Maternal-Fetal Genetic Interactions, Imprinting, and Risk of Placental Abruption

Abstract

Maternal genetic variations, including variations in mitochondrial biogenesis (MB) and oxidative phosphorylation (OP), are associated with placental abruption (PA). However, the role of maternal-fetal genetic interactions (MFGI) and parent-of-origin (imprinting) effects in PA remain unknown. We investigated MFGI in MB-OP, and imprinting effects in relation to risk of PA. Among Peruvian mother-infant pairs (503 PA cases and 1,052 controls), independent single nucleotide polymorphisms (SNPs), with linkage-disequilibrium coefficient <0.80, were selected to characterize genetic variations in MB-OP (78 SNPs in 24 genes) and imprinted genes (2713 SNPs in 73 genes). For each MB-OP SNP, four multinomial models corresponding to fetal allele effect, maternal allele effect, maternal and fetal allele additive effect, and maternal-fetal allele interaction effect were fit under Hardy-Weinberg equilibrium, random mating, and rare disease assumptions. The Bayesian information criterion (BIC) was used for model selection. For each SNP in imprinted genes, imprinting effect was tested using a likelihood ratio test. Bonferroni corrections were used to determine statistical significance (p-value<6.4e-4 for MFGI and p-value<1.8e-5 for imprinting). Abruption cases were more likely to experience preeclampsia, have shorter gestational age, and deliver infants with lower birthweight compared with controls. Models with MFGI effects provided improved fit than models with only maternal and fetal genotype main effects for SNP rs12530904 (p-value=1.2e-04) in calcium/calmodulin-dependent protein kinase [CaM kinase] II beta (CAMK2B), and, SNP rs73136795 (p-value=1.9e-04) in peroxisome proliferator-activated receptor-gamma (PPARG), both MB genes. We identified 320 SNPs in 45 maternally-imprinted genes (including potassium voltage-gated channel subfamily Q member 1 [KCNQ1], neurotrimin [NTM], and, ATPase phospholipid transporting 10A [ATP10A]) associated with abruption. Top hits included rs2012323 (p-value=1.6E-16) and rs12221520 (p-value1.3e-13) in KCNQ1, rs8036892 (p-value=9.3E-17) and rs188497582 in ATP10A, rs12589854 (p-value=2.9E-11) and rs80203467 (p-value=4.6e-11) in maternally expressed 8, small nucleolar RNA host (MEG8), and rs138281088 in solute carrier family 22 member 2 (SLC22A2) (p-value=6.8e-9).We identified novel PA-related maternal-fetal MB gene interactions and imprinting effects that highlight the role of the fetus in PA risk development. Findings can inform mechanistic investigations to understand the pathogenesis of PA.

Introduction

Disturbances that involve mitochondrial biogenesis (MB) and oxidative phosphorylation (OP) underlie pathologic mechanisms leading to placental abruption (PA) - a complex multifactorial disease characterized by premature separation of the placenta from the wall of the uterus [1-3]. Genome-wide and candidate gene association studies have identified common maternal single nucleotide polymorphisms (SNPs) in several MB and OP genes that are associated with PA risk [4-6]. However, findings were inconsistent, in line with other previous reports of genetic associations in complex diseases [7,8].

Investigators have suggested that assessment of maternal-fetal genetic interactions and assessment of imprinting effects, where risk is conferred depending on the parent-of-origin, may explain the missing heritability of complex diseases, particularly those with perinatal origins such as PA [7-9]. PA is a consequence of complex interplay of maternal and fetal genetics, epigenetics, and metabolic factors. For instance, the fetal genome influences placental growth and development, placental implantation and vascularization [10], all of which have been related to PA risk. In addition, many known imprinted genes affect embryonic or trophoblast growth [11] and have been implicated in preeclampsia [12], a known risk factor of PA [13]. Interactions between maternal and fetal genetic variations have previously been demonstrated in preterm delivery – another pregnancy complication with complex origin [14]. However, only one prior study, a study from our group, examined maternal-fetal genetic interactions in relation to risk of PA [4]. Using maternal-placental pairs from 222 PA cases and 198 controls, Denis et al. [4] reported maternal-fetal genetic interactions on PA risk for two SNPs in the PPARG gene (chr3:12313450 and chr3:12412978) and imprinting effects for multiple SNPs in imprinted (C19MC and IGF2/H19) regions. Using the largest assembled mother-infant dyad of PA cases and controls (503 PA case and 1,052 control mother-infant pairs) to date, that includes participants from the previous report [4], and an expanded set of SNPs (>3,035) in MB-OP genes and imprinted genes, we investigated maternal-fetal genetic interactions and imprinting effects in relation to PA risk.

Methods

Study setting and study populations

The study was conducted among participants of the Peruvian Abruptio Placentae Epidemiology (PAPE) and Placental Abruption Genetic Epidemiology (PAGE) studies, case-control studies of PA conducted in Lima, Peru. Both PAPE and PAGE studies had similar study objectives, setting and designs, but subjects recruited to each of the studies did not overlap, and have been reported before [4-6]. Briefly, participants were recruited among women who were admitted for obstetrical services to antepartum wards, emergency room, and labor and delivery wards of participating hospitals. The PAPE study participating hospitals were Hospital Nacional Dos de Mayo, Instituto Especializado Materno Perinatal, and Hospital Madre-Niño San Bartolomé (also known as Hospital San Bartolome) in Lima, Peru. Participants were recruited between August 2002 and May 2004 and between September 2006 and September 2008. The PAGE study participating hospitals were Instituto Nacional Materno Perinatal, Hospital Rebagliati, Hospital San Bartolome, Hospital Hipolito Unanue, Hospital Arzobispo Loayza, Hospital Dos de Mayo, and Hospital Maria Auxiliadora. Participants were recruited between March 2013 and December 2015. The underlying prevalence of PA in these study settings was 0.7%. The characteristics of the study population were similar between the two studies [5].

Participants who were <18 years, delivered infants from a multifetal pregnancy, had medical records that were insufficient to determine the presence or absence of PA (described below), had other diagnoses associated with third trimester bleeding (e.g. placenta previa), or reported taking blood thinning medications were excluded from the studies. The original sample sizes of the two studies were 490 PA cases and 500 controls in PAPE, and 522 PA cases and 1,147 controls in PAGE after quality control [5]. In the present study, we included PAPE participants who provided maternal blood and placental samples at delivery (176 PA case and 185 control mother-infant pairs) and PAGE participants who provided maternal saliva and newborn buccal cells at delivery (327 PA case and 867 control mother-infant pairs). In total, the pooled sample size after sample quality control steps (described below) was 503 PA case and 1,052 control mother-infant pairs. Study protocols of both studies were approved by the Institutional Review Boards (IRBs) of participating institutions and the Swedish Medical Center, Seattle, WA, where the studies were based. All participants provided written informed consent.

Data collection

PAPE and PAGE study participants were interviewed by trained personnel using standardized, structured questionnaires to collect information on sociodemographic characteristics and risk factors including maternal age, medical history, and smoking (both current and pre-pregnancy). Information on the course and outcomes of the pregnancy were abstracted from maternal medical records. Emergency room admission logbooks, labor and delivery admission logbooks, and the surgery report book were reviewed to determine a diagnosis of PA based on evidence of retro-placental bleeding (fresh blood) entrapped between the decidua and the placenta or blood clots behind the placental margin, accompanied by any two of the following: (i) vaginal bleeding at ≥20 weeks in gestation that is not due to placenta previa or cervical lesions; (ii) uterine tenderness and/or abdominal pain (without other causes, such as those due to hyperstimulation from pitocin augmentation); and, (iii) non-reassuring fetal status or fetal death [15]. Control participants, who did not have a diagnosis of PA in the current pregnancy, were randomly selected from eligible pregnant women who delivered at the participating hospitals during the respective study periods. Women with medical records that were insufficient to determine the presence or absence of PA, and those that reported taking blood thinning medications were not included in PAPE or PAGE studies. Additionally, mothers with other diagnoses associated with third trimester bleeding (e.g. placenta previa) were not included. These criteria were applied to both cases and controls.

In PAPE, maternal blood was obtained for maternal DNA extraction. In addition, placentas were collected immediately after delivery, weighed, double bagged and transported in coolers. Tissue biopsies (approximately 0.5 cm³ each) were obtained from 8 sites (4 maternal and 4 fetal) by stripping the chorionic plate and overlying membranes. The biopsy samples were taken from the fetal side and sampled for fetal genomic DNA extraction by placing them in cryotubes, snap frozen in liquid nitrogen, and stored at −80°C until analyses. In PAGE, maternal saliva and newborn buccal cells were collected, plated and stored using the Oragene™saliva cell collection kits (OGR500 and OGR250, DNA Genotek Ottawa, Canada), for DNA extraction and genotyping.

Genomic DNA extraction in the PAPE study were conducted using Gentra PureGene Cell kit (Qiagen, Hilden, Germany). SNP genotyping to characterize genome-wide variations were performed using Illumina Cardio-Metabochip (Illumina Inc., San Diego, CA) platform. In the PAGE study, genomic DNA were extracted using the Qiagen DNAeasy™ system and manufacturer protocols (Qiagen, Valencia, CA). SNP genotyping to characterize genome-wide variations were performed using the Illumina HumanCore-24 BeadChip (Illumina Inc., San Diego, CA) platform.

Maternal and fetal SNP data quality control procedures were applied using identical criteria (in PAPE and PAGE studies) before data analyses. SNPs were excluded if they had excessive missing genotype (SNPs with genotype call rate of <95%), deviated from Hardy-Weinberg equilibrium (HWE; p-value<1e-05), and had low minor allele frequency (MAF<0.05). The total number of SNPs, directly genotyped, that remained for further analyses in PAPE and PAGE studies were 128,371 and 241,301, respectively. Maternal-fetal pairs (PAPE n=23; PAGE n=10) were excluded if they were duplicates or related (Identity by Decent [IBD] value>0.9), had more than 5% of genotyping failure rate (PAPE n=45; PAGE n=51), and had excess heterozygosity rates (outside the range of mean ± 3 standard deviations of heterozygosity rate; PAPE n=5; PAGE n=15). PAGE and PAPE genotype data were then imputed to infer unobserved genotypes using identical steps. The genotype data were phased using SHAPEIT [16] to infer haplotypes and improve imputation accuracy using the 1000 Genomes haplotypes as the reference. Phased haplotypes were then used to impute the non-typed SNPs using IMPUTE2 [17]. After imputation and further quality control (filtering SNPs with imputation certainty score (Info)<0.3, HWE <0.00001, genotyping call rate<0.05, and MAF<0.05), a total of 5,553,176 and 5,314,631 SNPs were available for selection of genes and SNPs (described below) in the PAPE and PAGE studies, respectively.

The goal of the present study is to conduct a candidate gene study and evaluate a priori selected SNPs that are in genes implicated in PA pathway and imprinting genes. Candidate genes with described functions in MB and OP were selected from previously published studies [18-24]. Among 785 (in 101 MB-OP genes) and 359 SNPs (in 26 MB-OP genes) that were genotyped/imputed in the PAPE and PAGE studies, respectively, 322 overlapping SNPs (in 24 MB-OP genes) were selected. Pair-wise linkage disequilibrium (LD) was assessed between these 322 SNPs using SNAP [25]. Testing all of the SNPs that include SNPs that are highly in linkage with one-another can lead to multiple testing issues and higher likelihood of statistically spurious positive results. Thus, we selected a total of 78 independent SNPs (LD<0.80 in the set) in the 24 MB-OP genes (see S3 Table) for maternal-fetal interaction analyses. Similarly, a total of 12,459 SNPs in 83 imprinted genes from PAPE study and 10,030 SNPs in 78 imprinted genes from PAGE study were identified using GeneImprint [26]. Out of 9,666 SNPs in 73 imprinted genes that overlap between the two studies, a total of 2,713 independent SNPs (in 35 imprinted genes) were selected for imprinting effect analyses.

Statistical analyses

Mean and standard deviations for continuous variables and proportions for categorical variables were used to compare the characteristics of PA case and control participants. In maternal-fetal interaction analyses, for each SNP, similar to Denis et. al. [4], four models corresponding to allele effects operating only at fetal level (Model F), allele effects operating at maternal level (Model M), an additive model of maternal and fetal effects (Model M+F), and a model that includes a maternal-fetal interaction effect including the main effects (Model I) were considered. For the latter, we applied a parametrization that introduces two interaction terms capturing incompatibility between maternal and fetal genotypes; the interaction effects operate when the infant has one copy and the mother has either zero or two copies of the risk allele. The Bayesian information criterion (BIC) was used for model selection. In addition to the maternal and fetal genotype effects, we estimated the risk ratio (RR) of disease when the infant has one copy and the mother has zero copies of the risk allele. The reference group is defined as mother-infant pairs carrying zero copies of the risk allele.

Imprinting due to parent-of-origin effects relate to scenarios when a disease is influenced by the risk allele depending on where it is inherited from (i.e. mother or father) [27]. To test for imprinting effect on PA risk, we examined the transmission of alleles from the parent to the offspring using a likelihood ratio test described by Ainsworth et. al. and Howey et. al. [28,29], and implemented in PREMIM/EMIM [29], a statistical software designed to conduct imprinting/parent-of-origin effects analysis. The imprinting/parent-of-origin effect (I_m) corresponds to the factor multiplying the disease risk if the infant inherits a risk allele from the mother [28,29]. The test compares the maximum likelihood estimates of parameters representing maternal, fetal and imprinting genetic effects to the maximum likelihood estimates of parameters representing maternal and fetal genetic effects. For example, an I_m<1 would suggest effect of a paternally-inherited allele (paternal over-transmission). The p-values are then computed from the likelihood ratio test of the chi-squared value with a two degree of freedom test.

Bonferroni corrections were applied to correct for multiple testing: p-value<6.4e-4 for the 78 maternal-fetal interaction tests, and, p-value<1.8e-5 for 2,713 imprinting effect tests. In post-hoc exploratory analyses, we examined functions and functional relationships of imprinted genes that were represented by SNPs with significant imprinting effects using Ingenuity Pathway Analysis (IPA, Ingenuity, Redwood, CA) [30]. In the IPA based on the Ingenuity Pathways Knowledge Base (IPKB), gene-enrichment of networks was assessed using network score, negative log of p-values of a modified Fisher’s exact test.

All analyses were conducted based on HWE, random mating, and rare disease assumptions. Statistical analyses were conducted using PREMIM [29], EMIM [29], R (version i386 3.1.2) and SAS (Version 13).

Results

PAPE and PAGE PA-case mother-infant pairs were similar to control mother-infant pairs with respect to maternal age, marital status, employment, planned pregnancy, infant sex, alcohol use, drug use and vitamin use. Compared to control mother-infant pairs, PA case mother-infant pairs were more likely to smoke during pregnancy, deliver earlier (i.e., shorter gestational age), deliver infants with lower birth weight, and have a diagnosis of preeclampsia in the current pregnancy (S1 Table).

Out of 70 SNPs in candidate genes that were evaluated for interaction effects, we found evidence for significant maternal-fetal genetic interaction on PA risk for two SNPs (rs12530904 [CAMK2B] and rs73136795 [PPARG]). The BIC value was smallest (best fitting) in Model I (interaction model) for rs12530904 (p-value=1.2e-04) and rs73136795 (p-value=1.9e-04) (S2 Table). The interaction model showed that the risk of PA associated with a maternal GG genotype and fetal AG genotype at the rs12530904 locus is 1.79-fold (95%CI: 1.19, 2.69) higher relative to the referent group (maternal GG and fetal GG genotype). Similarly, the risk of PA associated with a maternal GG genotype and fetal AG genotype at the rs73136795 locus was 2.58-fold (95%CI: 1.64, 4.07) higher relative to the referent group (maternal GG genotype and fetal GG genotype) (Table 1 and S2 Table).

Table 1.

Association estimates of SNPs selected with maternal-fetal interaction as best fitting model.

						Fetal Effect Estimates	Maternal Effect Estimates	Interaction Effect Estimates
Gene	chr:pos	rsID	Risk Allele	Other Allele	Risk Allele Frequency	R1 (95% CI)	S1 (95% CI)	γ01 (95% CI)	γ21 (95% CI)	P-value
CAMK2B	7:44376785	rs12535537	A	G	0.16	1.05 (0.87, 1.27)	1.25 (1.04, 1.51)	1.87 (1.26, 2.77)	1.47 (0.85, 2.55)	1.6E-03
CAMK2B	7:44377171	rs12530904	A	G	0.14	1.00 (0.82, 1.22)	1.34 (1.11, 1.61)	1.79 (1.19, 2.69)	1.18 (0.66, 2.12)	1.2E-04*
PPARG	3:12468410	rs73136795	A	G	0.12	0.99 (0.79, 1.23)	0.79 (0.63, 0.99)	2.58 (1.64, 4.07)	0.77 (0.27, 2.24)	1.9E-04*
PPARG	3:12439348	rs35812816	C	G	0.17	0.51 (0.42, 0.62)	1.16 (0.94, 1.45)	1.69 (1.15, 2.50)	1.47 (0.42, 5.00)	1.1E-02

chr:pos : dbSNP build 37 hg19 chromosome:position

R1: Risk of PA associated with a copy of fetal risk allele (Fetal genotype effect)

S1: Risk of PA associated with a copy of maternal risk allele (Maternal genotype effect)

γ₀₁: Risk of PA associated with 1 copy of risk allele from fetus and zero copy of the risk allele from the mother

γ₂₁: Risk of PA associated with 1 copy of risk allele from fetus and 2 copies of the risk allele from the mother

^*statistically significant after Bonferroni correction for multiple testing.

p-values correspond to the log-likelihood ratio test comparing the main effects to the interaction effect

Furthermore, we identified additional loci rs12535537 [CAMK2B] and rs35812816 [PPARG] genes with Model I as the best fitting model (Table 1 and S2 Table). However, the corresponding p-values were not statistically significant after Bonferroni correction (rs12535537 [p-value=1.6e-3], and rs35812816 [p-value=0.01]). The risk of PA associated with a maternal GG genotype and fetal AG genotype at rs12535537 locus was 1.87-fold (95%CI: 1.26, 2.77) higher relative to the referent group (maternal GG genotype and fetal GG genotype). Similarly, the risk of PA associated with a maternal GG genotype and fetal CG genotype at rs35812816 locus was 1.69-fold (95%CI: 1.15, 2.50) higher relative to the referent group (maternal GG genotype and fetal GG genotype).

We identified 320 SNPs (having 224 maternally, 91 paternally, 1 isoform-dependent and 4 random expressed-alleles) with imprinting effects (that reached statistical significance after Bonferroni correction) on PA risk in 45 known imprinted genes (Table 2 and S4 Table). These imprinted genes included 41 SNPs in potassium voltage-gated channel subfamily Q member 1 (KCNQ1), 48 SNPs in neurotrimin (NTM), 15 SNPs in ATPase phospholipid transporting 10A (ATP10A) genes, 8 SNPs in maternally expressed 8 (MEG8) gene and 5 SNPs in solute carrier family 22 member 2 (SLC22A2) gene. Top hits included rs2012323 (p-value=1.6E-16) and rs12221520 (p-value1.3e-13) in KCNQ1, rs8036892 (p-value=9.3E-17) and rs188497582 in ATP10A, rs12589854 (p-value=2.9E-11) and rs80203467 (p-value=4.6e-11) in maternally expressed 8, small nucleolar RNA host (MEG8), and rs138281088 in solute carrier family 22 member 2 (SLC22A2) (p-value=6.8e-9) (Table 2 and S4 Table). The average imputation Info score of the 320 SNPs across PAGE maternal, PAGE fetal, PAPE maternal and PAPE fetal studies range from 0.40-0.94. Our post-hoc exploratory IPA analyses of the 35 imprinted genes identified that the top two enriched gene networks were a network (Score=29) of cell cycle, cell morphology (S5 Table and S1 Fig), and a network (Score=29) of cardiovascular disease and free radical scavenging (S5 Table and S2 Fig).

Table 2.

SNPs of imprinted genes with parent-of-origin effect on PA risk (top 15 SNPs out of 320 that were significant after Bonferroni correction [p-value<1.84e-5])

Gene	Name	Chr:pos1	rsID2	A1 3	RAF4	Expressed Allele	P- value5	Im (95% CI)6	Info Score7
KCNQ1	Potassium voltage-gated channel subfamily Q member 1	11:2604437	rs2012323	A	0.13	Maternal	1.6E-16	0.11 (0.04, 0.26)	0.74
SLC22A18	Solute carrier family 22 member 18	11:2942798	rs426359	G	0.11	Maternal	2.7E-16	0.05 (0.01, 0.23)	0.59
SNRPN	Small nuclear ribonucleoprotein polypeptide N	15:25175735	rs377264185	T	0.12	Paternal	3.2E-19	0.02 (0.00, 0.20)	0.48
ATP10A	ATPase phospholipid transporting 10A	15:25973151	rs8036892	G	0.10	Maternal	9.3E-17	0.04 (0.01, 0.22)	0.53
SNRPN	Small nuclear ribonucleoprotein polypeptide N	15:25179795	rs671362	C	0.18	Paternal	3.1E-15	0.14 (0.07, 0.29)	0.53
SNRPN	Small nuclear ribonucleoprotein polypeptide N	15:25171908	rs77979767	A	0.14	Paternal	5.7E-14	0.07 (0.01, 0.39)	0.45
KCNQ1	Potassium voltage-gated channel subfamily Q member 1	11:2584917	rs12221520	T	0.18	Maternal	1.3E-13	0.08 (0.01, 0.44)	0.63
GNAS_AS1	GNAS antisense RNA 1	20:57478448	rs2295583	T	0.11	Paternal	1.8E-13	0.12 (0.05, 0.34)	0.58
NTM	Neurotrimin	11:132036752	rs74954142	G	0.09	Maternal	4.0E-13	0.05 (0.00, 0.47)	0.54
SGK2	SGK2, serine/threonine kinase 2	20:42193900	rs3827067	G	0.16	Paternal	5.9E-13	0.12 (0.04, 0.36)	0.48
NLRP2	NLR family pyrin domain containing 2	19:55489862	rs28376680	G	0.14	Maternal	6.8E-13	0.13 (0.04, 0.40)	0.61
AIM1	Enoyl-CoA hydratase/isomerase family	6:106854918	rs149264472	C	0.11	Paternal	7.3E-13	0.16 (0.07, 0.35)	0.77
SNRPN	Small nuclear ribonucleoprotein polypeptide N	15:25174629	rs2736705	T	0.22	Paternal	9.0E-13	0.16 (0.07, 0.33)	0.49
MEG3	Maternally expressed 3	14:101325962	rs10144253	C	0.16	Maternal	1.1E-12	0.17 (0.08, 0.37)	0.55
ATP10A	ATPase phospholipid transporting 10A	15:25966731	rs188497582	C	0.17	Maternal	2.2E-12	0.17 (0.07, 0.38)	0.53

¹Build 37 Chromosome Position

²Build 37 SNP rsID

³Risk Allele

⁴Risk Allele Frequency

⁵P-value for a two degree of freedom test comparing the maximum likelihood estimates of parameters representing maternal, fetal and imprinting genetic effects to the maximum likelihood estimates of parameters representing maternal and fetal genetic effects

⁶I_m: Relative risk estimate (A multiplicative factor by which the probability of disease is multiplied if the child receives a maternal copy of the risk allele from its mother) of the likelihood ratio test

⁷Imputation Info Score across the four studies; rs2012323 (11:2604437; KCNQ1) was directly typed in the fetal PAGE study

Discussion

In this study, we identified several novel maternal-fetal MB SNP interactions and imprinting effects on PA risk. Maternal-fetal interactions were observed for SNPs in CAMK2B (rs12530904) and PPARG (rs73136795) while potential maternal-fetal interactions were observed for two other SNPs in the same genes (rs12535537and rs35812816 in CAMK2B and PPARG, respectively). Imprinting effects were observed for 310 SNPs in imprinted genes including KCNQ1, NTM, ATP10A, MEG8, and SLC22A2.

In the only other similar published study related to PA, our team reported maternal-fetal interaction for two PPARG SNPs (chr3:12313450 and chr3:12412978) and imprinting effect for eight SNPs, six in the C19MC region and two in IGF2-H19 [4]. While we found maternal and infant interactions on PA risk for PPARG SNP rs73136795 (chr3:12468410), the two previously reported PPARG SNPs were not in the set of SNPs we evaluated in the current study as they failed imputation quality (Info<0.3) in the PAGE study. We decided to discard SNPs below 0.3 info score after examining the Minor-allele score bins vs Imputation Info score plot for our data. In the plot, most of the poor imputation quality SNPs appeared to the left of info score=0.3. Similarly, we identified imprinting effect on PA risk for IGF2 SNP rs11564732 (chr1:2150895, p-value=9.3e-06); however, the previously reported SNPs in C19MC or IGF2-H19 genes were not evaluated in the current study because they were not genotyped or imputed in the PAGE study. Other studies have previously investigated interactions between maternal and fetal genetic variations on maternal and infant outcomes [14,31] [32]. For instance, interaction between maternal and fetal genetic variations at the G308A locus of TNF-alpha gene on risk of preterm delivery (PTD) has been reported by a study conducted among 250 PTD and 247 control Han Chinese families [14]. The combined maternal-fetal genotype GA/GA at the locus was associated with reduced risk of PTD (risk ratio=0.20 [95%CI: 0.07, 0.58])[14]. Goddard et. al. [32] observed evidence for maternal-fetal interaction at the rs5742620 loci in IGF1 gene on preeclampsia risk. Maternal-fetal pairs with AC/AC genotype at this locus had a 2.4-fold (p-value=0.0035) increased risk of preeclampsia compared to maternal-fetal pairs with CC/CC genotypes [32]. None of the SNPs investigated by the previous investigators were evaluated in our present study. However, collectively, findings from across available studies support possibility of maternal-fetal genetic interactions in pregnancy complications.

Our observation of PPARG maternal-fetal interaction on PA risk is noteworthy, not only because we found similar, although in different SNPs, interactions in our previous study, but also because the gene has been well-described in relation to placental growth, development, and function [4,33,34]. PPARG belongs to the PPAR-family of genes and is regulated by PPARG coactivator 1 alpha gene (PGC-1alpha) that is associated with enhanced MB [35-37]. PPARG is highly expressed in the placenta, inhibits trophoblast invasion through oxidized LDL in cytotrophoblasts of cells involved in invasion of the uterus [33,34], and is associated with the development of preeclampsia [34], an established risk factor of PA [13]. The other gene where we found significant maternal-fetal interactions was CAMK2B, a CaMK-family gene implicated in contraction-induced regulation of calcium handling in skeletal muscle and activation of MB [38-40]. CAMK2B is among several genes in myometrial relaxation and contraction pathways that are either transcribed in myometrial muscle cells or act upon the myometrium to regulate contraction [41], through pathways that involve oxytocin receptor activation [42]. Preterm uterine contractions are associated with PA risk [43]. CAMK2B is lowly expressed in the placenta, however miRNAs with low level of expression are shown to have key regulatory roles in vascular cells [44].

We found strong evidence for imprinting effect of several imprinted genes (including KCNQ1, NTM, and ATP10A) on PA risk. All SNPs we report herein have maternally-expressed alleles. The maternal imprinting effect estimates with I_M<1 suggest effect of a paternally-inherited allele. For example, the top imprinting result was a paternal over-transmission of rs2012323 (KCNQ1) A allele on chromosome 11 associated with PA risk. This may suggest evidence for non-expression of the maternally derived A allele in the offspring associated with PA [45,46]. Imprinted genes may affect maternal-fetal interactions that affect placental development [12,47-49]. Imprinted maternal alleles are required for the development of the embryo, and imprinted paternal alleles regulate formation of the placenta and the surrounding membranes of the embryo [11]. KCNQ1 encodes a voltage-gated potassium channel required for the repolarization phase of action potential [50]. KCNQ1 is expressed in the placenta and implicated in embryonic and placental growth. In mice, maternally inherited target deletion of ASCL2, a gene that resides in KCNQ1 cluster, is lethal due to failure of placental formation [49,51]. Both NTM, involved in neuronal cell adhesion, and ATP10A, ATPase phospholipid transporting 10A, genes are expressed in the placenta [52]. The imprinting effects of KCNQ1, NTM, or ATP10A on pregnancy-related outcomes have not been reported before.

We found imprinting effect of PA risk for rs11564732 in H19, a maternally imprinted gene near insulin like growth factor 2 (IGF2). We previously reported similar imprinting effects of PA risk by IGF2 [4]. H19/IGF2 regulates the development of the embryo and differentiation of cytotrophoblast cells [53], and was implicated in preeclampsia [54,55], a known risk factor of PA [13].

In our post-hoc exploratory analyses, the 35 imprinted genes that were represented by SNPs with significant imprinting effects showed gene enrichment for network of cell cycle and cell morphology, and a network of cardiovascular disease and free radical scavenging. Both of these networks align well with what is known about PA pathogenesis [20].

The underlying genetic architecture of PA has been examined in previous genome-wide and candidate gene association studies [4-6], which reported predominantly common, non-coding variants with modest effects and limited replication. Important strategies to address subsequent missing heritability include family studies that assess gene-gene interactions and imprinting effects [7,8]. Family studies are advantageous because affected relatives are more likely to share two nearby epistatic loci in LD that would be unlinked in unrelated individuals [8] and ignoring imprinting effects can mask true associations and diminish the proportion of heritability explained [8]. Other sources of missing heritability may be low frequency variants with intermediate effect. These should be tractable through larger sized studies and imputation of genome-wide data [8].

Our study is the most comprehensive investigation, to date, of maternal-fetal genetic interactions and imprinting effects on PA risk. These findings have the potential for enhancing our understanding of genetic variations in maternal and fetal genome that contribute to PA, a multi-factorial heritable disorder. The study addresses the potential limitations of sample size in previous studies. By conducting 1000 genomes genotype imputations, we analyzed a comprehensive set of SNPs in MB-OP and imprinted genes. However, in the current study, SNPs that did not overlap between PAPE and PAGE were excluded, not allowing us to examine some previously reported SNPs. Other limitations of our study include potential misclassification of sub-clinical PA (i.e. those with less placental disruption and consequent bleeding), which may limit the interpretation of the study results or reduce statistical power. We also did not distinguish between severe and mild cases of PA, which may have different risk factors or underlying mechanisms [2]. Our study was restricted to live births and interaction or imprinting effects that contribute to stillbirth, a common complication of PA, may have been missed. We assessed maternal-fetal genetic interaction on PA risk using MB-OP candidate SNPs. There could be similar interactions in other metabolic functions. This study may still be underpowered for small effects and rare genotypes. Finally, findings from the current study population may not be generalizable to other populations with different population genetic structure or PA risk pattern. Therefore, replication studies in different populations are critical to fully understand maternal-fetal genetic interactions and imprinting effects on PA risk. In an effort to validate our findings, we conducted post-hoc analyses by selecting a random set of “candidate” SNPs outside the proposed candidate genes and tested them for interaction and imprinting effect on PA risk. We did not see evidence for imprinting effect for these set of random SNPs.

In sum, findings in this study confirm the role of interactions between maternal and fetal genetic variations in MB and imprinting effects in PA. These findings highlight the potential of understanding the complex interplay between maternal and fetal genetic factors in explaining the missing heritability of PA and PA-related risk stratification. This may inform potential preventative and therapeutic targets of PA.

Supplementary Material

S1 Table

Selected characteristics of study participants

S2 Table

SNPs selected with maternal-fetal interaction as best fitting model.

S2 Figure

Significant networks (P-value=1.0e-29) represented by 12 imprinted genes with imprinting effect on PA risk. Molecules highlighted in purple represent cardiovascular disease, cell morphology, and free radical scavenging pathway

S1 Figure

Significant networks (P-value=1.0e-28) represented by 12 imprinted genes with imprinting/parent-of-origin effect on PA risk. Molecules highlighted in purple represent cell cycle and cell morphology pathway.

S3 Table

Mitochondrial biogenesis and/or oxidative phosphorylation pathway genes evaluated in the current study.

S4 Table

SNPs of imprinted genes with imprinting effect on PA risk (320 that were significant after Bonferroni correction [p-values<1.5e-5]).

S5 Table

Networks represented by 35 imprinted genes identified for imprinting effect on PA risk.

Data Availability

All relevant data are within the paper and its Supporting Information files. The participant level data are from the Peruvian Abruptio Placentae Epidemiology (PAPE) and Placental Abruption Genetic Epidemiology and Triggers (PAGE) studies whose authors may be contacted at mawilliams@hsph.harvard.edu. The local ethical committee does not allow public deposition of the data.