Molecular subtyping of hypertensive disorders of pregnancy

Abstract

Hypertensive disorders of pregnancy (HDP), including preeclampsia, affect 1 in 6 pregnancies, are major contributors to maternal morbidity and mortality, yet lack precision medicine strategies. Analyzing transcriptomic data from a prospectively-collected diverse cohort (n = 9102), this study reveals distinct RNA subtypes in maternal blood, reclassifying clinical HDP phenotypes like early/late-onset preeclampsia. The placental gene PAPPA2 strongly predicts the most severe forms of preeclampsia in individuals without pre-existing high risk factors, months before symptoms, and its overexpression correlates with earlier delivery in a dose-dependent manner. Further, molecular subtypes characterized by immune genes are upregulated in less severe forms of HDP. These results reclassify HDP clinical phenotypes into two distinct molecular subtypes, placental-associated or immune-associated. Validation performance for placental-associated HDP yields an AUC of 0.88 in the advanced maternal age population without pre-existing high risk factors. Molecular subtypes create new opportunities to apply precision-based medicine in maternal health.

Fig. 1. CONSORT diagram for Miracle of Life study.

HDP hypertensive disorders of pregnancy, GHTN gestational hypertension, PE preeclampsia. *For sites where source document verification was possible, otherwise site is only used in training.

Fig. 2. Effect size of PAPPA2 in different populations.

a Samples are grouped by their clinical phenotype (diagnosis time <37w, delivery time <37w, and/or presentation of severe features), into 8 groups and non HDP controls. 13 samples were missing sufficient information to include in subgroups not shown. b Effect size of PAPPA2 is shown next to each clinical phenotype, blending into molecular subgroups, placental-associated or immune-associated. Statistical comparisons of placental-associated or immune-associated versus non HDP (baseline) calculated using a two-sided Mann–Whitney U test. c Effect size of PAPPA2 for placental-associated and immune-associated HDP split by presence/absence of maternal high risk factors (RF). Statistical comparisons of each subgroup versus the non HDP (baseline) with the corresponding presence/absence of maternal high RF calculated using a two-sided Mann–Whitney U test. d PAPPA2 expression in early-onset PE split into placental-associated and immune-associated HDP among individuals with no high risk factors. e PAPPA2 expression in ante- and intrapartum PE with severe features split into placental-associated and immune-associated HDP among individuals with no high risk factors. Box and whisker plots display a box with lower bound, center line, and upper bound at the 25th, 50th, and 75th percentile, respectively, where whiskers extend to 1.5 times the inter-quartile range with points outside this range depicted individually as circles. Statistical comparisons depicted in box and whisker plots are calculated using a two-sided Mann–Whitney U test. HDP hypertensive disorders of pregnancy. Combined mean effect size for PAPPA2 for PE is represented in white, above-average in reds and below average in blues.

Fig. 3. Differential gene expression analyses in immune-associated HDP.

Quantile-quantile (QQ) plot of ranked Mann–Whitney U p-values, colored by absolute Cohen’s d effect size for a immune-associated PE versus non HDP (genomic inflation factor λ = 1.2, 29 significantly differentially expressed genes), b immune-associated GHTN versus non HDP (λ = 1.8, 38 significantly differentially expressed genes). Top 10 genes are named, and gene names are bolded based on overlap with immune and inflammatory GO biological processes. Statistical comparisons in QQ plots are calculated using a two-sided Mann–Whitney U test with multiple test correction by Benjamini–Hochberg. c Overlap in differentially expressed genes. Significance of the overlap was evaluated using a hypergeometric test, where the reported probability corresponds to the likelihood of observing an overlap of a given size or larger by random chance under the null hypothesis that the sets are independent. PE, preeclampsia; GHTN, gestational hypertension; HDP, hypertensive disorders of pregnancy.

Fig. 4. Differential gene expression and classifier predictions for placental-associated HDP.

Quantile-quantile (QQ) plot of ranked Mann–Whitney U p-values and colored by absolute Cohen’s d effect size for a placental-associated HDP versus all others (genomic inflation factor λ = 2.2), and b placental-associated HDP versus all others with no high risk factors (λ = 5.4). Statistical comparisons in QQ plots are calculated using a two-sided Mann–Whitney U test with multiple test correction by Benjamini–Hochberg. c Effect size for classifier genes in a population with no high risk factors, split into clinical categories based on diagnosis time <37wks, delivery time <37wks, and/or presentation of severe features. 11 samples were missing sufficient information to include in subgroups and were thus excluded from this plot. d Classifier prediction across placental-associated HDP by gestational age at delivery in individuals with no high risk factors. HDP Hypertensive disorders of pregnancy.

Fig. 5. Validation of the full classifier in all individuals with no high risk factors.

a Receiver operator characteristic (ROC) for all individuals with predicted outcome as placental-associated HDP (blue); all individuals with predicted outcome as placental-associated HDP and ≤35 week delivery (green); individuals with advanced maternal age (≥35 years) with predicted outcome as placental-associated HDP (orange). Performance at the prespecified cutpoint is annotated with a X on each curve. b Comparison of the positive likelihood ratio for each of the groups with USPSTF performance shown in striped lighter tone and the full classifier in solid color matching the ROC. AMA advanced maternal age (≥35 years), HRF High Risk Factor, USPSTF United States Preventive Services Task Force.