Preeclampsia prediction with maternal and paternal polygenic risk scores: the TMM BirThree Cohort Study

doi:10.21203/rs.3.rs-4883948/v1

Download PDF

Article

Preeclampsia prediction with maternal and paternal polygenic risk scores: the TMM BirThree Cohort Study

https://doi.org/10.21203/rs.3.rs-4883948/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Genomic information from pregnant women and the paternal parent of their fetuses may provide effective biomarkers for preeclampsia (PE). This study investigated the association of parental polygenic risk scores (PRSs) for blood pressure (BP) and PE with PE onset and evaluated predictive performances of PRSs using clinical predictive variables. In the Tohoku Medical Megabank Project Birth and Three-Generation Cohort Study, 19,836 participants were genotyped using either Affymetrix Axiom Japonica Array v2 (further divided into two cohorts—the PRS training cohort and the internal-validation cohort—at a ratio of 1:2) or Japonica Array NEO (external-validation cohort). PRSs were calculated for systolic BP (SBP), diastolic BP (DBP), and PE and hyperparameters for PRS calculation were optimized in the training cohort. PE onset was markedly associated with maternal SBP-, DBP-, and PE-PRSs in internal- and external-validation cohorts and with paternal SBP- and DBP-PRSs only in the external-validation cohort. Maternal DBP-PRS calculated using “LDpred2” presented the most improvement in prediction models and provided additional predictive information on clinical predictive variables. Paternal DBP-PRS improved prediction models in the internal-validation cohort. In conclusion, Parental PRS, along with clinical predictive variables, is potentially useful for predicting PE.

BirThree Cohort Study

family history

genome-wide association study

hypertensive disorders of pregnancy

LDpred2

Preeclampsia (PE) is a multisystem disorder characterized by de novo hypertension and proteinuria, affecting approximately 3.4% and 2.7% of pregnant women in the USA[1] and Japan[2], respectively. It causes approximately 45,900 maternal deaths worldwide each year.[3] In addition to effective interventions, such as moderate exercise,[4] aspirin administration,[5] and calcium supplementation,[6] increased vigilance throughout the high-risk pregnancy can aid in the early detection and treatment of PE, potentially reducing adverse pregnancy outcomes and leading to better outcomes[7] for mothers and fetuses. Therefore, developing accurate PE prediction models is crucial for clinical practice.

More than a hundred prediction models for PE are reported,[8] with some using genomic information from pregnant women (here, “maternal genomic information”).[9–12] As the heritability of PE was estimated to be 55% in a family study, with maternal and fetal contributions accounting for 35% and 20%, respectively,[13] maternal genomic information is expected to provide a good biomarker for PE prediction. Integrating maternal genomic information as a polygenic risk score (PRS) can potentially improve existing predictive models; however, relevant research remains insufficient. In previous studies,[9–12] PRS for blood pressure (BP) and PE were utilized but variables such as actual measured BP values and family history of PE, which are important clinical predictive variables, were not simultaneously incorporated into the prediction models. Actual measured BP values in early pregnancy are an effective PE biomarker and are included in many clinical prediction models.[8, 14] Similarly, family medical history is a predictive variable for diseases and reflects the genetic load and shared environmental factors.[15, 16] A recent study revealed that the PRS and family medical history provide complementary information on noncommunicable diseases.[17] As actual measured BP values and family history of hypertensive disorders of pregnancy (HDP) are more accessible than PRS, their combined clinical usefulness with genomic information should be explored. Although a previous study[18] reported no improvement when PRS was incorporated into a machine learning-based prediction model utilizing clinical predictive variables, this study had a small number of cases (< 100) and was not externally validated.

Given that genomic information from the paternal parent of the fetus (here, “paternal genomic information”) is transmitted to both fetal and placental tissues, the paternal genomic information holds predictive potential for PE. As obtaining the fetal genome is difficult, combining the maternal and paternal genomes may capture a comprehensive genetic predisposition to PE. However, no previous studies have employed such information in PE prediction models.

Herein, we examined the relationship of maternal and parental genomic information with PE onset. Additionally, we investigated whether parental PRSs have predictive information in addition to clinical models, including actual measured BP values and family history of HDP.

Participants

The Tohoku Medical Megabank (TMM) Project Birth and Three-Generation Cohort Study (the BirThree Cohort Study)[19, 20] recruited pregnant women and their families between 2013 and 2017. More than 50 obstetric clinics and hospitals in the Miyagi Prefecture, Japan, participated, registering 23,406 pregnant women and 8,823 participants who self-reported as fathers of the fetuses.

We excluded the following pregnant women from the study: those who withdrew consent, had multiple pregnancies, had stillbirth before 20 weeks of gestation, and missed the diagnosis of PE or delivery date. Herein, we included only the first-time participations of those who had multiple participations in the TMM BirThree Cohort Study.

Participants were genotyped using either the Affymetrix Axiom Japonica Array v2 (JPA v2) or Japonica Array NEO (JPA NEO). Participants were genotyped using JPA v2 and JPA NEO based on whether their genotypes were analyzed in the first or second half, respectively, of the genotyping process of the TMM Project. Because the date of enrollment was different according to the study area, the proportions of regions and hospitals from where the participants were recruited were different between participants subjected to the two arrays. In addition to the date of enrollment, the order of the analysis prioritized participants presenting with the data of their full family. Those with missing genotyping data were excluded after a standard quality control procedure.[21] Those genotyped using JPA v2 were divided into two cohorts—the PRS training cohort and the maternal internal validation cohort—at a ratio of 1:2. Those genotyped using JPA NEO were defined as the maternal external validation cohort. The paternal parents of the fetuses were genotyped using the same genotyping array platform as the participants. Cohorts with paternal genotypes in the maternal internal and external validation cohorts were defined as the parental internal and external validation cohorts, respectively (Fig. 1). Ethical approval was obtained from the Ethics Committee of the Tohoku Medical Megabank Organization (2013-1-103-1 and 2023-4-025), and informed consent for research participation was obtained from all participants. This study was performed in accordance with the Declaration of Helsinki.

Predictive variables

Based on previous research on PE prediction models,[22–26] we selected the following predictive variables: maternal age at conception, pre-pregnancy body mass index (BMI), chronic hypertension (CH), systemic lupus erythematosus (SLE), type 1 and 2 diabetes mellitus (DM), maternal family (mother or sisters) history of HDP, conception via in vitro fertilization (IVF), parity (nulliparous, parous with or without previous HDP), gestational age at the previous delivery, the inter-birth interval, and BP at the first antenatal care during 10–13 weeks of gestation. We collected BP data at 10–13 weeks of gestation when the majority of the population underwent the first or second antenatal care. Mean arterial pressure (MAP) was calculated and converted to log₁₀ transformed multiple of the median (log MoM) for the prediction model. Paternal age and family history of HDP were obtained for parental analyses.

Polygenic risk score

Genotyping and PRS calculations are described in Supplementary Methods. In brief, PRSs for three phenotypes, systolic BP (SBP), diastolic BP (DBP), and PE, were calculated using two methods, genome-wide clumping and thresholding (C + T) and a Bayesian approach using LDpred2.[27, 28] Although LDpred2 is generally considered more accurate than C + T, C + T is easier to implement and interpret.²⁷ The hyperparameters were optimized in the PRS training cohort. The number of variants included in each PRS is summarized in Supplementary Table S1.

Outcome measurement

PE was identified according to the guidelines of the American College of Obstetricians and Gynecologists[29, 30], using medical records at antenatal care. PE superimposed on the CH was included in the definition of PE. PE was automatically diagnosed using an algorithm based on medical record data and validated by a physician. Additionally, PE was classified as early or late onset based on if it occurred earlier than 34 weeks of gestation. The details are given elsewhere.[23, 29]

Model development

Herein, the competing risk model[31] was applied, which has been validated both internally and externally in Europe[32] and yielded comparable results in the TMM BirThree Cohort Study.[22, 23] This model assumes that all pregnant women will develop PE during pregnancy, but only some of them will actually develop PE because of the competition of delivery without PE. Additionally, a parametric survival model with a Gaussian distribution was applied.[31] Delivery without PE was considered censored. Missing predictive variables were imputed using k-nearest neighbor imputation with k = 140 (square root of the total study population[33]). To assess the model discrimination, Harrell’s C-statistic was calculated for the whole gestational age. To assess model calibration, the calibration slope was calculated by regressing observed survival outcomes on predicted gestational age at delivery with PE in the same parametric survival models.[34]

Statistical analysis

Herein, baseline characteristics were compared by PE status or internal/external validation cohorts. P-values were calculated using the t-test for continuous variables and the chi-square test or Fisher’s exact test for categorical variables. Pearson’s correlation coefficients were calculated for all combinations of parental PRSs using LDpred2 in the parental internal validation cohort.

Association analyses were performed using SBP-PRS, DBP-PRS, and PE-PRS using LDpred2. In the maternal internal and external validation cohorts, the associations between maternal PRSs and PE onset were examined using logistic regression analysis adjusted for maternal age and four genetic principal components (PC). Two models were developed, one with PRSs as continuous values and the other as tertile values. In the parental internal and external validation cohorts, associations between both maternal and paternal PRSs and PE onset were examined using logistic regression analysis adjusted for parental age and four genetic PCs. Maternal and paternal PRSs for the same phenotype were simultaneously included in the model. The results of internal and external validation cohorts were merged using inverse-variance weighting for meta-analysis. Interactions between maternal and paternal PRSs were also examined. As subanalyses, participants with CH were excluded from the study population, and analyses were performed again. To further assess the robustness of the results, linear regression analyses were performed with PRSs as exposures and BP in early pregnancy as the outcome. Additionally, the early- and late-onset PE were analyzed as outcomes to determine if the association with PRS differed by PE pathology.

The prediction model was developed in two stages: identification of a suitable PRS for prediction models and development of clinical prediction models using parental PRSs. In the first stage, prediction models were developed using each maternal PRS, age, and four genetic PCs. Based on the C-statistics of prediction models, we selected either C + T or LDpred2 as the PRS calculation algorithm, and SBP-PRS, DBP-PRS, or PE-PRS as the PRS phenotype. In the second stage, we created four models: reference model, family-history model, at-pregnancy-confirmation model, and in-early-pregnancy model (Supplementary Method). C-statistics and calibration slopes were calculated for each model. The 95% confidence intervals (CIs) and internal validation used the bootstrap method with 1,000 resamplings. A simulation study was performed to confirm the reproducibility of the first step. The cohort genotyped with JPA v2 was divided into PRS training and internal validation cohorts, with split ratios of 1:2, 2:3, and 1:1, repeated with 100 random numbers, to determine the PRS with the best predictive ability.

All analyses were conducted using R (4.1.0) unless otherwise noted. We considered a two-tailed P-value of < 0.05 as significant.

A total of 19,836 participants were eligible for the present study: 3,384 in the PRS training cohort, 6,768 in the maternal internal validation cohort, and 9,684 in the maternal external validation cohort (Fig. 1). In the maternal internal and external validation cohorts, 3,673 (54.3%) and 2,616 (27.0%) participants, respectively, had paternal genotyping data.

In total, 352 of 6,768 (5.2%) and 268 of 9,684 (2.8%) participants developed PE in the maternal internal and external validation cohorts, respectively (Table 1). Of the 3,673 and 2,616 participants in the parental internal and external validation cohorts, respectively, 216 (5.9%) and 29 (1.1%) participants developed PE (Supplementary Table S2). Participants with PE tended to have an earlier delivery; were older; had a higher BMI; had a history of CH, DM, and SLE; were nulliparous or parous with previous PE; had a shorter last delivery gestational age; conceived via IVF; and had a higher BP at 10–13 weeks of gestation, with slight variations among the maternal and parental cohorts (Table 1 and Supplementary Table S2). Notable differences in the characteristics between the internal and external validation cohorts were observed (Supplementary Tables S3 and S4). Strong correlations existed between maternal PRSs or among paternal PRSs, whereas weak correlations existed between maternal and paternal PRSs (Supplementary Figure S1).

Table 1

Baseline characteristics in the maternal internal and external cohort, stratified by the preeclampsia status
		Maternal internal validation cohort (n = 6,768)				Maternal external validation cohort (n = 9,684)
		PE	Not affected	P-value	n of missing	PE	Not affected	P-value	n of missing
		n = 352	n = 6,416			n = 268	n = 9,416
Gestational age, weeks		38.6 ± 2.2	39.2 ± 1.6	< 0.001	0	38.3 ± 2.7	39.1 ± 1.8	< 0.001	0
Maternal age at conception, years		31.9 ± 5.6	31.5 ± 5.0	0.143	0	32.2 ± 5.2	31.4 ± 5.0	0.005	0
Pre-pregnancy BMI, kg/m²		23.6 ± 4.9	21.6 ± 3.4	< 0.001	113	23.5 ± 5.1	21.4 ± 3.3	< 0.001	124
CH, %		104 (29.5)	177 (2.8)	< 0.001	0	103 (38.4)	192 (2.0)	< 0.001	0
DM (type 1 or 2), %		8 (2.3)	17 (0.3)	< 0.001	2,162	0 (0.0)	9 (0.1)	> 0.999	3,818
SLE, %		2 (0.6)	5 (0.1)	0.048	2,162	2 (0.7)	3 (0.0)	0.007	3,818
Maternal family history of HDP, %		12 (3.4)	136 (2.1)	0.155	2,162	3 (1.1)	158 (1.7)	0.632	3,818
Parity				< 0.001	30			< 0.001	167
	Nulliparous, %	176 (50.0)	2,796 (43.6)			102 (38.1)	3,615 (38.4)
	Parous with previous PE, %	30 (8.5)	174 (2.7)			29 (10.8)	260 (2.8)
	Parous with no previous PE, %	146 (41.5)	3,446 (53.7)			137 (51.1)	5,541 (58.8)
	Inter-birth interval, years	3.9 ± 3.5	3.6 ± 2.6	0.166	206	3.8 ± 2.9	3.6 ± 2.6	0.304	399
	Last delivery gestational age, weeks	38.6 ± 2.1	39.0 ± 1.6	< 0.001	1,257	38.4 ± 1.8	39.0 ± 1.6	< 0.001	2,020
Conception by IVF, %		33 (9.4)	298 (4.6)	< 0.001	62	19 (7.1)	415 (4.4)	0.052	335
MAP at 10–13 weeks of gestation, mmHg		89.3 ± 11.5	80.4 ± 9.3	< 0.001	973	91.8 ± 11.7	79.6 ± 9.3	< 0.001	1,328
PE: preeclampsia; BMI: body mass index; CH: chronic hypertension; DM: diabetes mellitus; SLE: systemic lupus erythematosus; HDP: hypertensive disorders of pregnancy; IVF: in vitro fertilization; and MAP: mean arterial pressure. Data are shown as mean ± standard deviation for continuous variables and n (%) for categorical variables. P-values were calculated using the chi-square or Fisher’s exact test for categorical variables and the t-test for continuous variables. Missing values were imputed using the k-nearest neighbor imputation with k = 140 (square root of the total study population).

In association studies on the maternal cohorts (Table 2), maternal SBP-, DBP-, and PE-PRSs were significantly associated with PE onset in the meta-analysis; odds ratio (OR) and 95% CI per 1 standard deviation (SD): 1.14 (1.05 to 1.24), 1.20 (1.11 to 1.31), and 1.10 (1.01 to 1.19), respectively, with significant heterogeneity (P for heterogeneity = 0.001, 0.003, and 0.030, respectively). In association studies for the parental cohorts (Supplementary Table S5), maternal and paternal PRSs were not related to PE onset in the meta-analysis. Only in the parental external validation cohort, paternal SBP-PRS and paternal DBP-PRS were associated with PE onset; OR and 95% CI per 1 SD: 1.93 (1.32 to 2.83) and 1.91 (1.29 to 2.84), respectively. Similar results were observed after excluding participants with CH. However, some results did not converge in the parental external validation cohort because of the small number of outcomes (Supplementary Tables S6 and S7). No interaction was noted between the maternal and paternal PRSs (data not shown). The associations of PRSs with BP during early pregnancy were similar to their associations with PE (Supplementary Table S8). When early- and late-onset PE were compared in relation to PRS, the effect sizes were generally comparable (Supplementary Table S9).

Table 2

Relationship between maternal polygenic risk scores and preeclampsia onset in the maternal cohorts by logistic regression analysis
		Internal validation cohort		External validation cohort		Meta-analysis
		OR (95% CI)	P-value	OR (95% CI)	P-value	OR (95% CI)	P-value	P for heterogeneity
PRS for SBP
	Continuous	1.01 (0.91 to 1.13)	0.794	1.33 (1.17 to 1.50)	< 0.001	1.14 (1.05 to 1.24)	0.002	0.001
	Tertile 1	Reference		Reference		Reference
	Tertile 2	1.09 (0.83 to 1.44)	0.532	1.02 (0.76 to 1.39)	0.874	1.06 (0.86 to 1.30)	0.570	0.761
	Tertile 3	0.96 (0.73 to 1.26)	0.768	1.61 (1.21 to 2.16)	0.001	1.22 (1.00 to 1.49)	0.046	0.010
PRS for DBP
	Continuous	1.08 (0.96 to 1.21)	0.186	1.39 (1.22 to 1.58)	< 0.001	1.20 (1.11 to 1.31)	< 0.001	0.003
	Tertile 1	Reference		Reference		Reference
	Tertile 2	0.80 (0.59 to 1.08)	0.150	1.33 (0.98 to 1.79)	0.063	1.03 (0.84 to 1.28)	0.768	0.020
	Tertile 3	1.05 (0.80 to 1.37)	0.748	2.18 (1.62 to 2.94)	< 0.001	1.46 (1.19 to 1.78)	< 0.001	0.000
PRS for PE
	Continuous	1.01 (0.91 to 1.13)	0.817	1.21 (1.07 to 1.37)	0.002	1.10 (1.01 to 1.19)	0.026	0.030
	Tertile 1	Reference		Reference		Reference
	Tertile 2	1.05 (0.81 to 1.35)	0.705	1.34 (0.96 to 1.88)	0.085	1.15 (0.94 to 1.41)	0.179	0.252
	Tertile 3	1.04 (0.79 to 1.36)	0.792	1.49 (1.08 to 2.05)	0.014	1.20 (0.98 to 1.48)	0.076	0.087
PRS: polygenic risk score; PE: preeclampsia; OR: odds ratio; CI: confidence interval; SBP: systolic blood pressure; and DBP: diastolic blood pressure. Two models were developed, one with PRS as a continuous value and the other as tertile values. All results were adjusted for maternal age at conception and four genetic principal components. P for heterogeneity was calculated by the Cochran Q test.

After parameter optimization in the PRS training cohort, the predictive performances of SBP-, DBP-, and PE-PRS calculated using the two methods, C + T and LDpred2, were compared in the maternal internal and external validation cohorts (Table 3). The DBP-PRS calculated using LDpred2 improved the discrimination performance the most in both the internal (0.571 in the model with PRS versus 0.567 in the model without PRS) and external (0.591 in the model with PRS versus 0.550 in the model without PRS) validation. The simulation study revealed that among the PRSs predicted by LDpred2, those for DBP were the most frequent with the highest predictive power in the external validation cohort (Supplementary Table S10).

Table 3

Prediction performance with maternal polygenic risk score derived from three phenotypes using two methods
		LDpred2	C + T	LDpred2	C + T	LDpred2	C + T	No PRS
C-statistics
	Apparent	0.573 (0.538 to 0.606)	0.574 (0.539 to 0.607)	0.580 (0.544 to 0.613)	0.576 (0.540 to 0.608)	0.576 (0.541 to 0.610)	0.574 (0.539 to 0.606)	0.574 (0.539 to 0.606)
	Internal validation	0.566	0.566	0.571	0.567	0.567	0.565	0.567
	External validation	0.564 (0.523 to 0.605)	0.546 (0.506 to 0.590)	0.591 (0.550 to 0.631)	0.543 (0.503 to 0.587)	0.557 (0.515 to 0.600)	0.550 (0.509 to 0.596)	0.550 (0.510 to 0.594)
Calibration slope
	Apparent	1.005 (0.582 to 1.369)	1.006 (0.566 to 1.379)	1.005 (0.570 to 1.349)	1.006 (0.573 to 1.381)	1.005 (0.559 to 1.378)	1.007 (0.569 to 1.376)	1.007 (0.568 to 1.375)
	Internal validation	0.827	0.824	0.832	0.827	0.822	0.822	0.853
	External validation	0.952 (0.462 to 1.400)	0.796 (0.297 to 1.278)	1.112 (0.645 to 1.520)	0.777 (0.272 to 1.256)	0.901 (0.388 to 1.360)	0.842 (0.341 to 1.312)	0.838 (0.334 to 1.312)
PRS: polygenic risk score. PRSs were calculated for three phenotypes, systolic blood pressure, diastolic blood pressure, and preeclampsia, using two methods, LDpred2 and C + T. Models without PRS included maternal age and four genetic principal components as predictive variables, and the others included additional PRSs. 95% confidence intervals and internal validation used the bootstrap method with 1,000 resamplings.

Thereafter, clinical prediction models were developed with maternal and paternal DBP-PRSs calculated using LDpred2. In the maternal cohort, models including maternal PRS demonstrated superior predictive discrimination compared to models without maternal PRS in the reference and family-history models in internal validation. This was also observed in all four external validation models, namely the reference, family-history, at-pregnancy-confirmation, and in-early-pregnancy models (Table 4). For example, in external validation, the C-statistics of the at-pregnancy-confirmation model were 0.776 without PRS and 0.783 with maternal PRS, indicating the utility of maternal PRS at pregnancy confirmation. Even in the in-early-pregnancy model, which included BP in early pregnancy, the C-statistics improved slightly by including maternal PRS (0.849 without PRS versus 0.850 with maternal PRS), suggesting the utility of maternal PRS in the prediction of early pregnancy. In the parental internal validation cohort (Table 5), paternal PRS, but not maternal PRS, improved the reference, family-history, and at-pregnancy-confirmation models. In the parental external validation cohort, maternal, but not paternal, PRS improved the reference, family-history, and At-pregnancy-confirmation models, although C-statistics and calibration had a wide 95% CI and the results were not stable. Models with both maternal and paternal PRSs did not demonstrate better discrimination ability than those with either maternal or paternal PRS (Table 5). C-statistics differences and 95% CIs between models with and without PRSs in the maternal and paternal external validation cohorts calculated by the bootstrap method are presented in Supplementary Table S11.

Table 4

Performance of prediction models with maternal polygenic risk score, baseline characteristics, and blood pressure in early pregnancy in the maternal cohorts
			No PRS	M-PRS
C-statistics
	Apparent
		Reference model	0.561 (0.533 to 0.592)	0.580 (0.544 to 0.613)
		Family-history model	0.564 (0.535 to 0.595)	0.579 (0.542 to 0.613)
		At-pregnancy-confirmation model	0.735 (0.703 to 0.769)	0.741 (0.709 to 0.775)
		In-early-pregnancy model	0.781 (0.751 to 0.811)	0.783 (0.752 to 0.812)
	Internal validation
		Reference model	0.561	0.571
		Family-history model	0.564	0.572
		At-pregnancy-confirmation model	0.730	0.729
		In-early-pregnancy model	0.777	0.776
	External validation
		Reference model	0.576 (0.545 to 0.612)	0.591 (0.550 to 0.631)
		Family-history model	0.571 (0.539 to 0.606)	0.584 (0.544 to 0.623)
		At-pregnancy-confirmation model	0.776 (0.740 to 0.810)	0.783 (0.747 to 0.817)
		In-early-pregnancy model	0.849 (0.820 to 0.874)	0.850 (0.820 to 0.877)
Calibration slope
	Apparent
		Reference model	1.002 (0.543 to 1.407)	1.005 (0.570 to 1.349)
		Family-history model	1.001 (0.555 to 1.408)	0.999 (0.573 to 1.370)
		At-pregnancy-confirmation model	0.998 (0.894 to 1.111)	1.000 (0.895 to 1.110)
		In-early-pregnancy model	1.000 (0.903 to 1.098)	0.999 (0.904 to 1.103)
	Internal validation
		Reference model	0.999	0.832
		Family-history model	0.963	0.835
		At-pregnancy-confirmation model	0.955	0.943
		In-early-pregnancy model	0.956	0.947
	External validation
		Reference model	0.576 (0.545 to 0.612)	0.591 (0.550 to 0.631)
		Family-history model	0.571 (0.539 to 0.606)	0.584 (0.544 to 0.623)
		At-pregnancy-confirmation model	0.776 (0.740 to 0.810)	0.783 (0.747 to 0.817)
		In-early-pregnancy model	0.849 (0.820 to 0.874)	0.850 (0.820 to 0.877)
PRS: polygenic risk score. “No PRS” indicates models without PRS and “M-PRS” indicates models with maternal PRS for diastolic blood pressure. Reference models included maternal age. Family-history models included maternal family history of hypertensive disorders of pregnancy and maternal age. At-pregnancy-confirmation models included variables available at pregnancy confirmation: maternal age, pre-pregnancy body mass index, chronic hypertension, systemic lupus erythematosus, diabetes mellites, maternal family history of hypertensive disorders of pregnancy, conception via in vitro fertilization, parity, gestational age at the previous delivery, and the inter-birth interval. At-early-pregnancy models added blood pressure in early pregnancy to at-pregnancy-confirmation models. Models with M-PRS also included maternal four genetic principal components. 95% confidence intervals and internal validation used the bootstrap method with 1,000 resamplings.

Table 5

Prediction performance with parental polygenic risk score, baseline characteristics, and blood pressure in early pregnancy in the parental cohorts
			No PRS	M-PRS	P-PRS	M-PRS and P-PRS
C-statistics
	Apparent
		Reference model	0.567 (0.531 to 0.606)	0.581 (0.538 to 0.623)	0.609 (0.568 to 0.651)	0.609 (0.568 to 0.649)
		Family-history model	0.572 (0.534 to 0.611)	0.583 (0.539 to 0.625)	0.614 (0.571 to 0.657)	0.614 (0.570 to 0.655)
		At-pregnancy-confirmation model	0.722 (0.682 to 0.763)	0.728 (0.686 to 0.765)	0.737 (0.694 to 0.778)	0.737 (0.693 to 0.776)
		In-early-pregnancy model	0.772 (0.734 to 0.808)	0.770 (0.733 to 0.804)	0.776 (0.739 to 0.813)	0.776 (0.739 to 0.811)
	Internal validation
		Reference model	0.567	0.568	0.598	0.593
		Family-history model	0.572	0.571	0.600	0.597
		At-pregnancy-confirmation model	0.711	0.708	0.721	0.718
		In-early-pregnancy model	0.763	0.758	0.762	0.759
	External validation
		Reference model	0.606 (0.491 to 0.723)	0.633 (0.497 to 0.746)	0.496 (0.356 to 0.648)	0.530 (0.397 to 0.665)
		Family-history model	0.597 (0.484 to 0.712)	0.624 (0.490 to 0.738)	0.488 (0.349 to 0.637)	0.521 (0.390 to 0.657)
		At-pregnancy-confirmation model	0.935 (0.857 to 0.985)	0.938 (0.864 to 0.983)	0.937 (0.872 to 0.982)	0.936 (0.872 to 0.982)
		In-early-pregnancy model	0.979 (0.962 to 0.989)	0.977 (0.958 to 0.988)	0.976 (0.961 to 0.987)	0.973 (0.955 to 0.986)
Calibration slope
	Apparent
		Reference model	0.998 (0.450 to 1.462)	0.997 (0.551 to 1.407)	0.992 (0.645 to 1.349)	0.997 (0.659 to 1.304)
		Family-history model	1.000 (0.499 to 1.445)	1.000 (0.585 to 1.378)	0.999 (0.634 to 1.343)	0.995 (0.660 to 1.303)
		At-pregnancy-confirmation model	0.999 (0.849 to 1.169)	1.000 (0.850 to 1.165)	0.998 (0.845 to 1.172)	0.998 (0.845 to 1.171)
		In-early-pregnancy model	0.997 (0.867 to 1.147)	0.997 (0.867 to 1.144)	0.998 (0.867 to 1.156)	0.998 (0.867 to 1.150)
	Internal validation
		Reference model	1.001	0.794	0.822	0.742
		Family-history model	0.945	0.809	0.781	0.723
		At-pregnancy-confirmation model	0.920	0.900	0.880	0.861
		In-early-pregnancy model	0.927	0.908	0.888	0.871
	External validation
		Reference model	1.142 (− 0.253 to 2.137)	0.944 (− 0.329 to 1.923)	0.261 (− 0.924 to 1.211)	0.268 (− 0.785 to 1.154)
		Family-history model	0.802 (− 0.421 to 1.633)	0.686 (− 0.423 to 1.529)	0.130 (− 1.006 to 0.989)	0.158 (− 0.842 to 0.946)
		At-pregnancy-confirmation model	1.556 (1.313 to 1.879)	1.519 (1.281 to 1.839)	1.508 (1.269 to 1.819)	1.473 (1.231 to 1.793)
		In-early-pregnancy model	1.479 (1.263 to 1.792)	1.410 (1.203 to 1.721)	1.450 (1.232 to 1.760)	1.386 (1.176 to 1.697)
PRS: polygenic risk score. “No PRS,” “M-PRS,” “P-PRS,” and “M-PRS and P-PRS” indicate models without PRS, models with maternal PRS for diastolic blood pressure, models with paternal PRS for diastolic blood pressure, and models with both maternal and paternal PRS for diastolic blood pressure, respectively. Reference models included maternal age. Family-history models included maternal family history of hypertensive disorders of pregnancy and maternal age. At-pregnancy-confirmation models included variables available at pregnancy confirmation: maternal age, pre-pregnancy body mass index, chronic hypertension, systemic lupus erythematosus, diabetes mellitus, maternal family history of hypertensive disorders of pregnancy, conception by in vitro fertilization, parity, gestational age at the previous delivery, and the inter-birth interval. At-early-pregnancy models added blood pressure in early pregnancy to at-pregnancy-confirmation models. Paternal age at conception was included in all the models with paternal PRS and paternal family history of hypertensive disorders of pregnancy was included in all the models with paternal PRS except for reference models. Models with M-PRS and P-PRS also included maternal and paternal four genetic principal components, respectively. 95% confidence intervals and internal validation used the bootstrap method with 1,000 resamplings

The maternal SBP-, DBP-, and PE-PRSs were associated with PE onset in the meta-analysis, with significant heterogeneity. The paternal SBP-, DBP-, and PE-PRSs were associated with PE onset only in the external validation cohort. Maternal DBP-PRS calculated using LDpred2 improved prediction models the most. Maternal DBP-PRS improved both at-pregnancy-confirmation and in-early-pregnancy models, indicating the clinical utility of maternal PRS. Paternal PRS improved prediction models in the parental internal validation cohort.

The present study shows that maternal PRSs for BP were associated with PE onset in both internal and external validation cohorts, consistent with previous studies in the European population,[9, 10] indicating cross-ethnicity generalizability. Regarding the onset time, the association between PRS and PE was not considerably affected, suggesting that PRSs are not pathology-specific biomarkers and broadly capture the genetic predisposition to hypertension. Maternal PRS improved the prediction model beyond a family history of HDP. While family history provides plentiful information on genetic predisposition and environment,[15] the amount of information is determined by family structure and relationships, as well as by disease prevalence. Owing to the demographic trend toward smaller families[35] and the need to obtain data retrospectively because of the transient nature of the condition during pregnancy, obtaining a comprehensive family history of HDP is difficult. Herein, < 2% of the participants and < 1% of the paternal parents of their fetuses reported a family history of HDP. Moreover, reliance only on the binary categorization of family history may result in overlooking the intricate genetic predisposition. Therefore, incorporating PRSs into risk prediction models in addition to family history is crucial for precision medicine. Maternal PRS improved the at-pregnancy-confirmation model and slightly improved the in-early-pregnancy model. This indicates that the maternal PRS is useful in clinical practice at pregnancy confirmation and even in early pregnancy, although its predictive significance is diminished by the fact that some genetic predisposition for high BP may manifest itself as the actual measured BP values in early pregnancy. In contrast, PE-RPS was not associated with the development of PE in this meta-analysis. The genome-wide association study (GWAS) for PE[36] used in our study had a relatively small number of cases, and the number of East Asian populations in the GWAS was limited (123 in the BioBank Japan, 1,031 in the UK Biobank, and 1,324 in the FinnGen), which might result in failure to reflect genetic PE predisposition.

Furthermore, in this study, paternal PRSs for BP were associated with PE onset in the external validation cohort. The paternal genetic contribution to PE onset has been predicted in family[13] and genetic studies.[37] However, as of date, the association between paternal PRS and PE onset has not been investigated. Our results were inconsistent between the internal and external validation cohorts. One possible reason for this is the biological differences between the parental internal and external validation cohorts. Placental and maternal factors are involved in the development of PE.[38] The baseline characteristics of the parental internal and external cohorts were quite different; the prevalence of CH in participants with PE was much higher in the external validation cohort than in the internal validation cohort (79.3% versus 25.0%). This may have led to differences between the two cohorts in terms of the ratio of placental to maternal contributions, resulting in different associations between paternal PRSs and PE. Paternal PRS improved the prediction models in the parental internal validation cohort, despite no significant association in the association analyses. The prediction models developed in the parental internal validation cohort did not have predictive power in the parental external validation cohort possibly, because of cohort differences, as mentioned previously. This suggests that the contribution of paternal PRS to PE onset may differ across populations and should be considered when developing predictive models.

Clinical and Research Implications

Risk prediction is crucial in clinical practice because it can lead to preventive interventions, allowing for intensive perinatal management, which facilitates early detection and treatment. Genomic information remains unchanged and can be used throughout life. This is a crucial difference from other biomarkers,[25] which are only reflective of temporary conditions. In addition, PRS is useful for predicting diseases other than those in the perinatal period.[39, 40] Given the unchangeability of genomic information, our study suggests that genomic information should be obtained during relatively healthy childbearing age rather than during middle or old age to avoid poor outcomes in pregnant women due to inadequate prediction of PE. We showed that maternal PRS can improve clinical PE prediction models and suggest its application in other perinatal diseases. In addition, we showed that the utility of paternal genomic information in predicting PE is inconsistent across cohorts; therefore, further studies should aim at identifying populations in which paternal PRS has predictive information for PE.

Strengths and Limitations

Our study is the first to investigate the utility of maternal and paternal PRS for PE prediction. We applied a well-validated prediction model; therefore, our results are close to clinical application. Two cohorts with genomic data from different genotype array platforms and baseline characteristics enabled us to examine the generalizability of the results.

Despite these strengths, our study had four limitations. First, our model did not include effective biomarkers specific to PE, such as the uterine artery pulsatility index and placental growth factor,[25] which may have improved the performance. However, given the widespread use of genomic information beyond perinatal diseases, prediction models using PRS may be more clinically applicable than those using disease-specific biomarkers. Second, there were several differences between the internal and external validation cohorts. The external validation cohort included more participants with CH, which was the strongest predictive variable, than the internal validation cohort, resulting in calibration slopes of > 1 in the external validation. Moreover, in the external cohort, a family history of HDP was not associated with PE, limiting the purpose of this study to confirm the predictive value of the PRS in addition to a family history of HDP. The differences between the two cohorts occurred possibly because of differences in the source populations, and although the study does not provide sufficient replication, it does provide some insight into the type of population for which the predictive model is useful. Third, because there is no paternal GWAS on PE, we had no choice but to apply the maternal GWAS to the fathers. Fourth, owing to the small number of parental cohorts, the effectiveness of paternal PRS in predicting PE may have been overlooked. Optimization was conducted only for maternal PRS, which may have resulted in an underestimation of the ability of paternal PRS to predict PE. In conclusion, parental PRS, along with clinical predictive variables, is potentially useful for predicting PE.

Data availability statement

The datasets used during the current study available from the corresponding author on reasonable request.

Acknowledgments: The authors would like to thank all the participants who consented to participate in this study and all the staff at the Tohoku Medical Megabank Organization, Tohoku University, Iwate Tohoku Medical Megabank Organization, and Iwate Medical University. A full list of the members of the Tohoku Medical Megabank Organization is available at https://www.megabank.tohoku.ac.jp/english/a230901/.

Author contributions

Conceptualization: H.O., M.I., and T.O. Methodology: H.O. and M.I. Visualization: H.O.

Supervision: S.K.

Writing – original draft: H.O. and M.I.

Writing – review & editing: All the authors

Sources of Funding: This work was supported by the Japan Agency for Medical Research and Development, Japan (Grant Nos. JP19gk0110039, JP17km0105001, JP21tm0124005, and JP21tm0424601) and Japan Society for the Promotion of Science KAKENHI (Grant No. JP21K10438).

Additional Information

K.M. is an employee of the Ministry of Education, Culture, Sports, Science and Technology, Japan. The other authors have no conflict of interests.

Paper presentation information

This manuscript has not been published in any journal or presented at any conference in part or in entirety and is not under consideration by another journal. It has been made publicly available as a preprint (DOI: 10.1101/2024.02.07.24302476).

Ananth, C. V., Keyes, K. M. & Wapner, R. J. Pre-eclampsia rates in the united states, 1980-2010: Age-period-cohort analysis. BMJ. 347, f6564 (2013).
Shiozaki, A., Matsuda, Y., Satoh, S. & Saito, S. Comparison of risk factors for gestational hypertension and preeclampsia in japanese singleton pregnancies. J Obstet Gynaecol Res. 39, 492-499 (2013).
GBD 2015 Maternal Mortality Collaborators. Global, regional, and national levels of maternal mortality, 1990-2015: A systematic analysis for the global burden of disease study 2015. Lancet. 388, 1775-1812 (2016).
Davenport, M. H. et al. Prenatal exercise for the prevention of gestational diabetes mellitus and hypertensive disorders of pregnancy: A systematic review and meta-analysis. Br J Sports Med. 52, 1367-1375 (2018).
Rolnik, D. L. et al. Aspirin versus placebo in pregnancies at high risk for preterm preeclampsia. N Engl J Med. 377, 613-622 (2017).
Woo Kinshella, M. L. et al. Calcium for pre-eclampsia prevention: A systematic review and network meta-analysis to guide personalised antenatal care. BJOG. 129, 1833-1843 (2022).
Magee, L. A., Nicolaides, K. H. & von Dadelszen, P. Preeclampsia. N Engl J Med. 386, 1817-1832 (2022).
De Kat, A. C., Hirst, J., Woodward, M., Kennedy, S. & Peters, S. A. Prediction models for preeclampsia: A systematic review. Pregnancy Hypertens. 16, 48-66 (2019).
Kivioja, A. et al. Increased risk of preeclampsia in women with a genetic predisposition to elevated blood pressure. Hypertension. 79, 2008-2015 (2022).
Nurkkala, J. et al. Associations of polygenic risk scores for preeclampsia and blood pressure with hypertensive disorders of pregnancy. J Hypertens. 41, 380-387 (2023).
Honigberg, M. C. et al. Polygenic prediction of preeclampsia and gestational hypertension. Nat Med. 29, 1540–1549 (2023).
Tyrmi, J. S. et al. Genetic risk factors associated with preeclampsia and hypertensive disorders of pregnancy. JAMA Cardiol. 8, 674–683. (2023).
Cnattingius, S., Reilly, M., Pawitan, Y. & Lichtenstein, P. Maternal and fetal genetic factors account for most of familial aggregation of preeclampsia: A population-based swedish cohort study. Am J Med Genet A. 130A, 365-371 (2004).
Wright, A., Wright, D., Ispas, C. A., Poon, L. C. & Nicolaides, K. H. Mean arterial pressure in the three trimesters of pregnancy: Effects of maternal characteristics and medical history. Ultrasound Obstet Gynecol. 45, 698-706 (2015).
Valdez, R., Yoon, P. W., Qureshi, N., Green, R. F. & Khoury, M. J. Family history in public health practice: A genomic tool for disease prevention and health promotion. Annu Rev Public Health. 31, 69-87 (2010).
Bennett, R. L. The family medical history. Prim Care. 31, 479-495 (2004).
Mars, N. et al. Systematic comparison of family history and polygenic risk across 24 common diseases. Am J Hum Genet. 109, 2152-2162 (2022).
Kovacheva, V. P. et al. Preeclampsia prediction using machine learning and polygenic risk scores from clinical and genetic risk factors in early and late pregnancies. Hypertension. 81, 264-272 (2024).
Kuriyama, S. et al. The tohoku medical megabank project: Design and mission. J Epidemiol. 26, 493-511 (2016).
Kuriyama, S. et al. Cohort profile: Tohoku medical megabank project birth and three-generation cohort study (TMM BirThree Cohort Study): Rationale, progress and perspective. Int J Epidemiol. 49, 18-19m (2020).
Yamada, M. et al. Genetic loci for lung function in japanese adults with adjustment for exhaled nitric oxide levels as airway inflammation indicator. Commun Biol. 4, 1288 (2021).
Ohseto, H. et al. Preeclampsia prediction model using the dipstick test for proteinuria during early gestation. Hypertens Res Pregnancy. 10, 88-96 (2022).
Ohseto, H. et al. Dietary calcium intake was related to the onset of pre-eclampsia: The tmm birthree cohort study. J Clin Hypertens (Greenwich). 25, 61-70 (2023).
Wright, D., Syngelaki, A., Akolekar, R., Poon, L. C. & Nicolaides, K. H. Competing risks model in screening for preeclampsia by maternal characteristics and medical history. Am J Obstet Gynecol. 213, 62 e61-62 e10 (2015).
O'Gorman, N. et al. Competing risks model in screening for preeclampsia by maternal factors and biomarkers at 11-13 weeks gestation. Am J Obstet Gynecol. 214, 103 e101-103 e112 (2016).
Bartsch, E., Medcalf, K. E., Park, A. L., Ray, J. G. & High Risk of Pre-eclampsia Identification, G. Clinical risk factors for pre-eclampsia determined in early pregnancy: Systematic review and meta-analysis of large cohort studies. BMJ. 353, i1753 (2016).
Choi, S. W., Mak, T. S. & O'Reilly, P. F. Tutorial: A guide to performing polygenic risk score analyses. Nat Protoc. 15, 2759-2772 (2020).
Prive, F., Arbel, J. & Vilhjalmsson, B. J. Ldpred2: Better, faster, stronger. Bioinformatics. 36, 5424-5431 (2020).
Mizuno, S. et al. Development of phenotyping algorithms for hypertensive disorders of pregnancy (hdp) and their application in more than 22,000 pregnant women. Sci Rep. 14, 6292 (2024).
Hypertension in pregnancy. Report of the american college of obstetricians and gynecologists' task force on hypertension in pregnancy. Obstet Gynecol. 122, 1122-1131 (2013).
Wright, D., Wright, A. & Nicolaides, K. H. The competing risk approach for prediction of preeclampsia. Am J Obstet Gynecol. 223, 12-23 e7 (2020).
Wright, D. et al. Predictive performance of the competing risk model in screening for preeclampsia. Am J Obstet Gynecol. 220, 199 e1-199 e13 (2019).
Zhang, Z. Introduction to machine learning: K-nearest neighbors. Ann Transl Med. 4, 218 (2016).
Rahman, M. S., Ambler, G., Choodari-Oskooei, B. & Omar, R. Z. Review and evaluation of performance measures for survival prediction models in external validation settings. BMC Med Res Methodol. 17, 60 (2017).
United Nations Department of Economic and Social Affairs Population Division. Patterns and trends in household size and composition: Evidence from a united nations dataset. ST/ESA/SER.A/433 (2019).
Sakaue, S. et al. A cross-population atlas of genetic associations for 220 human phenotypes. Nat Genet. 53, 1415-1424 (2021).
Galaviz-Hernandez, C. et al. The paternal polymorphism rs5370 in the edn1 gene decreases the risk of preeclampsia. Pregnancy Hypertens. 6, 327-332 (2016).
Burton, G. J., Redman, C. W., Roberts, J. M. & Moffett, A. Pre-eclampsia: Pathophysiology and clinical implications. BMJ. 366, l2381 (2019).
Briggs, S. E. W. et al. Integrating genome-wide polygenic risk scores and non-genetic risk to predict colorectal cancer diagnosis using uk biobank data: Population based cohort study. BMJ. 379, e071707 (2022).
Wedekind, L. E. et al. The utility of a type 2 diabetes polygenic score in addition to clinical variables for prediction of type 2 diabetes incidence in birth, youth and adult cohorts in an indigenous study population. Diabetologia. 66, 847-860 (2023).
Delaneau O, Zagury JF, Marchini J. Improved whole-chromosome phasing for disease and population genetic studies. Nat Methods. 10, 5–6 (2013).
O'connell J, et al. A general approach for haplotype phasing across the full spectrum of relatedness. PLoS Genet. 10, e1004234 (2014).
Bycroft C, et al. The UK Biobank resource with deep phenotyping and genomic data. Nature. 562, 203-209 (2018).
Tadaka S, et al. 3.5KJPNv2: an allele frequency panel of 3552 Japanese individuals including the X chromosome. Hum Genome Var. 6, 28 (2019).
Genomes Project Consortium, et al. A global reference for human genetic variation. Nature. 526, 68–74 (2015).
Howie BN, Donnelly P, Marchini J. A flexible and accurate genotype imputation method for the next generation of genome-wide association studies. PLoS Genet. 5, e1000529 (2009).

Competing interest reported. K.M. is an employee of the Ministry of Education, Culture, Sports, Science and Technology, Japan. The other authors have no conflict of interests.

Supplementary.docx

Download PDF

Reviews received at journal
15 Sep, 2024
Reviewers agreed at journal
13 Sep, 2024
Reviewers agreed at journal
11 Sep, 2024
Reviewers invited by journal
08 Sep, 2024
Editor assigned by journal
03 Sep, 2024
Editor invited by journal
12 Aug, 2024
Submission checks completed at journal
08 Aug, 2024
First submitted to journal
08 Aug, 2024

You are reading this latest preprint version

Preeclampsia prediction with maternal and paternal polygenic risk scores: the TMM BirThree Cohort Study

Status:

Version 1

Abstract

Figures

Introduction

Methods

Participants

Predictive variables

Polygenic risk score

Outcome measurement

Model development

Statistical analysis

Results

Discussion

Clinical and Research Implications

Strengths and Limitations

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1