Genetic effects on the timing of parturition and links to fetal birth weight

The timing of parturition is crucial for neonatal survival and infant health. Yet, its genetic basis remains largely unresolved. We present a maternal genome-wide meta-analysis of gestational duration (n = 195,555), identifying 22 associated loci (24 independent variants) and an enrichment in genes differentially expressed during labor. A meta-analysis of preterm delivery (18,797 cases, 260,246 controls) revealed seven associated loci and large genetic similarities with gestational duration. Analysis of the parental transmitted and nontransmitted alleles (n = 136,833) shows that 15 of the gestational duration genetic variants act through the maternal genome, whereas 7 act both through the maternal and fetal genomes and 2 act only via the fetal genome. Finally, the maternal effects on gestational duration show signs of antagonistic pleiotropy with the fetal effects on birth weight: maternal alleles that increase gestational duration have negative fetal effects on birth weight. The present study provides insights into the genetic effects on the timing of parturition and the complex maternal–fetal relationship between gestational duration and birth weight.


Introduction
In humans, similar to mammals broadly, the timing of delivery is crucial for neonatal survival and health. Preterm delivery is the world-leading direct cause of death in neonates and children under five years of age 1 . While the rate of neonatal mortality has substantially decreased in recent years, the reduction attributable to preterm delivery is one of the lowest among the major causes of mortality 2 . This partly reflects the relatively A full list of affiliations appears at the end of the paper. poor knowledge of the processes governing the timing of delivery in humans. Parturition may be initiated by a diversity of biological and mechanical pathways. Some of these are part of the physiological timing process, while others may override pregnancy maintenance with fail-safe mechanisms (e.g., in the case of uterine infection) 3 . The diversity of the mechanisms has led to the conceptualization of preterm delivery as a syndrome 4 , with various pathophysiological processes contributing to its etiology. Diversity is echoed at the transcriptomic level, with differentially expressed genes showing considerable heterogeneity between clinical subtypes and tissues 5 . To date, studies of genetic variation have not reported distinct effects on different phenotypes of the timing of parturition.
Gestational duration is the major determinant of birth weight (i.e., the longer the gestation, the heavier the newborn). Yet, uterine load is one of the known triggers of parturition 6 , evidenced by half of twin pregnancies delivering preterm 7 . In essence, both shared as well as distinct factors shape the interplay between gestational duration and birth weight. Shared factors include the maternal and fetal genomes, as revealed in recent genome-wide association studies (GWAS) [8][9][10][11] . Over evolutionary time, the maternal and fetal genomes may have even conflicted on gestational duration and birth weight, as proposed in the hypothesis of the genetic conflicts of pregnancy 12 . This hypothesis suggests that the maternal genome favors slightly shorter gestations and lower birth weight while the fetal genome favors the opposite. Co-adaptation theory, instead, suggests that maternal and fetal genomes may invest resources to achieve an optimal gestational duration or birth weight that increases fitness 13 . These known contributions, potential conflicts, and coadaptation of gestational duration and birth weight may ultimately shape the association between the two birth outcomes.
How distinct are the maternal genetic effects on gestational duration and preterm delivery? What is the relationship between fetal growth and gestational duration? Is there evidence suggesting maternal-fetal co-adaptation on these traits? To address these questions, we conducted a GWAS meta-analysis of gestational duration and preterm and post-term delivery in >190,000 maternal samples with spontaneous onset of delivery. We further analyzed these results using the parental transmitted and non-transmitted alleles in >135,000 parent-offsprings.

Genome-wide association analyses
We conducted a GWAS meta-analysis of gestational duration in 195,555 women of recent European ancestry, a four-fold increase in sample size compared to the largest published maternal GWAS of gestational duration 8 . After quality control, genetic variants at 22 loci were associated with gestational duration at genome-wide significance (Figure 1, Table S1, and Figure  S1). Approximate Conditional and Joint (COJO) analysis revealed two conditionally independent signals at EBF1 and KCNAB1 gene regions. Sixteen of the loci did not overlap with any previously reported gestational duration-associated locus 8 . Effect sizes were relatively small, ranging from 7 (HIVEP3/ EDN2) to 27 (MRPS22) hours of gestation per allele (average duration of gestation = 282 days, 40.3 weeks).
To prioritize candidate causal genes, we performed colocalization analysis 14 with cis-expression quantitative trait loci (eQTL, within 250 kb of the gene body) in induced pluripotent stem cells (Table S2) 15 . Colocalization analysis revealed six loci (OPRL1, ZBTB38, RGS19, TET3, DGUOK-AS1 and COL27A1) with strong evidence of sharing the causal SNP with eQTLs for protein coding genes (posterior probability > 0.90, except for COL27A1, with a posterior probability = 0.88). illustrating the GWAS discovery for gestational duration (top) and preterm delivery (bottom). Each genome-wide significant locus is labeled by their nearest protein-coding gene. Blue, previously identified locus; pink, newly identified locus. (B) Clustering of the effect origin for the index SNPs for gestational duration using transmitted and non-transmitted parental alleles (n= 136,833). Numbers depicted above the heatmap are the highest probability observed for that SNP and group names define the cluster to which the highest probability refers to. Heatmap represents effect size and effect direction for the parental transmitted and non-transmitted alleles. For comparison purposes, the maternal alleles with positive effects were chosen as reference alleles. Three major groups were identified according to the highest probability: maternal only effect, fetal only effect and maternal and fetal effect. Within variants with both maternal and fetal effects two clusters were observed: same ("SD") or opposite ("OD") effect direction from maternal and fetal genomes. One of the fetal effects was further clustered as having a parent-of-origin effect ("PoE"), specifically, an effect from the maternal transmitted allele.
The nearest protein coding genes to the 24 independently associated SNPs were > 3 times more likely to be loss-of-function intolerant genes (Fisher exact test p-value = 4.5×10 -3 ; expected rate = 0.16; observed rate = 0.41). We observed no enrichment in genes deemed to follow autosomal recessive inheritance or autosomal dominant inheritance ( Figure S2).
RNA tissue-specific enrichment highlighted the endometrium and other reproductive and smooth muscle tissues (Figure S3), which suggests that the effects of these loci on gestational duration occur at late stages of pregnancy and/ or during labor. Nonetheless, previous evidence suggests a genome-wide enrichment of genes differentially expressed after stromal cell decidualization 16 , which occurs at the very beginning of pregnancy, in the endometrium. Stratified LD-score regression ( Figure S4) revealed an enrichment in background selection, super enhancers, CpG content, H3K23ac and DNA methylation.
We also performed GWAS meta-analyses of preterm (< 37 completed weeks, controls = 260,246, cases = 18,797) and post-term delivery (> 42 completed weeks, controls = 115,307, cases = 15,972) in women of recent European ancestry ( Figure 1A, Table S1 and Figure S5-6). We observed a lower number of associated loci: 6 and 1 for preterm and post-term delivery, respectively. COJO analysis identified a secondary conditionally independent SNP associated with preterm delivery at the EBF1 gene region. Although one could expect a nearly perfect genetic correlation between gestational duration and preterm delivery, the observed estimate is modest ( Figure S7, r g = -0.62; 95% CI = -0.72, -0.51), suggesting that some differences exist in the genetic effects on the two phenotypes. Post-term delivery, instead, showed a perfect genetic correlation with gestational duration (r g = 1.17; 95% CI = 0.93, 1.41), suggesting no differences in their genetic architecture.
In concordance with the modest genetic correlation, we identified a locus associated with preterm delivery within the LRP5 gene region (rs312777, p-value = 6.6×10 -9 ) that showed weak evidence of association with gestational duration (p-value = 3.9×10 -3 ), suggesting it influences mechanisms only involved in early parturition. LRP5 encodes a transmembrane low-density lipoprotein receptor-related to Wnt-protein binding and Wnt-activated receptor activity, a pathway related to embryonic development and gestational duration through Wnt4 8 .

Resolving maternal-fetal effect origin
The genetic effects on pregnancy traits in general, and on gestational duration in particular, may be driven by two correlated genomes: the maternal and the fetal. To investigate whether the signals originate in either or both genomes, we used phased genotype data to estimate the effects of the parental transmitted and non-transmitted alleles from 136,833 parent-offspring trios or mother-child duos ( Figure 1B, Table S3 and Figure S8; the maternal samples of these parent-offspring duos/ trios were part of the discovery GWAS meta-analysis). Based on pattern similarity using Gaussian mixture model-based clustering 11 , SNPs were assigned to three large groups, including variants with fetal only effects. Of the 24 index variants, 15 had the highest probability of a maternal effect. Seven SNPs had a high probability of both maternal and fetal effects: five, with opposite effect directions, and the remaining two, with the same direction (ADCY5 and TET3/ DGUOK-AS1 gene region). Finally, two variants were grouped as having a fetal only effect; the first, independent of the parent of origin (TFAP4, probability= 0.57), the second limited to the maternal transmitted allele (EEFSEC), although caution should be taken considering its low probability (0.47). In addition, the EEFSEC locus was identified in a previous GWAS 8 (top SNP, rs2955117), yet it was only replicated at nominal significance (p-value = 2.3×10 -3 ) in a meta-analysis after excluding the data from the previous GWAS.
The index SNP at ADCY5 locus (rs28654158) had both maternal and fetal effects on gestational duration with the same effect direction. Interestingly, a SNP also located in the first intron of ADCY5 harbors maternal and fetal effects on birth weight, but in opposite directions, attributed to the fetal insulin hypothesis 10,11 . The two index SNPs for gestational duration (rs28654158) and birth weight (rs11708067) are located 50kbp apart from each other and are in low LD (R 2 < 0.2). The birth weight SNP, also implicated in diabetes, likely acts through ADCY5 17 , but it is unknown whether the gestational duration variant also acts through the same gene. Despite being physically close to each other, differences between the two loci are evident in the traits they colocalize with. The gestational duration locus also affects fat-mass-related traits, while the birth weight locus affects glucose-related ones ( Figure S9).

Gestational duration polygenic score
Owing to the fact that the maternal genetic effects on gestational duration and preterm delivery are similar, we built a polygenic score for the former in the MoBa cohort (including the X chromosome) using LDpred2 18 and estimated its effect on both traits. The polygenic score showed a significant relationship with gestational duration (beta = 1.31 days; 95% CI = 1.03, 1.58; n = 3,943), explaining 2.2% of its variance. The lowest decile had a mean gestational duration of 278 days (95% CI = 278, 279) while the highest polygenic score decile had a mean gestational duration of 283 days (95% CI = 282, 284) ( Figure 2). Analogously, the polygenic score was significantly associated with preterm delivery ( Table S4,

Genetic relationship with other female reproductive traits
To examine the shared genetic basis between the timing of parturition and other traits, we estimated the genetic correlations between 14 female reproductive traits and the maternal effects on gestational duration and preterm delivery ( Figure 3A). These estimates were generally comparable, with the latter being consistently higher. Calculated bioavailable testosterone (CBAT, r g = 0.40; 95% CI = 0.26, 0.54), testosterone (r g = 0.35; 95% CI = 0.19, 0.51) and sex-hormone binding globulin (SHBG, r g = -0.16; 95% CI = -0.27, -0.06) in women were modestly genetically correlated with preterm delivery, whereas there was little genetic correlation with levels of the same hormones in men (Table S5). We observed a positive genetic correlation between preterm delivery and the number of live births and while this may be counter-intuitive, it is in line with a positive genetic correlation reported between miscarriage and the number of live births 19 . The genetic correlation between preterm delivery and the number of live births was twice as high in cohorts where the women's whole reproductive history was available (r g = 0.27; 95% CI = 0.11, 0.43) compared to cohorts based on a random pregnancy (r g = 0.13; 95% CI = 0.00, 0.26), indicating an increased probability of preterm delivery with an increasing number of live births. We also detected a negative genetic 5 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 5, 2022. ; correlation with age at first birth and age at menopause.

3.
Genetic correlations between gestational duration and preterm delivery and other female reproductive traits. (A) Genetic correlations using LD-score regression. Dots are the genetic correlation estimate and error bars, the 95% CI. The direction of the genetic correlations with preterm delivery was flipped so that term deliveries were considered as cases and preterm deliveries as controls. Hence, the direction of the genetic correlations of preterm delivery matches that of gestational duration, providing a clear comparison of the 95% CI. Pink, preterm delivery; blue, gestational duration. (B) Latent causal variable analysis with female reproductive traits that were significantly associated (p-value < 0.05 / 14) with preterm delivery or gestational duration. The genetic causality proportion is only shown for traits where GCP was significantly different from 0 (p-value < 0.05 / 6). Pink, preterm delivery; blue, gestational duration.
Genetic correlations can arise due to pleiotropy or due to a trait being causally upstream of the other. To distinguish these, we used a latent causal variable (LCV) 20 model between genetically correlated traits ( Figure  3B and Table S6). We observed evidence for full or nearly full genetic causality of CBAT, testosterone, and SHBG on preterm delivery, but not on gestational duration. The results from the LCV model were further supported by instrumenting the concentrations of these sex-hormones (Table S7) in a two-sample Mendelian randomization analysis (Table S8). We observed no evidence of proportional horizontal pleiotropy (MR-Egger intercept p-values ≥ 0.176). Finally, we aimed to determine whether the effect of such hormones originated in the mother or the fetus by instrumenting CBAT, testosterone and SHBG levels with a polygenic score for the parental transmitted and non-transmitted alleles in individual level parent-offspring data from Iceland and Norway (MoBa and HUNT; n = 46,105 parent-offsprings, Table S9). We observed limited evidence for an association between the polygenic scores for all three hormones and gestational duration. Nonetheless, the maternal non-transmitted alleles polygenic scores for CBAT and testosterone were nominally significantly associated with gestational duration.

Genetic association between gestational duration and birth weight
We sought to understand the genetic relationship between gestational duration and birth weight and how the interplay between the maternal and fetal genomes affect this relationship. We used published summary statistics of birth weight (limited adjustment for gestational duration) derived from four different models 9,10 : maternal effect, fetal effect, maternal only effect (adjusted by fetal 6 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted May 5, 2022. ; effects) and fetal only effect (adjusted by maternal effects). The last two models were obtained using weighted linear modeling, and provide quasi-unbiased estimates for the maternal and fetal effects, respectively. We only present the results using the summary statistics after adjusting the maternal effects on birth weight by the fetal effects and vice-versa. The fetal effects on gestational duration were obtained from a previously published GWAS 9 .
We observed a strong genetic correlation between the maternal effects on gestational duration and birth weight ( Figure S10, r g = 0.65; 95% CI = 0.54, 0.75). Conversely, neither the maternal (r g = -0.05; -0.15, 0.04) nor the fetal (r g = -0.02; 95% CI = -0.15, 0.11) effects on gestational duration were genetically correlated with the fetal only effects on birth weight. We suggest the maternal effects on birth weight are, at least partially, mediated by gestational duration, while the effects of the fetus on birth weight are not.
We next sought to understand to what extent the maternal and fetal effects on birth weight were dependent on the maternal effects on gestational duration. Using multi-trait COJO analysis 21 , we conditioned the effects on birth weight by the maternal effects on gestational duration ( Figure 4A). Conditioning the fetal effects on gestational duration was not possible due to a lack of power in the fetal GWAS 9 . After conditioning, we split the genome into approximately LD-independent regions 22 and selected the SNPs with the lowest p-value on birth weight (p-value < 5×10 -6 ) from each region. We observed a reduction in effect size in > 75% of the 87 SNPs with a suggestive maternal only effect on birth weight (median reduction = 11%, Wilcoxon rank-sum test p-value = 6.0×10 -8 ). The genes tagging the SNPs with a modest reduction in effect size after conditioning (relative difference of effect > -0.2) were enriched in GO terms related to glucose and insulin metabolism and KEGG pathways also related to insulin and type II diabetes (g:Profiler, Table  S10). This suggests glucose and insulin metabolism to be another relevant mediator of the maternal genetic effects on birth weight. Distribution of the relative difference in effect size before and after conditioning the effect on birth weight by the maternal effect on gestational duration using approximate multi-trait conditional and joint analysis. In blue, relative difference in effect sizes for the maternal only effects on birth weight before and after conditioning; in pink, relative difference in effect sizes for the fetal only effects on birth weight after conditioning. After 7 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted May 5, 2022. conditioning, we split the genome into approximately LD-independent regions and selected the SNPs with the lowest p-value on birth weight (p-value < 5×10 -6 ) from each region (n SNPs maternal effect = 87; n SNPs fetal effect = 108). (B) Scatterplot for two-sample Mendelian randomization analysis for the effect of gestational duration on birth weight (maternal effect). Each dot represents one of the gestational duration index SNPs. Effect sizes and standard errors from the index SNPs for gestational duration derived from the maternal non-transmitted alleles were obtained from the meta-analysis of parent-offspring data (n= 136,833). The maternal only effects on birth weight were extracted from a previous GWAS meta-analysis (n= 210,248). The x-axis shows the SNP effect of the maternal non-transmitted alleles on gestational duration, and the y-axis the maternal only effect on birth weight. Horizontal and vertical error bars represent the 95% CI. The gray line depicts the inverse-variance weighted method estimate, and the gray-dashed line the MR-Egger estimate. Colors represent the clustering of the SNP effects on gestational duration. (C) Scatterplot for the association between maternal effects on gestational duration and fetal effects on birth weight. Each dot represents one of the index SNPs on gestational duration. Effect sizes and standard errors from the index SNPs for gestational duration derived from the maternal non-transmitted alleles were obtained from the meta-analysis of parent-offspring data (n= 136,833). The fetal only effects on birth weight were extracted from a previous GWAS meta-analysis (n= 297,356). The x-axis shows the SNP effect of the the 95% CI. The gray line depicts the inverse-variance weighted method estimate, and the gray-dashed line the MR-Egger estimate. Colors represent the clustering of the SNP effects on gestational duration. maternal non-transmitted alleles on gestational duration, and the y-axis the fetal only effect on birth weight. Horizontal and vertical error bars represent The multi-trait COJO analysis provided no evidence of a change in fetal effect on birth weight after conditioning on gestational duration (median reduction = 0%, Wilcoxon rank-sum test p-value = 0.858).
The SNP heritability of the maternal effects on birth weight dropped by 53% after conditioning on gestational duration (p-value = 9.4×10 -7 , Table S11): approximately half of the maternal genetic effects on birth weight are mediated by the maternal effects on gestational duration. Conditioning the fetal effects on birth weight by the maternal effects on gestational duration did not affect the heritability estimates (0.03%, p-value = 0.801).
In summary, while the maternal effects observed on birth weight are partially driven by gestational duration, we found no evidence for this to be true for the fetal effects on birth weight.

Bi-directional effects between gestational duration and birth weight
It is widely accepted that longer gestations lead to heavier newborns. In turn, fetal growth has implications on the timing of delivery itself due to increased uterine load. Here, we sought to obtain causal estimates of the effect of gestational duration on birth weight and the effect of fetal growth on gestational duration. We used the index SNPs from our discovery GWAS and the effect estimates from the maternal non-transmitted alleles as genetic instruments in a two-sample Mendelian randomization analysis (Figure 4B and Figure  S11) on the maternal only effects on birth weight (derived using a weighted linear model 10 ). The maternal non-transmitted gestational duration-increasing alleles were associated with higher birth weight (beta = 0.06 z-scores per day; 95% CI = 0.05, 0.08; p-value = 1.7×10 -16 ). The estimated effect (approximately 23 g per day) is concordant with the phenotypic association between gestational duration and birth weight (25 g per day in 18,452 samples from the MoBa cohort). The LCV model confirmed a full or nearly full causal (GCP = 0.6, p-value = 0.002, Table S12) effect of gestational duration on birth weight. We observed no effect from the paternal transmitted gestational duration-increasing alleles on birth weight.
Given that we observed no genetic correlation between gestational duration and the fetal effects on birth weight, we considered the fetal contribution to birth weight to be mainly driven by its effects on fetal growth. To evaluate the impact of fetal growth on gestational duration, we used individual-level data to instrument fetal growth using 68 SNPs with fetal only effect on birth weight (n= 35,280 and 48,741 parent-offsprings, Table S13) 10 . 8 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted May 5, 2022.
Higher paternal transmitted birth weight score was associated with shorter gestational duration and the estimated effect was larger when gestational duration was estimated using the last menstrual period (beta = -1.9 days per z-score, p-value = 4.0×10 -4 ) than ultrasound. The observation of an effect limited to gestational duration estimated by last menstrual period suggests that ultrasound estimation of gestational duration, based on measurements of fetal growth during pregnancy, is heavily biased by fetal growth. We caution on the use of gestational duration estimated by ultrasound when assessing the impact of growth-affecting traits on gestational duration. This result supports previous evidence showing faster fetal growth is associated with shorter duration of gestation 23 .

Maternal gestational duration loci exhibit antagonistic pleiotropy with fetal effects on birth weight
Maternal and fetal effects on gestational duration and birth weight have likely conflicted or coadapted through evolutionary time to increase fitness in either or both of them. To study this, we borrowed methods from two-sample Mendelian randomization. Yet, the results in this section should not be interpreted under a causal framework due to the use of summary statistics derived from two distinct genomes (maternal and fetal).
We explored pleiotropy between the maternal and fetal genomes over gestational duration and birth weight, using the maternal gestational duration-increasing alleles and the fetal effects on birth weight. This may provide a measure of the fetal ability to adapt to the maternal control of gestational duration. Both maternal transmitted and non-transmitted gestational duration increasing alleles were associated with a lower fetal only effect on birth weight ( Figure 4C and Table S14, maternal transmitted: beta = -0.02 z-scores per day; 95% CI = -0.03, -0.01; p-value = 3.4×10 -4 ; maternal non-transmitted: beta = -0.01 z-scores per day; 95% CI = -0.02, -0.01; p-value = 6.2×10 -3 ). Used as a negative control, the paternal transmitted gestational duration increasing alleles were not associated with fetal only effects on birth weight.

Gestational duration loci have distinct evolutionary histories
As previously shown, gestational duration and preterm delivery are likely shaped by a mosaicism of evolutionary forces 24 . We aimed to understand the evolutionary forces shaping all regions that colocalized with eQTLs in iPSC lines (TET3/ DGUOK-AS1, ZBTB38 and OPRL1) using the MOSAIc pipeline 24 . This analysis has been previously reported for OPRL1 24 . The three genomic regions described, each have distinct evolutionary histories (Figure S12-13), except for a relatively high LINSIGHT score, which is suggestive of regulatory effects and is concordant with their colocalization with eQTLs. The intronic index SNP at the gene TET3 (rs34555419, ancestral allele frequency across populations > 0.9) showed a significant enrichment in GERP, LINSIGHT, phastCons, and phyloP. These metrics indicate that this variant occurs in a highly conserved region of the genome. The ZBTB38 gene region (index SNP: rs7650602, ancestral allele frequency = 42%) showed significant enrichment for iES, Fst between East Asian and African populations, and LINSIGHT.

Discussion
The timing of parturition is crucial for neonatal survival and health. Yet, understanding of its genetic underpinnings lags behind that of other pregnancy traits such as birth weight 10 and fetal growth 11 . In this GWAS meta-analysis of parturition timing, we identified 17 loci not previously reported, one of which was more strongly linked to preterm delivery than to gestational duration. The results further support modest differences in the genetic architecture of gestational duration and preterm delivery. By including parent-offspring data with a similar sample size to that of the discovery GWAS, we were able to discern maternal from fetal effects with high certainty in most index SNPs. Finally, the results show a complex genetic relationship between the maternal and fetal effects on gestational duration and birth weight. The maternal effects on birth weight are largely mediated by the maternal effects on gestational duration 9 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted May 5, 2022. ; https://doi.org/10.1101/2022.05.04.22274624 doi: medRxiv preprint and the maternal gestational duration increasing alleles have negative fetal effects on birth weight.
Heterogeneity at the phenotypic and transcriptomic level has been long recognized in deliveries initiated at different times of pregnancy 5,25 . The present results suggest small differences in the genetic effects on the gestational duration (average = 282 days, 40.3 weeks) and preterm delivery (< 260 days, 37 weeks) to be modest. For example, while the polygenic score of gestational duration is still inadequate to be used clinically, it had a reasonable performance on preterm delivery (AUC = 0.61). We identified one locus exclusively linked to preterm delivery, an intronic variant at the LRP5 gene (rs312777). This SNP is an eQTL for LRP5 gene and long-range chromatin interactions also suggest this gene is the likely target of rs312777 in mesendoderm cells and trophoblasts 26,27 . LRP5 is a co-receptor for Wnt signaling, which has been implicated in gestational duration.
The maternal contribution to gestational duration is known to be larger than that of the fetus 28,29 , leading us to anticipate that the identified effects were driven mainly by the maternal genome. Here we show that, even if GWAS was conducted on maternal samples, the correlation between the maternal and fetal genomes may lead to the identification of fetal or maternal effects via the maternal transmitted alleles. Using the parental transmitted and non-transmitted alleles we were able to discriminate SNPs with maternal, fetal, or both maternal and fetal effects. For this analysis, we used the largest sample of phased haplotypes in parent-offsprings assembled to date, a sub-sample of the discovery cohort with available parent-offspring data. The similar sample size to the discovery cohort is reflected in the high classification probabilities we obtained. Nonetheless, some of the loci were not classified with high probability (i.e., parent-of-origin effect at EEFSEC), and should be thus regarded with caution.
The establishment and maintenance of pregnancy are tightly regulated by several hormones. While this has been suggested to be true also for parturition initiation, the evidence in humans has been so far limited. Here, we provide support for a causal relationship between maternal sex hormones, particularly, testosterone and SHBG, and preterm delivery. Yet, the LCV model supported only a partial causal effect on gestational duration, suggesting that the effects of these sex-hormones on the timing of parturition may be driven by pleiotropy or by an intermediary trait highly genetically correlated to these two hormones.
Gestational duration is the major determinant of birth weight. While the maternal genome affects offspring birth weight through many different causal pathways (e.g., maternal glucose levels 10,11 ), approximately half of its effects on birth weight are mediated by the effects of gestational duration. This has implications on the interpretation of GWAS of birth weight and downstream analyses, such as Mendelian randomization. In contrast with this, the fetal genetic effects on birth weight are not mediated by gestational duration, suggesting that the identified fetal birth weight loci mainly act by modulating fetal growth. Interestingly, the maternal gestational duration increasing alleles were associated with lower effects on birth weight when present in the fetus. These antagonistic effects, coupled with the negative effect of fetal growth on gestational duration suggest the fetal effects on birth weight have likely coadapted to increase the fitness of the fetus in pregnancies genetically predisposed to a shorter duration. It has been suggested that both gestational duration and birth weight are under balancing selection, with intermediate values of these traits having highest fitness 3,30 . As exemplified here, this could lead to selection favoring the coadaptation of maternal and fetal effects to attain optimal gestational duration and birth weight 13 .
In conclusion, the present results provide evidence of modest genetic differences between gestational duration and preterm delivery and further our understanding of the complex relationship between gestational duration and birth weight, likely shaped by 10 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted May 5, 2022. ; https://doi.org/10.1101/2022.05.04.22274624 doi: medRxiv preprint strong evolutionary forces. Particularly, we showed that the maternal effects on birth weight are largely driven by gestational duration and that maternal and fetal effects have antagonistic pleiotropic effects on gestational duration and birth weight.

Phenotype definition
In this study we included pregnancies with a singleton live birth and a spontaneous onset of delivery: medically initiated deliveries (either by induction or planned cesarean section) were excluded or part of controls for preterm delivery. Gestational duration in days was estimated using either the last menstrual period date or ultrasound. We excluded pregnancies lasting <140 days (20 completed weeks) or >310 days (44 completed weeks), as well as women with health complications prior to or during pregnancy and congenital fetal malformations. Spontaneous preterm delivery was defined as a spontaneous delivery <259 days (37 completed gestational weeks) or by using the ICD-10 O60 code, and controls as a delivery occurring between 273 and 294 days (39 and 42 gestational weeks). Post-term delivery was defined as a delivery occurring >294 days (42 completed weeks) or ICD-10 O48 code, and controls as a spontaneous delivery between 273 and 294 days (39 and 42 gestational weeks). Given the perfect genetic correlation between gestational duration and post-term delivery GWAS, and the small power of the latter, all downstream analyses are focused on gestational duration and preterm delivery. Each individual cohort applied specific quality control procedures, data imputation and analysis independently following the consortium recommendations. Unless more stringent, samples were excluded if genotype call rate <95%, autosomal mean heterozygosity >3 standard deviations from the cohort mean, sex mismatch or major recent ancestry was other than European (HapMap central european). Genetic variants were excluded if genotype call rate <98%, Hardy-Weinberg equilibrium p-value <1×10 -6 or minor allele frequency (MAF) <1%. Reference panels for imputation were either 1000 Genomes Project (1KG) 31 , Haplotype Reference Consortium (HRC) 32 , 10KUK, or a combination of one of the mentioned reference panels and own whole-genome sequencing data (deCODE, HUNT, DBDS, and FINNGEN). Each individual cohort performed a GWAS using an additive linear regression model adjusted for, at least, genetic principal components or relationship matrix on autosomal chromosomes and chromosome X. Summary statistics for each individual cohort were stored centrally and underwent quality control procedures before meta-analysis. Genetic variant ids were converted to 'CHR:POS:REF:EFF' (positions were mapped to the Genome Reference Consortium Human Build 37, hg19), where EFF was the alphabetically higher allele -effect sizes were aligned accordingly. Alleles for insertion/deletions were coded as 'I/D', respectively. Only sequence variants from the Haplotype Reference Consortium panel or 1000 Genomes Project were included in the 11 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted May 5, 2022. ; meta-analysis and genetic variants with a MAF > 0.05%, minor allele count > 6, an imputation INFO score > 0.4, MAF +-20% compared to HRC or 1KG, and a reported p-value with a less than 10% difference with a calculated p-value (from the z-score) in the -log 10 scale were included.

Meta-analysis of genome-wide association studies
After quality control, individual-cohort GWAS summary statistics were pooled using fixed effects inverse-variance weighted meta-analysis with METAL 33 without genomic control correction. We also performed an analysis of heterogeneity of effects (Cochran's Q-test). After meta-analysis, we removed genetic variants not available for at least half the maximum sample size, resulting in 9-10 million genetic variants. LD-score regression intercepts were substantially lower than genomic inflation factors, suggesting that the inflation in test statistics was mostly due to polygenicity (Table S7). Test statistics were not further adjusted for genomic control for any of the phenotypes. Initially, we naively defined independent loci based on physical distance, where SNPs within 250 kb from the index SNP were considered to be at the same locus. Novel loci were defined as loci not overlapping previously reported gestational duration loci in the largest GWAS performed to date 8 . Finally, we used conditional analysis to resolve independent loci (see below).

Conditional analysis
We looked for conditionally independent associations within each locus using approximate Conditional and Joint (COJO) analysis 34 implemented in Genome-wide Complex Trait Analysis (GCTA) software 35 . We ran a stepwise model selection (-cojo-slct) to identify conditionally independent genetic variants at p-value< 5×10 -8 for each of the genome-wide significant loci (using a radius of 1.5 Mb from the index SNP). Overlapping loci were merged into a single locus (only two loci overlapped, at 3q23). LD between genetic variants was estimated from 19,092 maternal samples from the Norwegian Mother, Father and Child Cohort, after excluding variants with imputation INFO score< 0.4. We converted the reference panel from BGEN files to hard-called PLINK binary format (.bed). As per default in COJO, genetic variants >10 Mb apart were assumed to be in complete linkage equilibrium.

Gene prioritization
To prioritize genes at the gestational duration loci identified, we set the baseline as the nearest protein coding gene to the index SNP at each independent locus. While naive, this approach has been consistently shown to outperform other single metrics for locus-to-gene mapping 36,37 . Next, we performed colocalization analysis for cis-eQTLs in 1,367 human iPSC lines 15 from the i2QTL resource. We decided to use these cell types due to a higher transcriptional homogeneity than that observed in bulk tissues from, for example, post-mortem samples (e.g. GTEx). Notably, Bonder et al. show that more eGenes were identified in the iPSC lines than in GTEx, despite the considerably smaller sample size 15 .

Colocalization
We utilized genetic colocalization to identify pleiotropic effects between gestational duration and expression quantitative trait locus (eQTLs) in iPSCs (see Gene prioritization). To this end, we applied COLOC 14 , which evaluates, in a Bayesian statistical framework, whether a single locus from two different phenotypes best fits a model where the associations are due to a single shared variant. For each tested locus, this information is summarized in the posterior probability of five hypotheses. Given phenotypes A and B: -No association for any of the two phenotypes -Association with phenotype A but not with phenotype B -Association with phenotype B but not with phenotype A -Association with both phenotypes, the causal variant is not shared -Association with both phenotypes, the causal variant is shared We ran pairwise colocalization with eQTLs for all variants within the gene body (±250 kb). Prior probabilities for each for the non-null 12 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted May 5, 2022. ; hypotheses were set as suggested by Wallace (prior probabilities that a random SNP in the loci is associated with phenotype A, phenotype B, or both phenotypes, 1×10 -4 , 1×10 -4 , and 5×10 -6 , respectively), which are considered more conservative than the ones set by default 38 .
Strong evidence of colocalization was defined as a posterior probability of colocalization >0.9.

Enrichment analysis
We tested for enrichment based on top loci and genome-wide using partitioned LD-score regression. For the nearest protein coding genes to the index SNPs, we analyzed over-representation for loss-of-function intolerant genes (pLI score >0.9) 39 , genes with dominant 40,41 and recessive effects 40,41 using Fisher's exact test and also tested for over-representation in tissue-specific RNA expression using data from the Human Protein Atlas (RNA consensus tissue gene data) 42 . For the latter, we performed a Wilcoxon rank-sum test on normalized RNA for genes within our set (above-mentioned) and all other genes. Significance for this test was set at Bonferroni correction for the number of tissues (p-value< 0.05 / 61), and suggestive evidence at p-value< 0.1/ 61. At the genome-wide level, we performed partitioned heritability using LD-score regression to test for enrichment in 97 different annotations 43,44 . Pre-computed partitioned LD-scores for subjects of recent European ancestry were employed (baseline-LD model v2.2).

Genetic correlations
We estimated genetic correlations by performing LD-score regression 45 locally using pre-computed LD-scores from 1000 Genomes Project samples of recent European ancestry. The MHC region (chr6:28477797-33448354) was removed prior to running LD-score regression.

Resolving effect origin
To classify the identified index SNPs for gestational duration as having maternal, fetal or maternal and fetal origin, we performed an association analysis using the parental transmitted and non-transmitted alleles on gestational duration. Basically, we used phased fetal genotype data to infer the parent-of-origin of the genotyped/ imputed alleles as previously described 23 . For each index SNP we fit the following linear regression model: = + + + where MnT and MT refer to the maternal non-transmitted and transmitted alleles respectively, and PT refers to the paternal transmitted alleles. The latter is interpreted as a fetal only genetic effect, while the effect of the maternal non-transmitted allele is a maternal only genetic effect. We first estimated the effects of the index SNPs in each birth cohort separately; effect sizes were then combined through fixed-effect meta-analysis, totalling a sample size of 136,833 (104,962 parent-offspring trios from Iceland with at least one genotyped individual, 17,024 parent-offspring trios from the MoBa cohort, 5,122 parent-offspring trios from the HUNT cohort, and 9,725 mother-child duos from the Avon Longitudinal Study of Parents and Children (ALSPAC), Finnish birth data set (FIN), the Danish National Birth Cohort (DNBC), the Genomic and Proteomic Network for Preterm Birth Research (GPN) and the Hyperglycemia and Adverse Pregnancy Outcome (HAPO)). The analysis of the Icelandic data was done on 104,962 parent-offspring trios with at least one genotyped individual. This includes 18,165 fully genotyped trios, 5,208 with only child and mother and 1,875 with only child and father genotyped, 40,182 with both parents genotyped but not the child, and 1,627, 24,965 and 12,868 with only child, mother or father genotyped, respectively. In order to estimate paternal, maternal and non-transmitted maternal effects we used a maximum likelihood framework and estimated parameters with the EM algorithm as described in 11 . To classify the identified genetic variants into classes with similar patterns of effect we used model based clustering 11 . Variants were clustered based on estimated effects of the transmitted and non-transmitted parental alleles into five clusters. Two clusters assume fetal effect only, one with effect independent of parent-of-origin and one where the effect is limited to the maternally transmitted allele; a cluster with maternal effect only; and two clusters with 13 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted May 5, 2022. ; both maternal and fetal effects, either in opposite or same direction.

Locus pleiotropy at 3q21
After identifying locus pleiotropy between the maternal effect on gestational duration and the fetal only effect on birth weight at the ADCY5 gene region, we set out to investigate differences between the two top SNPs in their colocalization with other traits. Phenome-wide colocalization for the two regions (defined as 1.5Mb around the index SNP) was performed using summary statistics from FINNGEN (data freeze 5) and Pan UK Biobank data 46 (in subjects of recent European ancestry). We included all phenotypes available from FINNGEN (n traits= 2,803), while for Pan UK Biobank, we reduced it to summary statistics with an estimated heritability >0.01 and that were labeled as biomarkers, continuous trait or ICD-10 codes (n traits = 832). Given the exploratory nature of this analysis, despite using 3,635 phenotypes, we used a lenient posterior probability of colocalization (>= 0.75).

Female reproductive traits
We obtained summary statistics for several female reproductive traits from different sources (minimum sample size 10,000). We included summary statistics from the following traits: miscarriage 19 , gestational duration (fetal genome) 9 , age at first birth, age at menarche 47 , age at menopause 48 , number of live births 47 , testosterone 49 , CBAT 49 , SHBG 49 , oestradiol (women) 47 , pelvic organ prolapse (FINNGEN), polycystic ovary syndrome ( 50 and FINNGEN), endometriosis 47 , leiomyoma uterus (FINNGEN) and pre-eclampsia 51 . For polycystic ovary syndrome, we meta-analyzed summary statistics from the largest published GWAS 50 and FINNGEN. We estimated genetic correlations between gestational duration and preterm delivery and these traits, performed colocalization analysis at the discovered GWAS loci, and latent causal variable analysis whenever a significant genetic correlation was identified. For traits with evidence of causal association with preterm delivery or gestational duration according to the latent causal variable analysis, we further explored causality using two-sample Mendelian randomization and inspected whether the effects originated in the maternal or the fetal genome (see below "Mendelian randomization").
To obtain GWAS estimates for preterm delivery independent of the number of live births, we split the cohorts into two groups and then meta-analyzed per strata: on one side, cohorts based on a random pregnancy per mother (the probability of having at least one preterm delivery is not affected by the number of previous or subsequent deliveries) and cohorts with whole reproductive history of a woman (i.e., cohorts using life-time ICD codes or with data on > 1 pregnancy for the same mother).

Latent causal variable analysis
We used latent causal variable analysis to distinguish (partial) causation from genetic correlation 20 . For this, we used traits genetically correlated with gestational duration or preterm delivery (birth weight or female reproductive traits). A latent variable mediates the genetic correlation between two phenotypes, with a causal effect on both traits. The genetic causality proportion, which quantifies the proportion of the genetic correlation that is due to causality, is then estimated using mixed fourth moments. Whenever GCP p-values were significant after Bonferroni correction (by-case definition), we defined GCP>= 0.6 between two traits as evidence of full or nearly full genetic causality, and GCP< 0.6 as evidence of limited partial causal implication.

Gestational duration polygenic score analysis
To obtain an independent sample for training and validation of a polygenic score, the meta-analysis for gestational duration was rerun, excluding the MoBa cohort. This new meta-analysis result was used as the base dataset to calculate the polygenic score. After applying the same exclusion criteria to it as used for the study samples in the meta-analysis, and removing duplicated samples and those with a kinship of greater than 0.125, the MoBa cohort was randomly split, using 80% (n=15,768) as the training cohort and the remaining 20% (n=3,942) as the validation cohort.
14 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted May 5, 2022. ; https://doi.org/10.1101/2022.05.04.22274624 doi: medRxiv preprint QC of training genotypes dataset From the training cohort, genotypes were excluded if they had a minor allele frequency of less than 0.01, info scores < 0.7, a Hardy-Weinberg equilibrium less than 1.0×10 -6 and genotype call rates less than 0.01.
Polygenic score calculation LDpred2 was used for the calculation of the polygenic score 18 . As we wanted to include the X chromosome in the polygenic score, we mapped the genotypes to the genetic map taken from Bolt-LMM 52 . polygenic scores were calculated for a range of models using a grid of hyperparameter values; proportion of causal SNPs from 10 -5 to 1 for 21 values, and proportions of heritability of 0.7, 1, and 1.4, calculated from a constrained LD-score regression. LDpred2 also uses a third hyperparameter that allows for sparse effect size estimates (i.e. some effects are exactly 0). This resulted in a total of 126 combinations of hyperparameter values for the range of grid models 18 . The variance explained was used to decide which of the grid models was the most appropriate polygenic score. We found the polygenic score (with ten principal components and adjusted for genotyped batch) that utilised the hyperparameters of proportion of causal SNPs of 0.0032, 0.7 of the heritability, and did not allow for sparse effect size estimates, was the most appropriate for the training cohort. This model accessed weighted betas from 1,123,366 variants to explain 2.3% of the variability in the testing sample. We then extracted the weighted betas for each variant from this model to be used in the polygenic score validation.
Polygenic score validation The remaining MoBa cohort was used for validation of the polygenic score. The SNPs included from the polygenic score model that explained the most variance in the training data were extracted from the genotyped data of the validation cohort. These SNPs were then used with the weighted betas from the training model to calculate a polygenic score for each individual in the validation cohort.
To test the performance of the polygenic score, a linear regression was conducted for gestational duration by the polygenic score. A second model was used that adjusted for 5 principal components and genotyped batch. R 2 was calculated for the models to quantify variance explained.
The utility of the polygenic score for the prediction of preterm delivery was also assessed.
Gestational duration was dichotomized into preterm delivery (less than 37 weeks) or full term (greater than or equal to 39 weeks and less than 41 weeks). Two models were analyzed, one assessing just the polygenic score and a second adjusting for 5 principal components and genotype batch. Receiver operating characteristic, area under the curve were calculated for each model and used as assessment of diagnostic accuracy.

Multi-trait conditional analysis
GCTA was used to perform bi-directional multi-trait COJO (mtCOJO) 21 analysis using summary statistics. The gestational duration GWAS was conditioned on the birth weight GWAS and vice-versa, using birth weight summary statistics from the largest GWAS meta-analysis of birth weight 10 . We obtained birth weight summary statistics from four different GWAS within the EGG Consortium: using the maternal genome -offspring birth weight, the fetal genome -own birth weight and using a weighted linear model to adjust the GWAS of offspring birth weight by the fetal genome, and the GWAS of own birth weight by the maternal genome. We estimated the heritability of the birth weight GWAS before and after conditioning for gestational duration using LD-score regression 53 . This method estimates polygenic heritability as the variance explained by common genetic variants (autosomal MAF>= 0.01). To test the significance of differences between heritability estimates before and after conditioning, we calculated a z-scored as follows 54 , where is the non-conditioned estimate, β1 β2 the conditioned estimate, and and the 1 2 standard errors for the non conditioned and conditioned estimates, respectively.

15
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted May 5, 2022. ; Enrichment analysis on variants that had a modest reduction in relative effect size after conditioning (> -0.2) was performed using g:Profile 55 . We introduced SNP rsids as inputs, and parameters were left as per default (enrichments were only checked for GO terms and KEGG pathways), multiple testing correction method applying significance threshold of 0.05.

Mendelian randomization
We performed Mendelian randomization to study the effects of gestational duration (maternal) on birth weight (maternal) and the effects of fetal growth (fetal effect on birth weight) and sex hormones on gestational duration.
To study the effect of gestational duration on birth weight, we employed two-sample Mendelian randomization. The 24 index SNPs (22 autosomal SNPs) from the present gestational duration meta-analysis and the effect sizes from the parental transmitted and non-transmitted alleles analysis were used to instrument gestational duration; birth weight was instrumented using summary statistics from a previous GWAS of offspring's birth weight not adjusted by gestational duration 10 . While there is a recent publication on fetal growth 11 , this analysis was largely adjusted for gestational duration. To avoid confounding due to the correlation between the maternal and fetal genomes, we used summary statistics derived using a weighted linear model 10 . This allowed us to obtain quasi-unbiased estimates for the fetal effects on birth weight (adjusting for the maternal effect): To obtain a causal estimate, we performed an inverse-variance weighted analysis with standard errors calculated using the delta method 56 . We assessed the impact of horizontal pleiotropy on the causal estimate with MR-Egger regression. The intercept was used to determine whether the average pleiotropic effect is not statistically different from zero (p-value > 0.100). In such cases, the inverse-variance weighted method estimate is a consistent estimate of the causal effect 57 . Whenever the MR-Egger intercept is significantly different from 0, we report the estimate from the MR-Egger analysis. Both inverse-variance weighted method and MR-Egger regression were performed on R using the MendelianRandomization package 58 .
We assessed the effect of sex hormones (testosterone, SHBG and CBAT) on gestational duration using two-sample Mendelian randomization and instrumenting the hormones using a polygenic score for the parental transmitted and non-transmitted alleles. For each sex hormone, we obtained a list of independent SNPs genome-wide associated with these traits (Table S7) by performing GWAS clumping (R 2 > 0.001) using the following PLINK command: plink --bfile <1000 Genomes> --clump {GWAS summary statistics} --clump-r2 0.001 --clump-kb 1000 --clump-p1 5e-8 --clump-p2 1e-5 Such variants were used as instrumental variables in the two-sample Mendelian randomization analysis and to construct the polygenic score for the parental transmitted and non-transmitted alleles. The current meta-analysis results were employed as outcome for the two-sample Mendelian randomization analysis (inverse-variance weighted and MR-Egger). We subsequently constructed the polygenic score for the maternal transmitted and non-transmitted alleles and the paternal transmitted alleles in 46,105 parent-offsprings from Iceland and Norway. We estimated the effects of the maternal non-transmitted (MnT PGS ) and transmitted (MT PGS ) and paternal transmitted (PT PGS ) alleles polygenic score using the following linear model: Again, effects from each of the three data sets (Iceland, MoBa and HUNT) were combined using fixed-effect inverse-variance weighted meta-analysis.
To understand the impact of fetal growth on gestational duration, we used individual genetic data from 35,280 (ultrasound-gestational duration) and 48,741 (last menstrual period-gestational duration) parent-offsprings from Iceland, the MoBa cohort and HUNT. To instrument fetal growth, we used 68 SNPs with fetal only effect on birth weight as classified in Warrington et al. using Structural Equation Modeling 10 . Based on these 68 SNPs, we constructed a fetal growth polygenic score for the parental transmitted and non-transmitted alleles and regressed these on gestational duration (estimated by ultrasound or last menstrual period, separately). We estimated the effects of the maternal non-transmitted (MnT PGS ) and transmitted (MT PGS ) and paternal transmitted (PT PGS ) alleles polygenic score using the following linear model: Effect estimates from each of the three data sets (Iceland, MoBa and HUNT) were pooled using fixed-effects inverse-variance weighted meta-analysis.
Testing for maternal-fetal coadaptation between gestational duration and birth weight We further investigated what are the fetal effects on birth weight for the maternal gestational duration increasing alleles, and the maternal effects on gestational duration for the fetal birth weight increasing alleles. To study maternal-fetal coadaptation, we borrow inverse-variance weighted analysis from Mendelian randomization, but using the two effects of two distinct genomes, the maternal and fetal. We caution that this should not be interpreted under a causal framework. To understand what the maternal gestational duration-raising alleles do to birth weight when present in the fetus, we used the effect sizes and standard errors of the parental transmitted and non-transmitted alleles for the 22 autosomal index SNPs on gestational duration and assessed its effects on the same SNPs with a fetal only effect on birth weight. To understand what the fetal birth weight-raising alleles do to gestational duration when present in the mother, we used the effect sizes and standard errors of 68 autosomal SNPs associated with fetal effects on birth weight and the effect sizes and standard errors from the current maternal GWAS of gestational duration.

Evolutionary analysis
To examine the evolutionary history of three regions identified in the GWAS meta-analysis, we ran the significant variants through the MOSAIc pipeline 24 . This pipeline is designed to detect enrichment in evolutionary signals using a variety of sequence-based metrics of selection. The sequence based evolutionary measures used in this method include: 1) Beta Score which detects balanced polymorphisms to infer balancing selection 59 , 2) ARGWEAVE uses ancestral recombination graphs to infer the evolutionary origin of regions 60 , 3) GERP uses sequence conservation to infer positive and negative selection 61 , 4) LINSIGHT uses sequence conservation to infer positive and negative selection 62 , 5) phastCONS100 uses sequence conservation to infer positive and negative selection, 6) PhyloP uses substitution rate to infer positive and negative selection, 7) iES uses haplotype homozygosity to infer positive selection, 8) XPEHH uses haplotype homozygosity to detect population-specific positive selection, and 9) Fst uses population differentiation to infer local adaptation. Variants from the GWAS that passed a significance threshold (p-value <1×10 -8 ) were clumped into regions using PLINK such that the clumps of variants had an R 2 >0.9 and were within 500 kb. We then obtained 5,000 control variants matched on variant count, LD structure and minor allele frequency. The evolutionary metrics were obtained for all variants, and the maximum value was extracted for analysis. Finally, the evolutionary metrics were also obtained for the control variants and further used to create a background distribution. Then a z-score and p-value were produced for each experimental genomic region compared to its unique background distribution.

Variant annotation
Variants were annotated using Ensembl's Variant Effect Predictor (hg19) command line tool 63 . Physical coordinates of protein coding genes were obtained from the UCSC

Data availability
Cohorts should be contacted individually for access to raw genotype data, as each cohort has different data access policies. Summary statistics from the meta-analysis excluding 23andMe and the summary statistics of the top 10,000 SNPs for each phenotype will be made available at the EGG website (https://egg-consortium.org/). Access to the full set, including 23andMe results, can be obtained after approval from 23andMe is presented to the corresponding author or by completion of a Data Transfer Agreement (https://research.23andme.com/dataset-access /), which exists to protect the privacy of 23andMe participants. Access to the Danish National Birth Cohort (phs000103.v1.p1), Hyperglycemia and Adverse Pregnancy Outcome (phs000096.v4.p1), and Genomic and Proteomic Network (phs000714.v1.p1) individual-level phenotype and genetic data can be obtained through dbGaP Authorized Access portal (https://dbgap.ncbi.nlm.nih.gov/dbgap/aa/wga. cgi?page=login). The informed consent under which the data or samples were collected is the basis for determining the appropriateness of sharing data through unrestricted-access databases or NIH-designated controlled-access data repositories. The summary statistics used in this publication other than the one generated are available at the following links: fetal GWAS of gestational duration (https://egg-consortium.org/), fetal and maternal GWAS of gestational duration (https://egg-consortium.org/), miscarriage (http://www.geenivaramu.ee/tools/misc_sumst ats.zip), age at first birth, oestradiol (women), endometriosis, number of live births and age at menarche (http://www.nealelab.is), age at menopause (https://www.reprogen.org), testosterone (women) 49

Code availability
Code for this project has been structured using a Snakemake workflow 66 and is available at (https://github.com/PerinatalLab/metaGWAS).