medications in pregnant women with hyperglycemia. Most antidiabetic drugs are safe to use during the perinatal period and reduce the risk of adverse perinatal outcomes. Of note, GLP-1 analogues may lea1d to a risk of low birth weight, while thiazolidinediones may lead to a reduction in fetal gestational age.
SUPPLEMENTARY MATERIALS
Supplementary materials related to this article can be found online at https://doi.org/10.4093/dmj.2024.0521.
Genome-wide association study summary data and expression quantitative trait loci studies’ data information
Instrumental variables for antidiabetic drugs
MR analysis results of positive control
Notes
CONFLICTS OF INTEREST
No potential conflict of interest relevant to this article was reported.
AUTHOR CONTRIBUTIONS
Conception or design: C.S., G.W., J.L.
Acquisition, analysis, or interpretation of data: C.S., X.H., X.C., J. T.
Drafting the work or revising: C.S.
Final approval of the manuscript: all authors.
FUNDING
This work was supported by grants from the Chinese National Natural Science Foundation (No. 82072527), Beijing Natural Science Foundation (L248018) to Jia Liu and Beijing Hospitals Authority’s Ascent Plan (DFL20220302) to Guang Wang.
ACKNOWLEDGMENTS
The authors appreciate all the subjects involved in this study.