Placental Genome and Maternal-Placental Genetic Interactions: A Genome-Wide and Candidate Gene Association Study of Placental Abruption

While available evidence supports the role of genetics in the pathogenesis of placental abruption (PA), PA-related placental genome variations and maternal-placental genetic interactions have not been investigated. Maternal blood and placental samples collected from participants in the Peruvian Abruptio Placentae Epidemiology study were genotyped using Illumina’s Cardio-Metabochip platform. We examined 118,782 genome-wide SNPs and 333 SNPs in 32 candidate genes from mitochondrial biogenesis and oxidative phosphorylation pathways in placental DNA from 280 PA cases and 244 controls. We assessed maternal-placental interactions in the candidate gene SNPS and two imprinted regions (IGF2/H19 and C19MC). Univariate and penalized logistic regression models were fit to estimate odds ratios. We examined the combined effect of multiple SNPs on PA risk using weighted genetic risk scores (WGRS) with repeated ten-fold cross-validations. A multinomial model was used to investigate maternal-placental genetic interactions. In placental genome-wide and candidate gene analyses, no SNP was significant after false discovery rate correction. The top genome-wide association study (GWAS) hits were rs544201, rs1484464 (CTNNA2), rs4149570 (TNFRSF1A) and rs13055470 (ZNRF3) (p-values: 1.11e-05 to 3.54e-05). The top 200 SNPs of the GWAS overrepresented genes involved in cell cycle, growth and proliferation. The top candidate gene hits were rs16949118 (COX10) and rs7609948 (THRB) (p-values: 6.00e-03 and 8.19e-03). Participants in the highest quartile of WGRS based on cross-validations using SNPs selected from the GWAS and candidate gene analyses had a 8.40-fold (95% CI: 5.8–12.56) and a 4.46-fold (95% CI: 2.94–6.72) higher odds of PA compared to participants in the lowest quartile. We found maternal-placental genetic interactions on PA risk for two SNPs in PPARG (chr3∶12313450 and chr3∶12412978) and maternal imprinting effects for multiple SNPs in the C19MC and IGF2/H19 regions. Variations in the placental genome and interactions between maternal-placental genetic variations may contribute to PA risk. Larger studies may help advance our understanding of PA pathogenesis.

As a multi-factorial disorder of complex origin, PA aggregates in families of women with the condition [24], suggesting a strong role for genetic predisposition, a thesis supported by a number of candidate gene studies [25][26][27][28][29]. Findings from recent PA-related genome-wide association studies (GWAS) and candidate gene association studies (mitochondrial biogenesis and oxidative phosphorylation pathway genes) in the maternal genome by our group provided suggestive evidence supporting associations of variation in maternal cardiometabolic genes with risk of PA [30,31]. Given these findings indicative of the importance of genetic susceptibility factors in PA and the evidence highlighting the central role of placental pathology in PA [5,28,32], we hypothesized that genetic variations in the placental genome, particularly those variants in mitochondrial biogenesis (MB) and oxidative phosphorylation (OP) pathways, are associated with PA risk. Further, using data from both maternal and placental genomes, we examined interactions between placental and maternal genetic variations (in MB-OP pathway genes and imprinted regions) on risk of PA.

Study Setting and Study Population
The current study was conducted in the setting of the Peruvian Abruptio Placentae Epidemiology (PAPE) study that has been described before [30,31]. Briefly, PAPE study participants were recruited and enrolled among patients admitted for obstetrical services to the Hospital Nacional Dos de Mayo, Instituto Especializado Materno Perinatal, and Hospital Madre-Niño San Bartolomé in Lima, Peru, between August 2002 and May 2004 and between September 2006 and September 2008. Hospital admission and delivery logs were monitored daily to identify PA cases among new admissions to antepartum, emergency room, and labor and delivery wards of participating hospitals. PA was diagnosed based on evidence of retroplacental bleeding (fresh blood) entrapped between the decidua and the placenta or blood clots behind the placenta and any two of the following: (i) vaginal bleeding in late pregnancy not due to placenta previa or cervical lesions; (ii) uterine tenderness and/or abdominal pain; and, (iii) non-reassuring fetal status or death. Controls were selected from among pregnant women who delivered at participating hospitals during the study period and did not have a diagnosis of PA in the current pregnancy. For the current study, investigating associations of placental genome with risk of PA, 280 PA cases and 244 controls who provided placental samples at delivery were included. A subset of these cases and controls that also provided blood samples (222 PA cases and 198 controls) were also included in the maternal-placental genetic interaction investigations.
Ethical approval for the study was granted by the Institutional Review Boards (IRB) of Hospital Nacional Dos de Mayo, Instituto Especializado Materno Perinatal, Hospital Madre-Niño San Bartolomé in Lima, Peru and the IRB of Swedish Medical Center, Seattle, WA. All participants provided written informed consent in accordance with the principles of the declaration of Helsinki.

Data collection
Standardized structured questionnaires administered by trained research personnel were used to collect information on socio-demographic characteristics (including maternal age, marital status, employment status during pregnancy, and smoking and alcohol consumption before and during pregnancy), and medical history. A brief physical examination was conducted to measure maternal height, weight, and mid-arm circumference. Medical records were reviewed to abstract information on course and outcomes of the pregnancy. At delivery, placental samples were collected for DNA extraction and genotyping as described below. obtained from 8 sites (4 maternal and 4 fetal). For this study, biopsy samples taken from the fetal side were sampled for genomic DNA extraction. Biopsies were placed in cryotubes, snap frozen in liquid nitrogen, and stored at 280˚C until analysis. The Gentra PureGene Cell kit for DNA preparations (Qiagen, Hilden, Germany) was used to extract DNA from placental samples. Genotyping was conducted using the Illumina Cardio-Metabochip (Illumina Inc, San Diego, CA) platform [30], a high-density custom array designed to include 217,697 SNPs that represent DNA variations at regions previously related to diseases and traits relevant to metabolic and atherosclerotic-cardiovascular endpoints [33,34]. During the assay manufacturing process 20,972 SNPs (9.6%) failed, resulting in 196,725 SNPs available for genotyping, downstream quality control and statistical analyses [34,35].

Candidate gene/SNP selection
For the candidate association study, 32 genes that were involved in mitochondrial biogenesis and oxidative phosphorylation were selected based on literature [31] and a total of 333 SNPs belonging to these genes and found in the Cardio-MetaboChip were included in the candidate gene association analyses. For the maternal-placental genetic interaction study, 325 of these SNPs that also passed quality control in maternal blood genomes, as well as SNPs in imprinted regions (5 SNPs in IGF2/H19 and 33 SNPs in C19MC) included in the Cardio-MetaboChip were analyzed.

Data quality control
Quality control and preprocessing were performed on the genotype data. Individuals with genotyping failure in more than 10% of SNPs were removed (n52). SNPs with minor allele frequency (MAF) less than 1% or that failed to be genotyped in more than 10% of the study samples were removed (n577, 276), as well as SNPs not in Hardy-Weinberg equilibrium (HWE) among controls (n5667). After these quality control procedures, a total of 118,782 genome-wide SNPs were examined among 280 PA cases and 244 controls. Similar quality control procedures were performed on the maternal blood genotype data, resulting in 222 PA case and 198 controls among maternal-placental pairs.

Statistical analyses
Univariate logistic regression model was used to estimate odds ratio (OR) and 95% confidence interval (95% CI) relating each SNP with risk of PA, in the genome-wide and candidate gene analyses. For multiple testing correction, a false discovery rate (FDR) procedure was used [36]. Functions and functional relationships of genes represented by the top 200 genome-wide SNPs were obtained by pathway analysis using the Ingenuity Pathway Analysis (IPA, Ingenuity Systems, www.ingenuity.com) software. Gene-enrichment network score based on a modified Fisher's exact test were calculated to rank biological significance of networks in relation to PA.
In multivariable analyses, we applied penalized logistic regression models to identify sets of SNPs that are jointly associated with the risk of PA. These penalized approaches have previously been applied in the context of GWAS and have shown promising results [37][38]. These methods allow the selection of relevant variables or groups of variables and the estimation of their regression coefficients [39]. The number of selected variables is guided by a penalty parameter: the larger the parameter, the smaller the selected subset. A 20-fold cross-validation approach was performed to select the penalty parameter and the value yielding the smallest prediction error was used. For the genome-wide SNP analysis, we applied a lasso regression [39]. One characteristic of lasso regression is that it selects a single variable among a set of correlated variables. To circumvent this, SNPs in high linkage disequilibrium with a selected SNP were also considered using an r 2 threshold of 0.8 within 500 kb. For SNPs in the candidate gene analyses, a group penalty approach was used to account for the membership in a gene [40]. Furthermore, we considered a bi-level selection approach that uses a composite minimax concave penalty [41,42] to select candidate genes associated with PA as well as relevant SNPs within those genes. These penalized regression methods do not accommodate missing values and the software BEAGLE version 3.3.2 [43] was used to impute missing genotypes.
For weighted genetic risk score (WGRS) analyses [44], a 10-fold crossvalidation procedure was implemented to protect against model over-fitting, which arises from using the same data to estimate the regression parameters used in computing WGRS and to evaluate the association between PA risk and WGRS [45]. The procedure consisted of randomly partitioning the data into 10 equal size subsamples, using nine of the subsamples as training set and the left-out one as validation set, with each subsample being used in turn as a test set. For each fold, a multivariate logistic regression model was fit on the training set using the SNPs selected from multivariate analyses. A weighted approach was then used to compute Genetic Risk Scores (GRS) in the validation set by multiplying the number of risk allele for each locus by its associated effect size estimated from the training set. Once the WGRS were obtained for all individuals, the subjects were categorized into four groups defined by the quartiles in the control. A logistic regression model was then fit to examine the association of the WGRS with PA risk using the lowest quartile (Group 1) as a reference and adjusting for infant sex and population admixture. This 10-fold cross-validation procedure was repeated 1000 times to account for the variability in randomly partitioning the data into subsamples. The receiver operating characteristics (ROC) curve for each of the replicates was evaluated. The estimated effect sizes and AUCs over the 1000 replicates were used to obtain the respective point estimates and confidence intervals.
Maternal-placental interaction analyses (for candidate genes and imprinted regions) were performed using a multinomial model proposed by [46] and implemented in the EMIM and PREMIM software tools [47]. The method requires some biological assumptions, such as Hardy-Weinberg equilibrium (HWE), random mating, and rare disease. For each SNP, four models were considered and a model selection procedure based on the Bayesian information criterion (BIC) was applied. The four models correspond to allele effects operating only at the fetal level (Model F), allele effects operating only at the maternal level (Model M), an additive effect of maternal and fetal effects (Model M+F), and a model that includes a maternal-placental interaction effect (Model I). For the latter, we used a parametrization that introduces two interaction terms capturing incompatibility between maternal and placental genotypes; the interaction effects operate when the infant has one copy and the mother has either zero or two copies of the risk allele [46,48]. Maternal imprinting effect, which corresponds to the factor multiplying the disease risk if the infant inherits a risk allele from the mother, was tested using a likelihood ratio test [46,49].
Adjustment for the first four principal components was done for all univariate and multivariable logistic regression models to take into account population stratification. The various statistical analyses were conducted using a combination of software tools: PLINK, PREMIM, EMIM, Haploview, and R. As for multivariable approaches the R packages ncvreg and grpreg were used [41,50]. The pathway analyses were conducted using the Ingenuity Pathway Analysis (IPA) software. Table 1 shows selected characteristics of PA cases and controls. Average maternal age in both groups was around 27 years. Alcohol use during pregnancy and preeclampsia/eclampsia were more common among PA cases than controls. As expected, infant birthweight and gestational age at delivery were lower for PA cases compared with controls.

Results
In the GWAS analyses, we did not observe significant genomic inflation (l51.17) (Fig. 1). None of the FDR-corrected p-values were lower than a 0.05 threshold ( Table 2). The top GWAS hits were rs544201, rs1484464 (CTNNA2), rs4149570 (TNFRSF1A) and rs13055470 (ZNRF3) (nominal p-values: 1.11e-05 to 3.54e-05) ( Fig. 2 and Table 2). Functions of 56 genes represented by the top 200 SNPs from univariate GWAS analyses were examined using IPA. The top five networks from these analyses with p-values less than 0.05 are shown in Table 3. The top network enriched by these genes was a network of cell cycle, growth and proliferation (score543, p-value52.12e-19) (Fig. 3). In the candidate gene analyses, none of the FDR-corrected p-values were lower than a 0.05 threshold. The top hits in these analyses were rs16949118 (COX10) and rs7609948 (THRB) (nominal p-values: 6.00e-03 and 8.19e-03, respectively) ( Table 4). In addition, several SNPs in the PPARG gene were among the top hits in the candidate gene analyses.
Among 118,782 SNPs included in the GWAS analyses, six SNPs were selected using lasso regression ( Table 5). All six SNPs were among the top 200 hits, including the top three hits, identified using the univariate logistic regression analyses. When fitting a multiple logistic regression model with the selected SNPs, all 6 SNPs had empirical p-values lower than 0.05. Using a group penalty and bi-level selection approach, we identified 22 SNPs (in 14 genes) among the .300 SNPs included in the candidate gene analyses (Table 6). In multiple logistic regression analysis that included these SNPs, 11 SNPs had empirical p-values less than 0.05 (Table 6). WGRS were computed using SNPs selected from the respective GWAS and candidate gene multivariable analyses (  Fig. and S2 Fig.). We observed evidence for maternal-placental genetic interaction (on PA risk) for 23 candidate SNPs (Table 8). In particular, maternal-placental genetic interaction on PA risk was found for two SNPs in PPARG (chr3:12313450 and chr3:12412978). The model selection procedure based on the BIC for the SNP chr3:12412978, is shown in Table 9. The smallest BICcorresponds to Model I meaning that Model I fits the data best. In the imprinting effect analyses, we found that six SNPs in the C19MC region (of the 33 that were examined) and two SNPs in the IGF2-H19 region (of the five that were examined) showed evidence for maternal imprinting effect (empirical p-value ,0.05) ( Table 10 and Table 11). In addition, borderline imprinting effects were detected for a number of other SNPs in these regions.

Discussion
In this placental GWAS and candidate gene study of PA, no SNP had a significant association with PA risk following correction for false discovery. The top GWAS hits were rs544201, rs1484464 (CTNNA2), rs4149570 (TNFRSF1A), and rs13055470 (ZNRF3). The top 200 SNPs of the GWAS were in genes of pathways involved in cell cycle, growth and proliferation. The top candidate gene hits were rs16949118 (COX10) and rs7609948 (THRB). Using repeated ten-fold crossvalidations, participants in the highest quartiles of WGRS based on SNPs selected from GWAS and candidate gene analyses had 8.40-fold (95% CI: 5.8-12.56) and 4.46-fold (95% CI: 2.94-6.72) higher odds of PA compared to participants in the respective lowest quartiles. We also found evidence of maternal-placental genetic interaction on PA risk for two SNPs in PPARG (chr3:12313450 and chr3:12412978) and maternal imprinting effects for multiple SNPs in the C19MC and IGF2/H19 regions. A number of studies have investigated genetic risk factors of PA. Most of these prior studies were candidate gene studies including investigations of thrombophilia and rennin-angiotensin system pathways, folate metabolism pathways, and interleukin receptor related pathways [24,28,29,51]. Besides inconsistencies between reports of previous associations, few studies evaluated genome-wide variations and PA risk [24,28,29,30,31,51]. Given the multi-factorial nature of PA pathogenesis, GWAS studies can potentially provide important information concerning possible genetic susceptibility factors and related novel pathways that play a role in occurrence of PA. To our knowledge, no prior study investigated PA risk and genetic variation in the placenta, where abnormal vasculature, thrombosis, lesions, and reduced perfusion, all culminate in the eventual occurrence of PA [5,52].
In the current study, we did not have top hits of the GWAS that passed statistical significance after correcting for multiple testing. However, a number of SNPs among the top hits deserve mention. Of note, the top 10 SNPs represented five known genes including CTNNA2, TNFRSF1A, ZNRF3, ACSL1, and LIPA. The CTNNA2 gene codes for a protein linking cadherin adhesion receptors with the cytoskeleton. The SNP we identified in this gene, rs1484464, has been investigated in relation to a number of phenotypes, including tobacco use disorder, smoking cessation, coronary artery disease, lipid metabolism (HDL, total cholesterol, triglycerides) disorders, waist-hip ratio, CRP, glucose levels, insulin resistance, body mass index, blood pressure, and interleukin levels; however, significant associations have not been reported [53][54][55][56][57][58][59][60][61][62]. The TNFRSF1A is a gene member of the TNFR superfamily that activates the transcription factor NF-KB, mediates apoptosis, and functions as regulators of inflammation. The SNP we identified in the current study has been associated with susceptibility to inflammatory bowel disease, as well as response to anti-TNF treatment, in a large Danish cohort [63,64]. It has also been associated with reduced expression of TNF alpha receptor [65].
While the ZNRF3 gene is encoding for an E3 ubiquitin-protein ligase that acts as a negative regulator of the Wnt signaling pathway and a tumor suppressor, no previous report, to our knowledge, exists on the ZNRF3 SNP we identified in the current study. The ACSL1 gene codes for an isozyme of the long chain fatty acid coenzyme A ligase family that converts free long-chain fatty acids into fatty acyl-CoA esters, a nuclear-encoded mitochondrial protein, and plays key role in lipid biosynthesis and fatty acid degradation. The rs9997745 SNP in ACSL1 we identified has been previously associated with metabolic syndrome, fasting glucose, insulin levels, and insulin resistance [66]. Similarly, the LIPA gene encodes for lipase A, also known as cholesterol ester hydrolase, an enzyme that functions in the lysosome to catalyze the hydrolysis of cholesteryl esters and triglycerides. The SNP in LIPA, rs792269, is associated with total cholesterol, LDL cholesterol, and triglycerides [67]. None of the other top ten hits in our list had any previous associations reported. Our pathway analyses revealed that genes participating in cell cycle, cell growth and proliferation, and gene expression were overly represented by SNPs that were among the top hit of our GWAS. It is well known that disruptions in underlying normal placental growth and development are key underlying pathways that may later lead to the occurrence of PA [68]. The top hits in the candidate gene association analyses included SNPs in the COX10 (rs16949118) and THRB gene (rs7609948). COX10 encodes for the cytochrome C oxidase protein, the terminal component of the mitochondrial respiratory chain that catalyzes the electron transfer from reduced cytochrome C to oxygen. Genetic variations in this gene have been related to several diseases with underlying mitochondrial dysfunction including Alzheimers' disease, neurodegenerative diseases and other childhood disorders [69][70][71]. The THRB gene encodes a protein that is a receptor for triidothyronine. Several studies have reported associations between disorders in thyroid (particularly low thyroid levels) and placental disorders including PA [72][73][74]. The SNPs identified in our study, however, have not been associated with phenotypes or clinical outcomes. In addition to these SNPs, several other SNPs in the PPARG gene, belonging to the PPAR-family of genes that have been well-described in relation to placental growth and development, were among the top hits in the candidate gene analyses [75,76].
We examined associations between GRS, calculated from top hits of the GWAS and candidate gene analyses, respectively, and risk of PA and demonstrated strong associations between GRS and PA risk in both instances. While we did not use separate training and testing samples, we have used a cross-validation approach to protect against overfitting. These preliminary analyses are helpful to summarize identified effects of genetic variations, and could help in the construction of predictive models in future studies [77,78]. In particular, such genetic prediction have advantages, over non-genetic prediction models, as they are highly stable over time and are more suited for assessment of lifetime risk. In fact, their utility improves over time [78]. In addition, the decreasing genotyping cost and minimal invasiveness associated with obtaining samples highlight the potential importance of genetic prediction scores. On the other hand, the need for large study populations, that comprise of training and testing sets, and identification of genetic variants that individually account for large effects, are potential challenges in this area of research.
In the current study, for two SNPs in PPARG (chr3:12313450 and chr3:12412978), models with maternal-genetic interaction on PA risk were found to fit the data best. A number of studies have previously reported interactions between maternal and fetal metabolic genes on maternal and fetal outcomes [79][80][81][82][83][84]. Liang et al. have previously reported significant interaction between maternal and fetal genetic variations at the G308A SNP of TNF-alpha on risk of preterm delivery [85]. Similarly, other investigators have also reported significant maternal-fetal genotype interactions in IL-1beta -511C/T genotype, 4845GG genotype of IL-1alpha, along with the G308A SNP of TNF-alpha on preterm delivery risk [84]. Sinsheimer et al. have particularly stressed that the complex interplay of maternal and fetal genetics can be important for phenotypes originating with the placenta, given the importance of both maternal and fetal related (paternal) risk factors to placenta-based diseases and demonstrated gene expression differences in placenta-related pathologies (e.g. preeclampsia) [83]. Interestingly, the gene that has been highlighted in relation to maternal genetic variation and fetal sex interactions on risk of gestational diabetes is the PPARG gene [81] More specifically, Hocher et al. have demonstrated that mothers carrying G alleles of the Pro12Ala polymorphism delivering a girl had a higher total glycated hemoglobin (6.81) versus mother carrying the same alleles but delivering boys (5.85) (p-value50.0015) [81]. However, to our knowledge, no prior study investigated maternal-placental genotype interactions in relation to PA risk. Similarly, placental growth and development is primarily under the control of fetal genes inherited from the father [5,86]. Imprinted paternal alleles regulate formation of placenta and membranes surrounding the embryo, whereas the development of the embryo itself requires contribution from the maternally derived alleles [5,86]. Our findings suggest maternal imprinting effects for multiple SNPs in the C19MC and IGF2/H19 regions. While these imprinting sites are well described for several conditions including placental growth and development, our findings are novel in terms of linking imprinting at these sites to PA risk [87,88]. Given our sample size and related concerns regarding limited available statistical power, in exploratory analyses, we conducted maternal-placental interaction analyses of the PPARG gene using a haplotype-based approach. Tag SNPs from this gene were identified and haplotype blocks were defined using the Haploview software version 4.2 [89,90]. A total of four haplotype blocks tagged by 29 SNPs were identified. For each haplotype block, three possible diplotypes HH, HH 0 , H 0 H 0 as described in [80] were defined with the haplotype ''H'' denoting the ''relevant'' haplotype and ''H 0 '' denoting all other haplotypes. Each haplotype with frequency greater than 0.05 was considered as a potential ''relevant'' haplotype. For each ''relevant'' haplotype a procedure similar to the SNP interaction analysis was performed through the EMIM and PREMIM software tools. The ''relevant'' haplotype with the smallest BIC after evaluating all potential ''relevant'' haplotypes in each haplotype block was selected. This haplotype was called the optimal ''relevant'' haplotype. Findings from the exploratory interaction analyses are presented in Table 12. For haplotype block 1 a maternal-only effect was selected whereas for haplotype block 2 there was no evidence of haplotype effect. For haplotype block 3 a placental-only effect was selected. For both haplotype blocks (1 and 3), there was evidence that group 3 had a significantly lower risk of PA compared to the reference group (RR50.21, 95% CI: 0.066-0.68 and RR50.51, 95% CI: 0.293-0.878, respectively). For haplotype block 4, there was evidence of maternal-placental interaction effect. The group 3 showed a significantly higher risk of PA (RR55.71 95% CI: 3.398-9.604) whereas group 5 presented a significantly lower risk (RR50.18, 95% CI: 0.037-0.884) compared to the reference group. While findings from these haplotype based interaction analyses are encouraging, caution is warranted in interpreting these results due to uncertainties in the haplotype estimation process and the simplified model that considers diplotypes rather than evaluating all haplotypes using a reference.
This study expands the literature concerning the genetics of PA in a number of respects. We evaluated placental genetic variations, assessed interactions between maternal and placental genetic variations and examined placental imprinting effects on PA risk. However, several study limitations should be considered when interpreting our findings. First, our GWAS study has limited power to examine associations between genetic variations and disease risk, particularly for SNPs that have low to very low minor allele frequencies. Similarly, our statistical power to detect SNP-SNP interactions on PA risk was limited. To the extent possible we have tried to employ approaches that involve genetic risk scores to assess placental genetic variations contributions to PA risk. Second, many of the identified top hits of our analyses, both in the main effect and interaction models, have not been previously described either in genetic epidemiology or basic science investigations. Therefore, an important next step, along with replication efforts, should involve characterization of functional effects of these variations. Third, our imprinting effect assessment was based on mother-offspring dyad data and could benefit from triad data that also has information on fathers. Fourth, our GRS-based analyses, because of limited statistical power, did not involve independent training and testing sets, which would have been ideal for evaluating their predictive power, but rather relied on repeated ten-fold cross-validations. The WGRS will provide a summary estimate of effects of multiple SNPs, and can provide specific hypotheses that can be tested in future studies. Finally, findings from our study need to be cautiously generalized to other populations with different genetic make-up, population admixture, and environmental risk factors.
In sum, we found that variations in the placental genome and interactions between maternal-placental genetic variations may contribute to PA risk. We reported several novel loci where placental genetic variations may be associated with PA risk as well as several novel loci for maternal-placental genetic interactions on PA risk. Future larger investigations may help advance our understanding of PA pathogenesis.