https://orcid.org/0000-0003-1803-5693
This is an uncorrected proof.
Figures
Abstract
Much remains unknown about the genetics of immune system changes during pregnancy. We used SNP data in a pregnancy cohort to genetically investigate 47 immune biomarkers at two timepoints, along with change between timepoints (Δ). We identified 19 biomarkers with significant SNP-based heritability and 34 with genome-wide significant signals, demonstrating genetic regulation. The same biomarkers measured in early- and mid-pregnancy shared about half of significant associations across timepoints, with enrichment for immune pathways. In contrast, Δ showed enrichment in transcription factors and developmental processes. About half of suggestive associations overlapped with non-pregnancy associations. However, these data leave a substantial fraction of potentially timepoint-specific and pregnancy-unique findings. Nearby genes were enriched for high expression in decidual cells at the maternal-fetal interface, reinforcing the novelty of our results. We additionally explored the relationship between immune genetic associations and prior GWAS of pregnancy complications. Overall, we present the first two-timepoint genetic study of immune profile in pregnancy.
Author summary
Much remains unknown about the genetics of immune system changes during pregnancy. We used a pregnancy cohort with genetic data and measurement of 47 immune biomarkers at two timepoints to investigate early and mid-pregnancy, as well as changes in biomarker levels between timepoints (Δ). We demonstrated genetic regulation of immune biomarkers using heritability estimation and genome-wide significant associations. We found shared and unique genetics of early vs. mid-pregnancy timepoints, and unique genetic influences to Δ (related to developmental and regulatory genes). About half of our top results were not shared by studies of the same immune biomarkers in non-pregnant adults, and we found enrichment for cell types at the maternal-fetal interface. We explored whether these genetic results might overlap with common pregnancy complications such as preeclampsia, preterm birth, and gestational diabetes. Overall, our results suggest strong and potentially pregnancy-specific genetic contributions to immune profile.
Citation: Cakir M, Traglia M, Alexeeff S, Ames JL, Ashwood P, Grosvenor LP, et al. (2026) Genetic survey of biomarkers at early and mid-pregnancy identifies pregnancy-specialized immune regulation. PLoS Genet 22(6): e1012204. https://doi.org/10.1371/journal.pgen.1012204
Editor: Heather J. Cordell, Newcastle University, UNITED KINGDOM OF GREAT BRITAIN AND NORTHERN IRELAND
Received: February 10, 2026; Accepted: June 4, 2026; Published: June 30, 2026
Copyright: © 2026 Cakir et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: Full summary statistics of each immune biomarker GWAS are publicly available from NHGRI-EBI GWAS Catalog (https://www.ebi.ac.uk/gwas/) with the identifiers between GCST90809085 - GCST90809225. Publicly available GWAS results used in this study were obtained from following sources: study 1 (ebi.ac.uk/gwas/publications/27989323), study 2 (ebi.ac.uk/gwas/publications/37563310), PE (ebi.ac.uk/gwas/publications/37285119), GDM (ebi.ac.uk/gwas/publications/38182742), and PTB (egg-consortium.org/Gestational-duration-2023.html). scRNA-seq expression datasets were obtained from supplementary dataset of Suryawanshi et al. (science.org/doi/10.1126/sciadv.aau4788) and Human Protein Atlas (proteinatlas.org/about/download). Data used in this study (beyond summary statistics described above) were provided by the Kaiser Permanente Research Bank from the Kaiser Permanente Research Bank collection, which includes the Kaiser Permanente Research Program on Genes, Environment, and Health. Access to data used in this study may be obtained by application to the Kaiser Permanente Research Bank at kp.org/researchbank/researchers.
Funding: This work was supported by the National Institute of Child Health and Human Development (https://www.nichd.nih.gov/) (Grant No: R01HD095128 [to LAC] and R01HD112326 [to LAC]). ES was supported by the National Center for Advancing Translational Sciences, National Institutes of Health (https://ncats.nih.gov/), through grant number UL1 TR001860 and linked award TL1 TR001861. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the National Institutes of Health.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Throughout pregnancy, maternal immune cells, uterine cells, and the placenta sustain a dynamic environment to support fetal growth. The maternal immune system during pregnancy must enable implantation, prevent rejection of the semi-allogeneic fetus, and protect the mother and baby against infections [1–3]. This complex balance is maintained by fluctuations between regulatory, pro- and anti-inflammatory conditions throughout pregnancy [4,5]. During the first trimester, a pro-inflammatory environment aids implantation and placentation. As pregnancy progresses through the second trimester, the environment shifts towards immune balance and growth factors. Finally, pro-inflammatory factors become critical again in the lead-up to labor. Dysregulation of this pattern can lead to pregnancy complications and hinder fetal development [3,6–8]. Therefore, insight into regulation of the maternal immune environment can inform practices that support healthy pregnancy.
Immune cells mediate the maternal environment through secretory molecules like cytokines, chemokines, and growth factors [8,9]. These factors recruit specialized immune cells, such as decidual natural killer cells and macrophages, to the maternal-fetal interface at the right time to maintain interactions between the fetus and placenta [7,10,11]. Various studies have been performed to characterize the temporal dynamics of these proteins during pregnancy. One longitudinal study revealed that abundance of 10% of 1125 plasma proteins changed significantly across gestational age [12]. These proteins were enriched in “defense response” and “leukocyte migration” pathways, highlighting immune system’s importance. Another longitudinal study that included samples obtained post-partum revealed substantial inter-individual differences that remained stable over time [13]. Despite growing knowledge of the importance of immune mediators in pregnancy, our understanding of the impact of genetics on their levels and changes during pregnancy is incomplete.
Previous studies in non-pregnant populations have identified genetic loci associated with levels of cytokines and chemokines [14,15], highlighting the fact that genetic variation contributes to inter-individual variation and indicating potential for germline genetics to inform causal relationships. However, the immune system changes dramatically in pregnancy, suggesting that there may be genetic factors, signaling pathways, or regulatory processes exclusive to pregnancy. Our previous study investigated genetic determinants of maternal immune biomarkers at a single timepoint in the second trimester [16]. We identified three genome-wide significant (GWS) loci associated with distinct biomarkers, two of which had not been previously observed [14,16]. This work provided insight into the mid-gestational timepoint, but considering the dynamic nature of immunity, it is possible that different mechanisms would be predominant at other pregnancy stages. Therefore, studying more than one timepoint may provide valuable insight into how genetics contribute to the maternal immune milieu and differentiate between factors that impact early or mid-stages of pregnancy. Additionally, the rate at which a biomarker’s levels change could be controlled by unique regulatory mechanisms. Studies that explored response to infection or other challenges revealed different genetic factors affecting inter-individual variation at baseline and in responsiveness [17–19]. By investigating the genetic loci mediating biomarker changes between two timepoints, we can explore how the immune system responds to the challenge of a semi-allogeneic fetus at different phases of pregnancy.
Our previous study took advantage of genotyping and immune biomarker data in the Early Markers for Autism (EMA) study to investigate the relationship between genetics and maternal immunity [16]. Although this cohort was ascertained for children with autism spectrum disorder (ASD) and other neurodevelopmental disorders (DD), we gained novel insight into how maternal and fetal genetics shape levels of immune mediators in pregnancy. Our current IMPaCT (Immune and Metabolic Markers during Pregnancy and Child Development) study also oversamples pregnancies resulting in children with ASD and DD, and matches general population (GP) controls to these demographics [20]. Despite sampling bias, under the (testable) assumption that genetic regulation of the immune system is similar regardless of outcome, we can utilize this large dataset to further improve our understanding of the interplay between genetics, pregnancy, and immune system. We test this assumption by estimating effect sizes for our detected loci in GP controls only to demonstrate minimal difference with the full IMPaCT study.
In this study, we identified genetic determinants of circulating levels of immune biomarkers at two different timepoints in pregnancy. To separately investigate genetic regulation of changes in biomarker levels, we also established a delta (Δ) variable based on the differences between measurements. We first established the existence of common genetic determinants using SNP-based heritability estimation and genome-wide association studies (GWAS). Next, we examined the biological processes enriched in associated loci across immune biomarkers to identify pathways important in the immune biology of early pregnancy, middle pregnancy, and responding over pregnancy. We also compared our GWAS-nominated loci with other studies of the same biomarkers in non-pregnant populations to discern putative pregnancy-specialized loci. Because the placenta and decidualization of endometrium create a pregnancy-exclusive environment, we examined enrichment of our loci in decidual cell types to further investigate immune regulation unique to pregnancy. Finally, pregnancy complications are associated with immune system dysregulation. To investigate whether genetic drivers of the immune environment contribute to adverse pregnancy outcomes, we examined overlap between the genetics of immune biomarkers and the genetics of preeclampsia, gestational diabetes, and preterm birth.
Results
Dataset characteristics
The IMPaCT cohort included 2436 mothers across three child outcome groups (ASD N = 339, DD N = 1193, GP N = 904) with blood sample measurements taken at two timepoints (14–147 days apart) [S7c Fig] per individual over the first and second trimesters (Fig 1a). To study the changes observed during the progression of pregnancy, we also computed Δ based on the difference between first (T1) and second timepoint (T2) biomarker levels, adjusted for the timing of each measurement as well as the amount of time between.
a. Schematic summarizing the study design and flow. b. Venn diagram of number of genes annotated to nearby GWS SNPs per time variable and overlap. c. T1 (blue) and T2 (red) effect size estimates of SNPs (that are GWS in at least one timepoint) are drawn with a connecting line representing the magnitude of their difference. In cases where the effect sizes are significantly different, the line is colored based on the timepoint with greater (absolute value) effect size, otherwise it is grey. Circle size represents the corresponding p-value. The y-axis lists the corresponding biomarker – SNP associations and the order matches Table 2.
To understand the independence of our genetic results per biomarker, we computed pairwise Pearson correlation matrices per time variable (S1a–S1c Fig). We identified a group of 11 (out of 47) biomarkers that were strongly correlated across T1, T2, and Δ (FGF-2, IL-2, GM-CSF, IL-12p70, IFNγ, Fractalkine, MCP-3, IL-22, TNFβ, IL-13, TNFα). Most of the remaining biomarkers showed limited correlation: only 21 additional pairs in T1, 20 pairs in T2, and 6 pairs in Δ had r > 0.5 out of 1081 pairs. Additionally, individual biomarkers’ T1-T2 measurements were highly correlated (S1d Fig). However, T1-Δ and T2-Δ for the same biomarkers showed limited correlation (S1e and S1f Fig), suggesting that, for instance, higher T1 measurement didn’t correspond with greater decrease (or less increase) over time to regress to the mean. Likewise, inter-biomarker pairs showed some differing patterns of correlation for Δ compared with T1 or T2 (S1a–S1c Fig). For instance, sCD40L-EGF showed an increase in correlation in Δ (T1/T2 r = 0.60, Δ r = 0.84) along with a negative correlation between sCD40L-IL-7 (r = -0.35) and EGF-IL-7 (r = -0.33) not observed in T1 or T2. Overall, these phenotypic relationships suggest that genetic analyses across biomarkers and time variables could provide some independent information.
Heritability
We estimated SNP-based heritability [21] with individual-level data for all biomarkers per time variable and, in total, identified 19 of 47 biomarkers with significant (P < 0.05) heritability for at least one time variable (T1:10, T2:7, Δ:7) (Table 1). This is substantially more than the 2–3 expected by chance for each time variable, even considering all biomarkers as independent. MCP-1 had a high heritability estimate for both T1 and T2 (T1 h2 = 0.50, T2 h2 = 0.55) and PDGF AB/BB had moderate heritability for both (T1 h2 = 0.31, T2 h2 = 0.38). In contrast, there were 2 biomarkers with high (TNFβ and IL-12p70) and 5 with moderate (IL-4, fractalkine, GROα, IL-17F, IL-12p40) heritability estimates only for T1. Additionally, IL-22 had moderate estimates for both T1 and Δ (T1 h2 = 0.43, T2 h2 = 0, Δ h2 = 0.46), and IL-1RA (T1 h2 = 0.058, T2 h2 = 0.25, Δ h2 = 0.31) and IL-27 (T1 h2 = 0.18, T2 h2 = 0.30, Δ h2 = 0.31) had moderate estimates for T2 and Δ. IL-1α had a high heritability estimate only for Δ, along with 3 biomarkers with moderate estimates only for Δ (IL-7, IL-5, and IL-13). Since there is a small subset of individuals with only one measurement, we further validated differences in T1 and T2 heritability estimates by including only individuals with both T1 and T2 measurements. In most of these cases (IL-22, IL-4, IL-17F, IL-12p40, and MIP-1α), the estimates remain substantially different since the non-significant timepoint still had a near-zero heritability estimate (S1 Table). However, due to the high standard error, we lack power to perform stringent tests for estimate differences. Overall, we observe many more significant heritability estimates than expected by chance, with some strikingly different estimates for the same biomarker measured in the same individuals in early- vs. mid-pregnancy.
GWAS results
We performed GWAS under an additive QTL model [22] for all 47 immune biomarkers at T1, T2, and Δ. To obtain permuted results that will enable us to test the significance of our observations, we also generated 100 datasets of biomarker levels randomized to different individuals and ran GWAS with the exact same parameters for each time variable. 34 out of 47 biomarkers had at least one genome-wide significant (GWS) SNP (P < 5x10-8) for at least one time variable: 20 biomarkers at T1 (644 SNPs), 21 at T2 (736 SNPs), and 9 for Δ (55 SNPs) (Table 2). Given that GWS represents α = 0.05, even counting all 141 GWAS (47 by 3 time variables) as independent, we see many more GWS associations than the 7 loci expected by chance. These GWS SNPs include highly-significant and previously-identified associations between rs4905 and IL-27 (T1 P = 2.05x10-94, T2 P = 4.83x10-72), and rs12075 and eotaxin, GROα, and MCP-1 (T1, T2 8.00x10-51 > P > 5.14x10-20) [14,15,23,24]. rs6565538 is associated with fractalkine and MCP-3, along with suggestive associations with GM-CSF, IFNγ, and IL-13. These 5 biomarkers are part of the 11-biomarker correlation cluster (S1a–S1c Fig).
We mapped GWS SNPs to nearest gene transcription start sites (within 1000kb) [25], annotating 47 genes for T1 (1 unannotated), 46 for T2, and 10 for Δ (1 unannotated). This mapping revealed that Δ results represent a distinct set of loci, as there were no shared genes with T1 or T2 (Fig 1b). In contrast, over half of GWS loci (26 genes) were shared between T1 and T2, the vast majority corresponding to genes that were identified as significant at both T1 and T2 for the same biomarker. These shared associations were typically autoregulatory or with their corresponding receptor, such as G-CSF–CSF3 (rs3826331), GROα–CXCL1 (rs3111692), MDC–CCL22 (rs654974), and ACKR1’s (rs12075) associations with eotaxin, GROα, and MCP-1, providing high confidence of their biological validity. As previously shown [14], rs12075’s association with three biomarkers with moderate to low correlation (eotaxin-GROα r = 0.22, eotaxin-MCP-1 r = 0.51, GROα-MCP-1 r = 0.29) exemplify how variation at one cytokine receptor (ACKR1) can pleiotropically impact levels of multiple biomarkers. rs4905, mapped to EBI3 (which encodes a subunit of IL-27 and IL-35), provides an interesting example of strong association with both its own and another biomarker’s levels (IL-1β), despite only modest phenotypic correlation (T1 r = 0.23, T2 r = 0.36). To better understand the relationship between EBI3, IL-27, and IL-1β, we further examined the association between the lead variant and IL-1β in a conditional analysis to distinguish between mediation (EBI3 genotype influences IL-27 levels which influence IL-1β levels) and pleiotropy (EBI3 genotype influences IL-27 levels AND EBI3 genotype influences IL-1β levels). Mediation analysis revealed that 78% of the effect of EBI3 variant on IL-1β was estimated as indirect effects mediated through IL-27 and the estimate of direct effect of EBI3 on IL-1β was not significant, helping us establish the directionality of this shared association (S2 Fig).
To better characterize observed differences between T1- and T2-associated loci, we compared effect size estimates for loci GWS in at least one timepoint (Fig 1c). In general, we observed minimal differences for loci that were GWS in both timepoints, such as IL-27–rs4905, MIP-1β–rs1015673, GROα–rs3111692, and G-CSF–rs3826331. However, associations like sCD40L–rs7159135 (timepoint difference P = 0.020), VEGF-rs4714719 (P = 0.014), and IFNα2–rs112624350 (P = 0.044) were GWS only in T1 and not suggestive in T2. This specificity to T1 was also reflected in the magnitude of differences between T1 and T2 effect size estimates. Similarly, there are T2-associated loci with much larger effect sizes in T2 than in T1, such as MIP-1α–rs74135708 (P = 0.023), sCD40L–rs138877592 (P = 0.011), and MIP-1α–rs6793722 (P = 0.045). Together these suggest timepoint-specific immune regulation.
Child outcome was included in our models as a covariate, as the study cohort oversampled pregnancies resulting in children with ASD and DD along with matched GP controls [20]. To test whether the observed associations were confounded by this study design, we ran secondary analyses with GP samples only, focusing on GWS loci associated with T1, T2, and Δ. S3 Fig shows the high correlation between effect size estimates of the full cohort results and GP-only results (T1, T2 r = 0.95, Δ r = 0.99), suggesting conserved genetic regulation.
Pathway enrichment analysis
To examine biological processes represented by observed associations per time variable, we performed pathway enrichment analysis after mapping suggestive SNPs (P < 10-6) to nearest genes (as above). All SNPs included in GWAS were also mapped to nearest genes to generate the appropriate background set of genes. Within each time variable, genes near suggestive SNPs across biomarkers were combined into gene sets for T1, T2, and Δ. Using Gene Ontology (GO) Molecular Function and GO Biological Process definitions [26], we observed enrichment related to cytokine/chemokine binding and immune receptor activity for T1 and T2, highlighting that the associations we identified have known immune roles (Figs 2 and S4).
Significantly enriched GO-Molecular Function pathways within each time variable’s set of query genes assigned to nearby suggestive SNPs (P < 10-6). Categories are ordered based on adjusted p-values and circles are color coded based on adjusted p-values. Count in x-axis denotes the number of genes in the query set that belong to the category. Circle sizes are proportional to gene ratio, which denotes the ratio of the number of genes in the query set in that category to the total number of genes in the category.
In line with the distinct genetic results in Δ, we identified differing enrichment. GO-Biological Process terms show a trend towards developmental processes and differentiation (S4 Fig). Transcriptional activity was the only enriched term with GO-Molecular Function (Fig 2). To further explore relationships with transcription, we obtained a curated list of known transcription factors from TFlink [27]. We observed that 11.1% of Δ-associated genes encode transcription factors, and this corresponded to a significant overrepresentation when compared against permuted GWAS results (P = 0.04) (S2 Table). Some of these transcription factors have known roles in embryonic development, suggesting a causal connection between variation in transcriptional developmental processes and changes in biomarker levels over pregnancy.
Comparison with non-pregnant population GWAS
Genetic regulation of immune biomarkers has previously been investigated in largescale GWAS of non-pregnant populations composed of adult men and women [14,15]. A comparison against these studies provides the opportunity to identify variants that are particularly relevant in the context of pregnancy. For this analysis, we examined Ahola-Olli et al. (study 1) [14] with 28 overlapping biomarkers and Zhao et al. (study 2) [15] with 25 overlapping biomarkers. Per biomarker, we extracted highly suggestive SNPs (P < 10-7) also present in the comparison study. We categorized these based on whether they are also suggestive (P < 10-7) or above this threshold in the comparison study, labeling them as shared or putative unique variants, respectively. We performed an additional test of putative unique variants for homogeneity of effect size, characterizing those with significantly different effect size estimates as unique. Using the (P < 10-7) threshold resulted in only a single putative unique variant that did not reject homogeneity of effect size, likely due to the high power of the comparison studies, clearly demarcating shared and unique loci (Fig 3). There was one locus (ACKR1) with variants spanning the (P < 10–7) threshold; we classified the locus as shared, although it is plausible that both shared and pregnancy-unique variants exist at the same locus. These comparisons were performed separately for study 1 and study 2.
Highly suggestive SNPs (P < 10-7) identified per biomarker are shown (dots) along with the associated biomarker and genes assigned to these variants (italicized). Each biomarker-gene pair represents an independent genomic region and in cases where this region contains multiple genes, one representative gene is used for annotation. The y-axis specifies p-value in the comparison non-pregnancy study, where the purple line is P = 10-7. Loci above the purple line are shared with the comparison study. Loci below the purple line are significant uniquely in our study and have significantly different effect size estimates, except for MCP-1-ACKR2 in (c) and (d). Circles are color coded based on chromosomes. a. T1-associated SNPs compared against study 1(14) b. T1-associated SNPs compared against study 2(15) c. T2-associated SNPs compared against study 1 d. T2-associated SNPs compared against study 2.
Approximately ~50% of T1 loci and ~60% of T2 loci are shared with non-pregnancy studies (Fig 3 – loci above purple line). There is strong concordance both between the two non-pregnancy studies and also between T1 and T2 timepoints (Fig 3); if a locus is shared between T1 and study 1, it is likely also shared between T1 and study 2 as well as with T2. For instance, the region mapped to ACKR1 (a chemokine receptor gene) is significant in both timepoints and in both comparison studies, providing further support for the functional relationship between this receptor and eotaxin, GROα, and MCP-1 levels. One exception to this trend was observed with the VEGF–TMEM63B association. This region is downstream of VEGFA and previous studies have established that rs6921438 and additional nearby SNPs are associated with VEGF levels [28,29]. However, in our study, this association was significant only at very early stages of pregnancy, after which VEGF levels show a notable decline (S5 Fig), suggesting a reduced role for this variant after 50 days of gestation.
Based on comparison with these two well-powered non-pregnancy studies, approximately ~50% of our highly suggestive T1 loci and ~40% of T2 loci are putative unique loci as their p-values are greater than 10-7 threshold (Fig 3 – loci below purple line). Most of these loci are not even nominally significant (P < 0.05) in comparison studies. In order to reduce the likelihood of false conclusions about these loci, we additionally compared effect size estimates (S6 Fig), demonstrating the strong correlation of overlapping loci and near-zero estimates for unique loci. The difference between these effect size estimates was significant for all but MCP-1-ACKR2 (T2) association, therefore this locus is the only one that is categorized as neither unique nor shared. In cases where a given biomarker was available in both studies, loci categorized as unique based on study 1 were also consistently unique based on study 2. IL-7–IL7 was the only exception, where it was significant in study 2 but not study 1 (Fig 3c and 3d). Unique loci include T1- and T2-associated G-CSF–CSF3 and IL-1β–EBI3, as well as T1-specific (IFNγ–RPL3, IL-13–TMCC3) and T2-specific (MIP-1α–MAGI1, FLT-3L–CATSPERG, eotaxin–PFN2) associations. Many of the loci unique to our study are supported by multiple SNPs (Fig 3), as well as the biological plausibility highlighted here. Overall, these comparisons suggest the presence of genetic drivers unique to pregnancy.
Decidual cell expression enrichment
The placenta is an organ exclusive to pregnancy, working in concert with maternal decidua to support healthy pregnancy. Therefore, studying the gene expression patterns of cell types within maternal-fetal interface has the potential to determine whether our variants of interest are enriched near genes with high expression in particular decidual cell types to further explore pregnancy-specialized results. Because we have only maternal genetic data, we focus specifically on cells of maternal origin for interpretability.
Suryawanshi et al.[30] performed scRNA-seq on placental villi and decidua samples to identify the composition and expression patterns of cells from fetal and maternal origins, respectively. We specifically examined expression profiles of 13 cell types identified within maternal decidua, ranging from decidual stromal cells and fibroblasts to various subtypes of immune cells. Per cell type, we first ranked genes by expression level. Then, we compared the rank distribution of genes annotated to suggestive SNPs against genes annotated to SNPs in permuted GWAS results to determine whether decidual cell types show higher than expected expression of genes near our true suggestive SNPs. Decidual immune cells were relevant for T1 (natural killers) and T2 (macrophages) (Table 3). T2 and Δ-implicated genes also showed enrichment for endometrial epithelial cells and smooth muscle cells, respectively.
The maternal cell types are highly specialized versions of blood (such as decidual natural killer cells, macrophages) and uterine (such as decidual stromal cells) cells. To test whether these cell type enrichments are specific to decidua residents or representing a typical expression profile of origin cells, we repeated this analysis using the single cell resource of the Human Protein Atlas [31]. The enrichments seen for decidual cell types were stronger than for corresponding cell types; in all cases decidua-specific enrichment was more significant in permutation testing (Table 3). It thus appears that maternal genetic drivers of immune biomarkers during pregnancy may be specific to decidual cell types.
Pregnancy complication overlap with immune biomarker genetics
We also explored whether the genetically-determined immune environment during pregnancy might influence the emergence of pregnancy complications. Based on the literature curation in Barbitoff et al., we first examined genes previously associated with the following pregnancy complications [32]: placental abruption (PA), preeclampsia (PE), gestational diabetes mellitus (GDM), and preterm birth (PTB). Across all four conditions, there were 34 genes overlapping with genes annotated to suggestive SNPs (Fig 4) (PA: 6 of 42, PE: 13 of 284, GDM: 12 of 170, PTB: 11 of 233). PA and GDM genes showed significant overlap with top immune biomarker loci (Fisher’s exact test P = 0.0042 and P = 0.037, respectively).
Four different pregnancy complications are color coded in these plots for T1 (a), T2 (b), and Δ (c): preterm birth (PTB) – green, preeclampsia (PE) – orange, gestational diabetes mellitus (GDM) – blue, and placental abruption (PA) – pink. Per condition, the outermost layer lists biomarkers of interest identified based on two separate analyses as follows. 1: The middle layer includes gene name for genes annotated to an associated SNP (P < 10-6) for that time variable that were previously shown to be associated with the given condition, with the corresponding biomarker in the outer ring. Genes that are associated with multiple conditions are included in all relevant conditions and shaded with the colors of all. 2: Darker “GWAS” sections in the middle layer are attached to the biomarkers that showed significant enrichment of suggestive PE, GDM, or PTB previous GWAS SNPs (P < 5x10-5) within the nominally significant SNPs (P < 0.05) of that biomarker. (Asterisk indicates significant after multiple testing correction.) No GWAS section is included with PA as comparison against an existing study wasn’t performed, and no PA layer is found in T1 as there was no overlap between PA and T1-associated genes.
For a more biomarker-specific view into genetic overlap between immune biomarkers and pregnancy complications, we performed comparisons against GWAS summary statistics of PE [33], GDM [34], and PTB [35]. We investigated whether suggestive SNPs (P < 5x10-5) in these studies were enriched within nominally significant SNPs (P < 0.05) of a given biomarker in comparison to permuted immune biomarker GWAS results. With PE, 5 biomarkers at T1, 4 at T2, and 4 for Δ were empirically significant for enrichment. IFNγ and G-CSF results are especially of interest since they are associated with PE genes (LAMC1 and HLA-DQA1, respectively). The GDM comparison identified 4 biomarkers at T1, 1 at T2, and 2 for Δ. IL-12p70 was the only biomarker also associated with a known GDM gene (RXRA). Finally, the PTB comparison identified 7 significantly enriched biomarkers at T1 and 6 at T2. Three of these also have suggestive associations with PTB-associated genes: G-CSF (HLA-DQA1, CSF3), IL-7 (IL7), and IL-4 (HLA-DQA1). A number of these enrichment results remain significant upon correction for multiple testing (5 T1, 2 T2, 3 Δ) (Fig 4). This exploratory comparison highlights the potential polygenic overlap between immune biomarkers and pregnancy complications and nominates biomarkers that may be of interest for further detailed assessment.
Due to the small size and the diversity of the IMPaCT dataset, as well as the differing overlap with pregnancy complication GWAS, we were unable to perform comprehensive co-localization analysis to follow up gene overlap. Due to the limited strong associations for each biomarker, we were also unable to perform multiple marker Mendelian Randomization analysis to further investigate genome-wide enrichment results. Even though more thorough future studies will be required to investigate causality, our preliminary observations raise the possibility that genetic determinants of immune biomarkers could also be associated with the development of pregnancy complications.
Discussion
In this study, we investigated how genetics impact maternal immune biomarkers at different stages of pregnancy and within a diverse cohort of individuals. Focusing on a large set of biomarkers allowed us to ‘survey’ the maternal immune environment at two timepoints, identify patterns in how genetic variation acts on immune biomarkers during pregnancy, and prioritize specific variants and loci that may have unique roles during pregnancy. Additionally, by studying the changes observed between two measurements, we identified distinct genetic determinants associated with fluctuations of biomarker levels between two timepoints.
Estimating SNP-based heritability identified 19 different heritable biomarkers, highlighting the contribution of genetic variation to the differences in immune biomarker levels observed across individuals. Interestingly, most biomarkers that showed evidence of heritability had a significant heritability estimate for only one time variable. This could imply that different genetic regulatory mechanisms are important (in comparison to non-genetic effects) at different timepoints in pregnancy. For instance, MIP-1α has a significant estimate only at T2 and it is associated with two loci that are GWS only in T2, corroborating the presence of differing genetic effects at different points in pregnancy. Additionally, these diverging estimates are driven by differences in estimated genetic variance and not phenotypic variance. In the majority of these cases, the non-significant timepoints have a near-zero genetic variance estimate, which we highlight as the higher confidence subset in terms of timepoint-specific heritability. This timepoint difference is surprising in light of the strong correlations between T1-T2 measurements of these biomarkers (0.78 < r < 0.90), and thus genetics would be acting in particular on the smaller independent portion of variation. However, we cannot disregard the possibility that limited power led to differences in detection due to noisy estimates near the threshold of significance, resulting in false negatives and limiting our ability to test the significance of differences between T1 and T2.
One of the strengths of our study is measurements from two timepoints in pregnancy per individual, enabling us to investigate distinct trends in early versus mid-pregnancy. Within associations categorized as unique to pregnancy, MIP-1α–MAGI1 and eotaxin–PFN2 were also specific to T2. These genes have previously been associated with pregnancy complications [36,37], suggesting that they could have specific roles during pregnancy. On the other hand, GWS SNPs shared between T1 and T2 are dominated by autoregulatory associations, such as G-CSF–CSF3, GROα–CXCL1, IL-12p40–IL12B, MDC–CCL22, and IL-27–EBI3. Variants near EBI3 are especially of interest as they were also associated with IL-1β at both timepoints. Mediation analysis showed that EBI3 variants affect IL-27, which in turn affects IL-1β. Interestingly, IL-1β – EBI3 (but not IL-27–EBI3 [23,24]) is also a putative pregnancy-specialized association, so the relationship between these biomarkers may change or strengthen during pregnancy.
Cross-referencing T1-vs-T2 specificity with unique-vs-shared with non-pregnancy categorization revealed a trend towards shared associations being significant in both timepoints, with VEGF being an interesting exception. The association between VEGF and rs6921438 has previously been explored [28,29], where one study observed enhancer-associated histone marks potentially contributing to the regulation of VEGF levels [38]. The temporal trend we observed raises the possibility of this mechanism becoming reduced in saliency after ~50 days of pregnancy. Yan et al. also identified this region to be associated with VEGF during pregnancy; however, they observed this association throughout first and second trimesters [39]. One key difference with our study is that their cohort includes a higher percentage of pregnancies with complications, such as preeclampsia and preterm birth, in which VEGF has established roles [40]. Their cohort was also focused on nulliparous women and had less genetic ancestry diversity. Detailed characterization of this locus can provide further insight into how it regulates VEGF levels during pregnancy, both in terms of temporal changes and connections to parity, genetic ancestry, or adverse pregnancy outcomes.
Another advantage of measuring two timepoints is the opportunity to study changes in the levels of these biomarkers and the extent to which genetics contributes to modulating these changes. For Δ, we identified 7 biomarkers with significant heritability highlighting a role for genetic variation in the dynamics of immune response to early-to-mid pregnancy progression, perhaps analogous to the unique genetics of response to immune challenge [17–19]. We also identified GWS associations for 9 biomarkers’ Δ. These loci are non-overlapping with T1/T2 associations and enriched in distinct biological processes, primarily developmental, in comparison to immune signaling enrichment in T1/T2. Another trend we observed with Δ-associated loci was the enrichment of transcription factor genes, several of which have already been shown to regulate expression of various immune biomarkers [27]. For instance, IL12B is a known target of transcription factor RXRA, which has a suggestive association with IL-12p70 [27,41]. Additionally, RXRA and other transcription factors such as MAF and RORA have known roles in progression of pregnancy and pregnancy complications [32,42,43]. These could imply transcriptional mechanisms guiding the changes in immune biomarker levels occurring between early and mid-pregnancy.
Our previous study in the EMA cohort is the only other available study of maternal genetics and immune biomarkers during pregnancy. Comparison with summary statistics showed limited overlap, which could be driven by differences between studies. The EMA cohort was smaller (N = 790), which limits its power in comparison to this study, and sampled from a southern (vs. northern) California population. It examined a subset of maternal biomarkers at only one timepoint (similar to T2). Ultimately, 6 of 20 biomarkers compared had at least one suggestive SNP also nominally significant in the current study, representing ~6% of independent loci at P < 0.05. The EMA study also included fetal genetics and biomarkers, revealing a strong influence of fetal genetics on maternal biomarkers and vice versa. One neonate locus near PLCL2 was associated with neonatal levels of MIG and 10 other biomarkers. In this current study, we also identified a MIG-associated suggestive SNP mapped to PLCL2. This could indicate overlap between maternal and fetal genetic regulation. A limitation of our current study is that we have neither fetal genetics nor neonatal biomarker measurements to assess specificity of maternal effects.
Throughout pregnancy, the placenta and its surrounding environment undergo transformations to support placentation and fetal development, including changes in the milieu of maternal immune cells in the uterus. A study on spatiotemporal dynamics of maternal-fetal interface [44] characterized a strong relationship between gestational age and maternal immune cell composition. They predominantly observed natural killer cells and T cells with immunosuppressive phenotypes at 6–8 weeks. Tolerogenic macrophages and extravillous trophoblasts became the dominant cell types by weeks 16–20. Likewise, in our data the expression signatures of decidual natural killer cells and T cells were enriched for T1 associations in contrast to macrophages and T cells at T2.
The maternal immune environment and its dysregulation have been linked to adverse pregnancy conditions, although the extent to which genetic factors causally contribute to this relationship is not understood [45,46]. Our preliminary investigation of overlap between genetics of immune mediators and pregnancy complications identified 20 biomarkers enriched for PE, PTB, and GDM-associated loci. Even though we were unable to investigate causality in more detail, some of the biomarkers our genetic data identified have already been studied in these conditions. For instance, previous studies examined the relationship of IFNγ and IL-10 with preeclampsia [47,48] and the possibility of IP-10 connecting inflammatory anti-angiogenic states [49], all three of which showed an enrichment of PE loci. Previous Mendelian randomization studies support a putative causal relationship implied by our data between IL-7 and IL-12p70 levels and GDM [50,51]. We observed enrichment of PE, PTB, and GDM-associated loci in IL-4, and disruption of its anti-inflammatory role in pregnancy has been associated with adverse outcomes [52]. G-CSF has roles in embryo implantation [53] and is being explored as a therapeutic option for recurrent pregnancy loss [54]. It showed an enrichment of PE and PTB loci, along with associations with known pregnancy complication genes CSF3 and HLA-DQA1, suggesting that genetic autoregulation could be critical in maintaining healthy levels of G-CSF. However, it’s worth noting that there were also biomarkers with critical roles in maternal immune environment, such as IL-6 [55] and IL-17A [56], where we didn’t observe this genetic overlap or any GWS signals, potentially implicating non-genetic factors more strongly in these associations. Follow-up investigation could examine the extent of genetic overlap between immune biomarkers and adverse pregnancy outcomes in more detail and help positively confirm ‘environmental’ contributions for the remaining inflammatory markers and nominated pregnancy outcomes. Disruption of inflammatory protein levels could also affect the health and neurodevelopmental outcomes of the baby [57,58]. Therefore, future studies may be able to use our genetic results to disentangle causal relationships between immune dysregulation, pregnancy complications, and neurodevelopmental outcomes.
To remain unbiased and consistent while mapping SNPs to nearby genes, we used annotation methodology prioritizing distance to the nearest transcription start site. However, there may be better candidate genes in the same region. For instance, fractalkine-associated rs1051112 was assigned to TMEM121, but this SNP is also near IGHG3 which encodes an immunoglobulin heavy chain. IL-15-associated rs112022139 was mapped to SUV39H2, but nearby DCLRE1C has roles in V(D)J recombination. TGFα-associated rs10918270 was annotated to OLFML2B, though it is in the intronic region of ATF6, an important arm of unfolded protein response and ER stress which contributes to preeclampsia [59]. rs73043012, which is a pregnancy unique variant associated with T2 only, was assigned to CATSPERG, involved in sperm cell hyperactivation. Regional eQTLs act on CATSPERG and SPINT2 [60]. SPINT2 is an inhibitor of HGF activator, which is highly expressed in endometrial epithelial cells (enriched in T2-implicated genes). Thorough finemapping across GWS loci may lead to additional insights into determination of the immune repertoire in pregnancy.
Our study has limited power for detecting heritability and GWS signals. However, even considering testing of 47 biomarkers and three time variables, we observe many more GWS loci and significant heritability estimates than would be expected based on α ~ 0.05. Certain confounders may further increase noise, including sampling conditions (e.g., time of day), subclinical infections or immune disorders. The lack of pre-pregnancy or post-partum samples limited our ability to assess which variants represent pregnancy-specialized associations. As an alternative, we focused on comparisons with existing studies in the general non-pregnant population. However, because our cohort is female-only and reproductive-aged, some of the observed differences could be driven by sex, age, population (or other) differences between cohorts and not specific to pregnancy. Our data were limited to maternal genetic and biomarker measurements and did not include functional or tissue-based validation experiments. Despite these (and other) limitations, our observations provide an examination of variants associated with immune biomarkers in early and mid-pregnancy along with genetic drivers of change observed during pregnancy, nominate putative pregnancy-unique associations, and suggest a relationship between genetics of maternal immune milieu and adverse pregnancy outcomes. We hope and anticipate that future studies will further investigate our observations and mitigate some of the limitations of our study design.
Methods
Ethics statement
Study procedures were approved by the Kaiser Permanente Northern California (KPNC) Institutional Review Board. All mothers had previously provided written consent to participate in the Research Program on Genes, Environment, and Health (RPGEH) pregnancy cohort [61], including donation of a blood sample during the first and second trimesters of pregnancy and permission to access their own and their child’s KPNC electronic health records for future studies.
Study population
The detailed description of the IMPaCT (Immune and Metabolic Markers during Pregnancy and Child Development) cohort is available in Croen et al.[20]. Briefly, participants were selected from children born at Kaiser Permanente Northern California (KPNC) from January 2011 to January 2016, who survived to 2 years of age, and whose mothers received health care during the 2 years prior to delivery. The latter criterion doesn’t introduce a strong bias into the selection pool as most individuals were members for 10 years on average before delivery [61]. In December 2019, pregnancies resulting in cases of ASD (N = 339), DD (N = 1193), and GP controls (N = 904) were selected for the study. Controls were randomly sampled from children with no ASD or DD diagnoses based on birth year to match the birth year distribution of cases.
The self-reported maternal race distribution of the study participants was as follows: 45.7% white, 24.6% hispanic, 20.7% asian, 5.4% black, and 3.6% other or unknown. Population structure was further explored by computing genetic principal components (PCs) using PLINK [22] (S7a Fig).
Genotyping and immune biomarker measurements
Maternal blood samples were collected during the first and second trimesters of pregnancy. ~ 90% of individuals had plasma samples and the remaining small proportion had only serum available. The detailed description of sample preparation and genotyping procedures is available in Croen et al.[20]. In brief, genotyping was performed on the Axiom Precision Medicine Research Array (PMRA) by the Genomics Core Facility at UCSF, with a standard quality control pipeline. Out of approximately 800,000 markers that were genotyped, 727,655 markers passed quality control. Imputation was performed using the Michigan Imputation Server [62] with the 1000 Genomes Reference panel [63], increasing the total number to 30,969,804 markers.
Concentrations of the following serum and plasma biomarkers were measured using Milliplex human cytokine/chemokine/growth factor 48-plex bead kit (HCYTA-60K-PX48) following the manufacturer’s instructions: sCD40L, epidermal growth factor (EGF), eotaxin, fibroblast growth factor 2 (FGF-2), Fms-related tyrosine kinase 3 ligand (FLT-3L), fractalkine, granulocyte colony-stimulation factor (G-CSF), granulocyte macrophage colony-stimulating factor (GM-CSF), GRO1 oncogene (GROα; CXCL1), interferon alpha 2 (IFNα2), IFNγ, interleukin (IL)-1α, IL-1β, interleukin-1 receptor antagonist (IL-1RA), IL-2, IL-3, IL-4, IL-5, IL-6, IL-7, IL-8, IL-9, IL-10, IL-12p40, IL-12p70, IL-13, IL-15, IL-17A, IL-17E/IL-25, IL-17F, IL-18, IL-22, IL-27, interferon gamma-induced protein 10 (IP-10), monocyte chemoattractant protein-1 (MCP-1), MCP-3, macrophage colony-stimulating factor (M − CSF), macrophage-derived chemokine (MDC; CCL22), monokine induced by gamma (MIG; CXCL9), macrophage inflammatory protein-1 alpha (MIP-1α), MIP-1β, platelet-Derived Growth Factor AA (PDGF AA), PDGF AB/BB, regulated upon activation, normal T cell expressed and secreted (RANTES; CCL5), transforming growth factor alpha (TGFα), tumor Necrosis Factor alpha (TNFα), TNFβ, and vascular endothelial growth factor (VEGF). This commercial array of cytokines, chemokines, and growth factors was selected to provide a broad range of analytes that represented inflammatory, regulatory, chemotactic, and developmental (cell development) factors that could be important to fetal and brain development. Serum and plasma samples were run on different plates to ensure the correct assay conditions for sample substrate types.
Data processing
Data processing and analysis steps described below were performed in R (v4.4.1) unless otherwise specified. There were 2227 individuals with first (T1) and 2098 individuals with second (T2) measurements available, as we only had one measurement available for a subset of individuals. Values below the limit of detection were treated as missing values. The median of percentage of missing values across biomarkers was 1.1% for T1 and T2. However, 11 biomarkers at T1 and 10 at T2 had more than 10% missing values. This did not lead to a non-normal distribution of these biomarkers’ measurements, nor genomic inflation in QQ plots (S9 Fig). IL-3 measurements were excluded from further studies due to a high rate of missing values below the limit of detection (~85%).
Each biomarker’s T1 and T2 measurements were log-transformed to normalize the distributions. Gestational age in days when samples were collected varied across individuals and covered a wide range within first and second trimesters (S7b Fig). On average, gestational age at first sample collection was 62.9 (sd: 15) and second sample collection was 121.8 (sd: 19). To account for variation associated with this, linear regression of gestational age against log-normalized T1 and T2 measurements was used to obtain residuals. These gestational age-corrected values were then used as the final T1 and T2 values in further analyses. However, unlike other biomarkers, IL-27 and VEGF T1 measurements displayed a non-linear trend across gestational age (S8 Fig). Therefore, we used piecewise linear regression (segmented R package v2.1.4) to better account for their temporal trends. Finally, for each biomarker, values + /- 3 standard deviations from the mean were excluded as outliers. The average number of outliers excluded was 24 for T1 and 22 for T2, and the maximum number of outliers excluded was 49 for T1 and 41 for T2.
To create a variable that reflects the changes between two measurements (Δ), we computed the difference between gestational age-corrected T1 and T2 measurements. Values + /- 3 standard deviations from the mean were excluded as outliers. The average number of outliers excluded was 29 and the maximum number of outliers excluded was 67. Δ variable was computed for the subset of individuals with two measurements that were obtained at least two weeks apart from each other, corresponding to 1892 individuals. The full distribution of the difference in gestational age between two measurements is shown in S7c Fig. The average gap between two measurements was 59 days (sd: 21.2).
The corrplot R package (v0.92) was used to generate heatmaps of pairwise Pearson correlation matrices.
SNP-based heritability
We used the Restricted Maximum Likelihood (REML) model implemented in GCTA software [21] to estimate SNP-based heritability for each biomarker and time variable, accounting for the following covariates: maternal education, maternal age, maternal race, fetal sex, birth year, child outcome, measurement plate number, and the first ten PCs. Heritability estimates with P < 0.05 were defined as significant. While comparing results across time variables, h2 > 0.5 was defined as high heritability and 0.2 < h2 < 0.5 was defined as moderate.
Genome-wide association study
We performed GWAS under an additive model using PLINK software [22] per biomarker and time variable combination, accounting for the following covariates: maternal education, maternal age, maternal race, fetal sex, birth year, child outcome, measurement plate number, and the first ten PCs. We included these variables as covariates as they were significantly associated with multiple biomarker levels and the majority of them didn’t have moderate or high correlation with each other. With the Δ variable, the difference between gestational age at second and first measurements was also included as a covariate to account for the variation in gap between measurements (S7c Fig). SNPs with minor allele frequency less than 1% were excluded, the final number of markers used in the study was 10,337,656. QQ plots generated for quality control were in line with expected distributions (S9 Fig). The ggmanh R package (v1.8.0) was used to create Manhattan plots for each GWAS result (S10 Fig). METAL [64] was used to test homogeneity of effect size estimates.
We used 5x10-8 as the p-value threshold for GWS SNPs and 10-6 as a suggestive threshold, unless specified otherwise. The GREAT online portal [25] was used to annotate SNPs to nearby genes using the “single nearest gene” option with default parameters (the extension limit of 1000kb). Variants that were not annotated to any gene with this method were not included in gene-based analyses.
For mediation analysis, mediation R package (v4.5.1) was used to identify the proportion of rs4905 genotype’s effect on T1 levels of IL-1β mediated through T1 levels of IL-27 (with bootstrapping parameter “sims” set to 1000).
GWAS permutations
To create a set of permuted GWAS runs, we generated randomized matrices of biomarker levels. We first created a combined matrix of T1, T2, and Δ values along with all covariates except for 10 PCs by concatenating each individual’s corresponding values. Individual IDs were then shuffled to create a randomized matrix. This preserves any relationship between biomarkers within an individual by keeping all measurements obtained from one individual together. 100 different randomized matrices were generated, and they were then separated into T1, T2, and Δ files to perform GWAS with each permuted dataset following the same protocol described above.
Suggestive SNPs within a given biomarker’s permuted run were obtained by selecting the same number of SNPs that passed the suggestive threshold in the original result, after sorting the permuted results based on p-values. This was done separately for each individual permuted result, generating 100 lists of permuted suggestive SNPs per biomarker.
Pathway enrichment analysis
After mapping suggestive SNPs (P < 10-6) to genes, we obtained the union of genes associated with all biomarkers to create a collective set of genes for T1, T2, and Δ separately. These were used to run pathway enrichment analysis with clusterProfiler [65] (v4.12.0). MSigDB [26] was used to obtain gene sets of GO-Biological Process and GO-Molecular Function terms (v2025.1), which provide functional annotations of sets of genes (though do not take tissue context into account). We also mapped every SNP that was included in GWAS to genes in order to generate a background gene set (17612 genes) for enrichment. TFlink database [27] was used to obtain the list of genes encoding transcription factors, based on annotations obtained from small-scale experiments.
Comparison against non-pregnant population GWAS on immune biomarkers
Full summary statistics of study 1 by Ahola-Olli et al.[14] (https://www.ebi.ac.uk/gwas/publications/27989323) and study 2 by Zhao et al.[15] (https://www.ebi.ac.uk/gwas/publications/37563310) were downloaded from GWAS catalog.
For this analysis, instead of using the suggestive threshold of 10-6, we tested a range of p-value thresholds to identify a value that would maximize the coverage of shared associations and limit the number of possible false positive unique associations. P < 10-7 provided an optimal solution, as relaxing the threshold beyond this value did not lead to the inclusion of any additional shared associations. Therefore, this threshold was used to identify the list of highly suggestive SNPs in both our study and comparison studies.
The following steps were performed per biomarker with a corresponding summary statistic available in study 1, separately for T1 and T2. We first clumped highly suggestive SNPs (P < 10-7) using PLINK [22] (with default parameters, except for clump-r2 = 0.2 and clump-kb = 500). We then identified the subset of index SNPs that is also present in study 1. If the index SNP itself was not present but at least one other SNP within the clump was found in the other study, the given clump was counted as an overlapping clump. Afterwards, we used the p-value of these index SNPs in study 1 to divide them into the “shared” or “unique” category based on the same P < 10-7 threshold. In cases where one independent genomic region contains multiple index SNPs, this categorization was done based on the index SNP with the lowest study 1 p-value. This process was then repeated for study 2.
The Manhattan plots include all highly suggestive SNPs that are present in both our and comparison study and additionally highlight the corresponding index SNPs of each clump. Genes assigned to each region are also annotated to the biomarkers they are associated with (where one representative gene is selected for regions spanning multiple genes), matching the color code of corresponding chromosomes. These plots were generated using ggmanh R package (v1.8.0). For each SNP included in these Manhattan plots, we also obtained effect size estimates and their standard error from studies 1 and 2 to plot them against the corresponding effect size estimates from immune biomarker GWAS. Study 1 effect size and standard errors were reported in SD-scaled units and therefore, we also standardized our effect size and standard errors when comparing against study 1.
An additional comparison was performed against our previous EMA study with a pregnancy specific cohort [16]. These were done only with T2 results, as EMA cohort’s sample collection time frame was within the second trimester. We focused on 20 of the biomarkers that were available in the maternal EMA dataset. We then identified the list of independent loci below a suggestive threshold (P < 10-5) in EMA that are also nominally significant (P < 0.05) in T2 (6 of 99).
Comparison against decidual gene expression
Suryawanshi et al.[30] performed scRNA-seq on first trimester placenta villi and decidua, identifying 9 fetal and 13 maternal cell types, respectively; we focused specifically on the maternal cells. For each cell type, we first ranked genes based on their expression level (with highest expression getting the first rank) to make results across cell types more comparable. Per time variable, we then identified the rank distribution of genes assigned as nearest to suggestive SNPs and determined the rank that corresponds to the 10th percentile of this distribution. This rank was used as a threshold to compare against each permuted GWAS result. We computed the number of times where a permutation’s percentage of genes with ranks lower than this threshold was greater than 10% to estimate a permutation p-value.
We downloaded “RNA expression levels per gene and cell type” dataset from Human Protein Atlas [31] website and repeated the analysis described above with the cell types that most closely resemble the origin of decidual cell types.
Comparison against pregnancy complication studies
Barbitoff et al.[32] curated genes associated with various pregnancy complications based on multiple sources. We specifically focused on the Public Health Genomics and Precision Health Knowledge Base (PHGKB v6.2.1) HuGE Navigator curation. Using the union of annotated genes per time variable, we identified the number of genes known to be associated with PA, PE, GDM, and PTB within the full set of curated genes.
To perform comparisons against GWAS related to these complications, we obtained full summary statistics of studies focusing on PE by Tyrmi et al.[33] (https://www.ebi.ac.uk/gwas/publications/37285119), GDM by Elliott et al.[34] (https://www.ebi.ac.uk/gwas/publications/38182742), and PTB by Sole-Navais et al.[35] (egg-consortium.org/Gestational-duration-2023.html). The following steps were performed against each of these studies per biomarker and time variable combination. We first identified suggestive SNPs (P < 5x10-5) in the comparison study that are also available in our study and clumped these SNPs. Afterwards, we computed the percentage of these independent loci that also have nominally significant (P < 0.05) associations with a given biomarker. This step is then repeated with 100 permuted GWAS results to estimate significance for the level of overlap observed.
Supporting information
S1 Table. SNP-based heritability estimates for full cohort and T1/T2 subset.
a. The table includes biomarkers where heritability estimate was significant in T1 but not in T2. Per biomarker, three rows are listed under Source column: V(G): estimated genetic variance, Vp: estimated phenotypic variance, V(G)/Vp: heritability estimate. SE columns list standard errors of these estimates. The left half of the table includes estimates from full T1 and T2 datasets. The right half with T1 Subset/T2 Subset includes estimates from repeated analyses with individuals that have both T1 and T2 measurements. Near-zero heritability estimates are shown in bold. b. Same as (a), with biomarkers where heritability estimate was significant in T2 but not in T1.
https://doi.org/10.1371/journal.pgen.1012204.s001
(PDF)
S2 Table. Δ-associated SNPs (P < 10-6) annotated to transcription factors.
If there are multiple SNPs annotated to the same gene, only the most significant SNP is displayed.
https://doi.org/10.1371/journal.pgen.1012204.s002
(PDF)
S1 Fig. Biomarker correlation.
a-c. Pairwise Pearson correlation matrices of T1 (a), T2 (b), and Δ (c). T1 heatmap (red - positive correlation; blue – negative correlation) is ordered based on hierarchical clustering, and biomarkers in T2 and Δ heatmaps are presented in the same order for comparison. The yellow squares highlight 11 highly correlated biomarkers, the purple squares highlight sCD40L-EGF-IL-7. d-f. Pairwise correlation matrices of T1 & T2 (d), T1 & Δ (e), and T2 & Δ (f). T1 labels are shown in black, T2 in green, and Δ in purple.
https://doi.org/10.1371/journal.pgen.1012204.s003
(PDF)
S2 Fig. Mediation analysis of effects of EBI3 variant on IL-1β mediated through IL-27.
Average causal mediation effect (ACME) corresponds to indirect effect mediated through IL-27 (estimate: -0.14, 95% CI: -0.17, -0.11). Average direct effect (ADE) corresponds to rs4905’s direct effect on IL-1β (estimate: -0.038, 95% CI: -0.12, 0.04). Total effect is the combined effect of the two (estimate: -0.18, 95% CI: -0.25, -0.11). Proportion of mediated effects is 0.78.
https://doi.org/10.1371/journal.pgen.1012204.s004
(PNG)
S3 Fig. Neurodevelopmental outcome sensitivity analysis.
Effect size estimates of T1 (a), T2 (b), and Δ (c) GWS SNPs are plotted against estimates obtained when GWAS is repeated with GP-only samples, excluding pregnancies with ASD or DD outcomes. Correlation coefficients are displayed on the plots.
https://doi.org/10.1371/journal.pgen.1012204.s005
(PNG)
S4 Fig. Gene Ontology enrichment.
Significantly enriched GO-Biological Process pathways within each time variable’s set of query genes assigned to nearby suggestive SNPs (P < 10-6). Categories are ordered based on adjusted p-values and circles are color coded based on adjusted p-values. Count in x-axis denotes the number of genes in the query set that belong to the category. Circle sizes are proportional to gene ratio, which denotes the ratio of the number of genes in the query set in that category to the total number of genes in the category.
https://doi.org/10.1371/journal.pgen.1012204.s006
(PNG)
S5 Fig. VEGF-rs6921438 association by gestational age.
Log-normalized VEGF level from each sample is plotted against the corresponding gestational age. Red line indicates 90 days, the threshold between first and second trimesters. Dashed purple line indicates 50 days. T1 data were divided based on this 50-day timepoint and p-values of the association between rs6921438 and VEGF levels within each group are shown in blue and orange. The p-values of the association between rs6921438 and VEGF levels within full T1 and T2 datasets are shown in red.
https://doi.org/10.1371/journal.pgen.1012204.s007
(PNG)
S6 Fig. Effect size estimates compared against study 1 and study 2.
Each SNP shown in Fig 3 is included here where x-axis shows the effect size estimates of the comparison study and y-axis shows the effect size estimates of T1/T2. Purple dots correspond to shared SNPs and green dots correspond to unique SNPs. Darker shades represent the most significant index SNP of each independent genomic region. Study 1 comparisons (a and c) show estimates in SD units. a. T1-associated loci compared against study 1 b. T1-associated loci compared against study 2 c. T2-associated loci compared against study 1 d. T2-associated loci compared against study 2.
https://doi.org/10.1371/journal.pgen.1012204.s008
(PNG)
S7 Fig. Data description.
a. Genetic principal components PC1 and PC2 are plotted to visualize genetic ancestry. Dots are colored based on self-reported maternal race/ethnicity. b. Histogram of gestational age in days at sample collection. c. Histogram of differences in gestational age between first and second measurement collection.
https://doi.org/10.1371/journal.pgen.1012204.s009
(PNG)
S8 Fig. Representative plots of biomarker levels across gestational age.
Log-normalized immune biomarker levels from each sample are plotted against the corresponding gestational age at sample collection. Dashed red line indicates 90 days, the threshold between first and second trimesters. A subset of biomarkers is selected to represent biomarkers with stable levels across time (a), biomarkers with linear relationship with gestational age (decreasing over time) (b), and biomarkers with non-linear relationship with gestational age (c). The fitted linear line is shown in blue to highlight these different trends.
https://doi.org/10.1371/journal.pgen.1012204.s010
(PNG)
S9 Fig. QQ-plots for all 141 GWAS.
QQ-plots for each T1 (a), T2 (b), and Δ (c) GWAS.
https://doi.org/10.1371/journal.pgen.1012204.s011
(PDF)
S10 Fig. Manhattan plots for all 141 GWAS.
Manhattan plots for each T1 (a), T2 (b), and Δ (c) GWAS. Dashed red line represents P = 5x10-8 and dashed blue line represents P = 10-6.
https://doi.org/10.1371/journal.pgen.1012204.s012
(PDF)
Acknowledgments
We thank Susan Fisher and Andrea Sandoval for their helpful comments and contributions.
References
- 1. Lash GE. Molecular cross-talk at the feto-maternal interface. Cold Spring Harb Perspect Med. 2015;5(12):a023010. pmid:26385089
- 2. Erlebacher A. Immunology of the maternal-fetal interface. Annu Rev Immunol. 2013;31:387–411. pmid:23298207
- 3. Colamatteo A, Fusco C, Micillo T, D’Hooghe T, de Candia P, Alviggi C, et al. Immunobiology of pregnancy: from basic science to translational medicine. Trends Mol Med. 2023;29(9):711–25. pmid:37331882
- 4. Mor G, Aldo P, Alvero AB. The unique immunological and microbial aspects of pregnancy. Nat Rev Immunol. 2017;17(8):469–82. pmid:28627518
- 5. Dutta S, Sengupta P, Liew FF. Cytokine landscapes of pregnancy: mapping gestational immune phases. Gynecol Obstet Clin Med. 2024;4(1):e000011.
- 6. Arck PC, Hecher K. Fetomaternal immune cross-talk and its consequences for maternal and offspring’s health. Nat Med. 2013;19(5):548–56. pmid:23652115
- 7. Yang F, Zheng Q, Jin L. Dynamic function and composition changes of immune cells during normal and pathological pregnancy at the maternal-fetal interface. Front Immunol. 2019;10:2317. pmid:31681264
- 8. Yockey LJ, Iwasaki A. Interferons and proinflammatory cytokines in pregnancy and fetal development. Immunity. 2018;49(3):397–412. pmid:30231982
- 9. Saito S. Cytokine network at the feto-maternal interface. J Reprod Immunol. 2000;47(2):87–103. pmid:10924744
- 10. Zhang Y, Liu Z, Sun H. Fetal-maternal interactions during pregnancy: a “three-in-one” perspective. Front Immunol. 2023;14:1198430. pmid:37350956
- 11. Jin F, Liu W, Cheng G, Cai S, Yin T, Diao L. The function of decidua natural killer cells in physiology and pathology of pregnancy. Am J Reprod Immunol N Y N. 2023;90(3):e13755. pmid:37641369
- 12. Romero R, Erez O, Maymon E, Chaemsaithong P, Xu Z, Pacora P, et al. The maternal plasma proteome changes as a function of gestational age in normal pregnancy: a longitudinal study. Am J Obstet Gynecol. 2017;217(1):67.e1-67.e21. pmid:28263753
- 13. Pazos M, Sperling RS, Moran TM, Kraus TA. The influence of pregnancy on systemic immunity. Immunol Res. 2012;54(1–3):254–61. pmid:22447351
- 14. Ahola-Olli AV, Würtz P, Havulinna AS, Aalto K, Pitkänen N, Lehtimäki T, et al. Genome-wide association study identifies 27 loci influencing concentrations of circulating cytokines and growth factors. Am J Hum Genet. 2017;100(1):40–50. pmid:27989323
- 15. Zhao JH, Stacey D, Eriksson N, Macdonald-Dunlop E, Hedman ÅK, Kalnapenkis A, et al. Genetics of circulating inflammatory proteins identifies drivers of immune-mediated disease risk and therapeutic targets. Nat Immunol. 2023;24(9):1540–51. pmid:37563310
- 16. Traglia M, Croen LA, Jones KL, Heuer LS, Yolken R, Kharrazi M, et al. Cross-genetic determination of maternal and neonatal immune mediators during pregnancy. Genome Med. 2018;10(1):67. pmid:30134952
- 17. Lee MN, Ye C, Villani AC, Raj T, Li W, Eisenhaure TM, et al. Common genetic variants modulate pathogen-sensing responses in human dendritic cells. Science. 2014;343(6175):1246980.
- 18. Oelen R, De Vries DH, Brugge H, Gordon MG, Vochteloo M, single-cell eQTLGen consortium. Single-cell RNA-sequencing of peripheral blood mononuclear cells reveals widespread, context-specific gene expression regulation upon pathogenic exposure. Nat Commun. 2022;13(1):3267.
- 19. Kim-Hellmuth S, Bechheim M, Pütz B, Mohammadi P, Nédélec Y, Giangreco N, et al. Genetic regulatory effects modified by immune activation contribute to autoimmune disease associations. Nat Commun. 2017;8(1):266. pmid:28814792
- 20. Croen LA, Ames JL, Qian Y, Alexeeff S, Ashwood P, Gunderson EP. Inflammatory conditions during pregnancy and risk of autism and other neurodevelopmental disorders. Biol Psychiatry Glob Open Sci. 2024;4(1):39–50.
- 21. Yang J, Lee SH, Goddard ME, Visscher PM. GCTA: a tool for genome-wide complex trait analysis. Am J Hum Genet. 2011;88(1):76–82. pmid:21167468
- 22. Chang CC, Chow CC, Tellier LC, Vattikuti S, Purcell SM, Lee JJ. Second-generation PLINK: rising to the challenge of larger and richer datasets. Gigascience. 2015;4(1):s13742-015-0047–8.
- 23. Folkersen L, Fauman E, Sabater-Lleal M, Strawbridge RJ, Frånberg M, Sennblad B, et al. Mapping of 79 loci for 83 plasma protein biomarkers in cardiovascular disease. PLoS Genet. 2017;13(4):e1006706. pmid:28369058
- 24. Folkersen L, Gustafsson S, Wang Q, Hansen DH, Hedman ÅK, Schork A, et al. Genomic and drug target evaluation of 90 cardiovascular proteins in 30,931 individuals. Nat Metab. 2020;2(10):1135–48. pmid:33067605
- 25. McLean CY, Bristor D, Hiller M, Clarke SL, Schaar BT, Lowe CB, et al. GREAT improves functional interpretation of cis-regulatory regions. Nat Biotechnol. 2010;28(5):495–501. pmid:20436461
- 26. Liberzon A, Subramanian A, Pinchback R, Thorvaldsdóttir H, Tamayo P, Mesirov JP. Molecular signatures database (MSigDB) 3.0. Bioinformatics. 2011;27(12):1739–40.
- 27. Liska O, Bohár B, Hidas A, Korcsmáros T, Papp B, Fazekas D, et al. TFLink: an integrated gateway to access transcription factor–target gene interactions for multiple species. Database. 2022;2022:baac083.
- 28. Debette S, Visvikis-Siest S, Chen M-H, Ndiaye N-C, Song C, Destefano A, et al. Identification of cis- and trans-acting genetic variants explaining up to half the variation in circulating vascular endothelial growth factor levels. Circ Res. 2011;109(5):554–63. pmid:21757650
- 29. Choi SH, Ruggiero D, Sorice R, Song C, Nutile T, Vernon Smith A, et al. Six novel loci associated with circulating VEGF levels identified by a meta-analysis of genome-wide association studies. PLoS Genet. 2016;12(2):e1005874. pmid:26910538
- 30. Suryawanshi H, Morozov P, Straus A, Sahasrabudhe N, Max KEA, Garzia A. A single-cell survey of the human first-trimester placenta and decidua. Sci Adv. 2018;4(10):eaau4788.
- 31. Karlsson M, Zhang C, Méar L, Zhong W, Digre A, Katona B. A single–cell type transcriptomics map of human tissues. Sci Adv. 2021;7(31):eabh2169.
- 32. Barbitoff YA, Tsarev AA, Vashukova ES, Maksiutenko EM, Kovalenko LV, Belotserkovtseva LD, et al. A data-driven review of the genetic factors of pregnancy complications. Int J Mol Sci. 2020;21(9):3384. pmid:32403311
- 33. Tyrmi JS, Kaartokallio T, Lokki AI, Jääskeläinen T, Kortelainen E, Ruotsalainen S, et al. Genetic risk factors associated with preeclampsia and hypertensive disorders of pregnancy. JAMA Cardiol. 2023;8(7):674–83. pmid:37285119
- 34. Elliott A, Walters RK, Pirinen M, Kurki M, Junna N, Goldstein JI, et al. Distinct and shared genetic architectures of gestational diabetes mellitus and type 2 diabetes. Nat Genet. 2024;56(3):377–82. pmid:38182742
- 35. Solé-Navais P, Flatley C, Steinthorsdottir V, Vaudel M, Juodakis J, Chen J, et al. Genetic effects on the timing of parturition and links to fetal birth weight. Nat Genet. 2023;55(4):559–67. pmid:37012456
- 36. Chen K-R, Chu S-K, Kuo P-L. Barrier genes are associated with preterm birth. Front Med (Lausanne). 2025;12:1580877. pmid:40625359
- 37. Liu H, Wang X. MiR-200b-3p is upregulated in the placental tissues from patients with preeclampsia and promotes the development of preeclampsia via targeting profilin 2. Cell Cycle. 2022;21(18):1945–57. pmid:35613309
- 38. Stathopoulou MG, Bonnefond A, Ndiaye NC, Azimi-Nezhad M, El Shamieh S, Saleh A, et al. A common variant highly associated with plasma VEGFA levels also contributes to the variation of both LDL-C and HDL-C. J Lipid Res. 2013;54(2):535–41. pmid:23204297
- 39. Yan Q, Zhang Y, Blue NR, Truong B, Guerrero RF, Honigberg MC, et al. Genetic associations with placental and pregnancy proteins in maternal serum identify biomarkers for hypertension in pregnancy. Am J Obstet Gynecol. 2026;234(3):779–815. pmid:41183724
- 40. Guo X, Yi H, Li TC, Wang Y, Wang H, Chen X. Role of vascular endothelial growth factor (VEGF) in human embryo implantation: clinical implications. Biomolecules. 2021;11(2):253. pmid:33578823
- 41. Gynther P, Toropainen S, Matilainen JM, Seuter S, Carlberg C, Väisänen S. Mechanism of 1α,25-dihydroxyvitamin D(3)-dependent repression of interleukin-12B. Biochim Biophys Acta. 2011;1813(5):810–8. pmid:21310195
- 42. Holdsworth-Carson SJ, Permezel M, Rice GE, Lappas M. Preterm and infection-driven preterm labor: the role of peroxisome proliferator-activated receptors and retinoid X receptor. Reproduction. 2009;137(6):1007–15. pmid:19279199
- 43. Qiu C, Gelaye B, Denis M, Tadesse MG, Luque Fernandez MA, Enquobahrie DA, et al. Circadian clock-related genetic risk scores and risk of placental abruption. Placenta. 2015;36(12):1480–6. pmid:26515929
- 44. Greenbaum S, Averbukh I, Soon E, Rizzuto G, Baranski A, Greenwald NF, et al. A spatially resolved timeline of the human maternal-fetal interface. Nature. 2023;619(7970):595–605. pmid:37468587
- 45. Mladenić T, Barišić A, Pereza N, Ostojić S, Peterlin B, Dević Pavlić S. Maternal genetic risk factors for spontaneous preterm birth: a systematic review and meta‐analysis. Int J Gynecol Obstet. 2025;169(2):458–73.
- 46. Stavros S, Panagopoulos P, Machairiotis N, Potiris A, Mavrogianni D, Sfakianakis A, et al. Association between cytokine polymorphisms and recurrent pregnancy loss: a review of current evidence. Int J Gynaecol Obstet. 2024;167(1):45–57. pmid:38706379
- 47. Kumar A, Begum N, Prasad S, Agarwal S, Sharma S. IL-10, TNF-α & IFN-γ: potential early biomarkers for preeclampsia. Cell Immunol. 2013;283(1–2):70–4. pmid:23880295
- 48. Spence T, Allsopp PJ, Yeates AJ, Mulhern MS, Strain JJ, McSorley EM. Maternal serum cytokine concentrations in healthy pregnancy and preeclampsia. J Pregnancy. 2021;2021:6649608. pmid:33680514
- 49. Gotsch F, Romero R, Friel L, Kusanovic JP, Espinoza J, Erez O, et al. CXCL10/IP-10: a missing link between inflammation and anti-angiogenesis in preeclampsia? J Matern Fetal Neonatal Med. 2007;20(11):777–92.
- 50. Pan L, Hong S, Li Y, Yuan L, Zhao L, Wen J. The causal relationship between 91 inflammatory cytokines and Gestational Diabetes Mmellitus: a bidirectional two-sample Mendelian randomization study. Diabetes Res Clin Pract. 2024;216:111838. pmid:39181454
- 51. Xu J, Xu W, Song J, Wang D, Sun J. Circulating inflammatory cytokines, lipoprotein particles, and gestational diabetes risk: a mendelian randomization study. Lipids. 2025;2025(8):70002.
- 52. Chatterjee P, Chiasson VL, Bounds KR, Mitchell BM. Regulation of the anti-inflammatory cytokines interleukin-4 and interleukin-10 during pregnancy. Front Immunol. 2014;5:253. pmid:24904596
- 53. Ding J, Wang J, Cai X, Yin T, Zhang Y, Yang C, et al. Granulocyte colony-stimulating factor in reproductive-related disease: function, regulation and therapeutic effect. Biomed Pharmacother. 2022;150:112903. pmid:35430390
- 54. Zhang B, Xu J, Wei X, Zhong X. The role of granulocyte colony stimulating factor in mitigating the risk of recurrent miscarriages: a comprehensive review and meta-analysis. Reprod Sci. 2025;32(5):1379–87. pmid:39623124
- 55. Vilotić A, Nacka-Aleksić M, Pirković A, Bojić-Trbojević Ž, Dekanski D, Jovanović Krivokuća M. IL-6 and IL-8: an overview of their roles in healthy and pathological pregnancies. Int J Mol Sci. 2022;23(23):14574. pmid:36498901
- 56. Otero AM, Antonson AM. At the crux of maternal immune activation: viruses, microglia, microbes, and IL-17A. Immunol Rev. 2022;311(1):205–23. pmid:35979731
- 57. Han VX, Patel S, Jones HF, Dale RC. Maternal immune activation and neuroinflammation in human neurodevelopmental disorders. Nat Rev Neurol. 2021;17(9):564–79. pmid:34341569
- 58. Estes ML, McAllister AK. Maternal immune activation: implications for neuropsychiatric disorders. Science. 2016;353(6301):772–7. pmid:27540164
- 59. Rashid TA, Farhan SR, Khalaf AA, Sanghvi G, Uthirapathy S, Jyothi R, et al. Navigating cellular stress: endoplasmic reticulum stress and the unfolded protein response in the molecular pathogenesis of preeclampsia. Cell Biochem Biophys. 2025;83(4):4181–99. pmid:40762761
- 60. Ward LD, Kellis M. HaploReg: a resource for exploring chromatin states, conservation, and regulatory motif alterations within sets of genetically linked variants. Nucleic Acids Res. 2012;40(Database issue):D930-4. pmid:22064851
- 61. Hedderson MM, Ferrara A, Avalos LA, Van den Eeden SK, Gunderson EP, Li DK, et al. The Kaiser Permanente Northern California research program on genes, environment, and health (RPGEH) pregnancy cohort: study design, methodology and baseline characteristics. BMC Pregnancy Childbirth. 2016;16(1):381. pmid:27899076
- 62. Das S, Forer L, Schönherr S, Sidore C, Locke AE, Kwong A, et al. Next-generation genotype imputation service and methods. Nat Genet. 2016;48(10):1284–7. pmid:27571263
- 63. Fairley S, Lowy-Gallego E, Perry E, Flicek P. The International Genome Sample Resource (IGSR) collection of open human genomic variation resources. Nucleic Acids Res. 2020;48(D1):D941-7. pmid:31584097
- 64. Willer CJ, Li Y, Abecasis GR. METAL: fast and efficient meta-analysis of genomewide association scans. Bioinformatics. 2010;26(17):2190–1. pmid:20616382
- 65. Wu T, Hu E, Xu S, Chen M, Guo P, Dai Z. clusterProfiler 4.0: a universal enrichment tool for interpreting omics data. The Innovation. 2021;2(3):100141.