RNA profiles reveal signatures of future health and disease in pregnancy

The study addresses the need for early, accurate assessment of pregnancy progression and prediction of adverse outcomes, particularly pre-eclampsia, a condition characterized by maternal endothelial dysfunction, new-onset hypertension, and increased lifetime cardiovascular risk. Human pregnancy research faces ethical constraints and limited applicability of animal models due to unique features of human placentation. The authors hypothesize that cfRNA in maternal plasma can capture dynamic molecular signals from maternal, placental, and fetal tissues, enabling accurate tracking of gestational progression and early identification of deviations indicative of impending pathology such as pre-eclampsia, independent of traditional clinical risk factors.

Ultrasound biometry is the standard for gestational age estimation but provides limited biological insight and reduced accuracy later in pregnancy. Conventional animal models have limited relevance for human-specific placentation and multifactorial pre-eclampsia pathogenesis. Prior biomarker studies and reviews have called for clearer evidence to support clinical utility of circulating biomarkers in pregnancy. The authors reference gene set resources (Gene Ontology, Molecular Signatures Database) and prior studies on placental biology and pre-eclampsia-related genes, noting prior associations (e.g., PAPPA2, CLDN family, inflammatory mediators) but a lack of comprehensive, longitudinal, non-invasive transcriptomic profiling spanning diverse populations and early asymptomatic stages.

• Cohorts and samples: Eight prospectively collected cohorts (labelled A–H) provided 2,539 plasma samples from 1,840 pregnancies across multiple ethnicities and geographies, spanning a broad range of gestational ages. A detailed description and methods are provided in Supplementary Information.
• cfRNA processing and normalization: Plasma cfRNA transcript counts were standardized across cohorts (mean-centering; Extended Data Fig. 1) prior to modeling and analyses.
• Gestational age (GA) modeling: Data from full-term pregnancies were split into training (n=1,908 samples) and test (n=479 samples) sets, stratified by GA. A Lasso linear model was trained to predict GA referenced to first-trimester ultrasound biometry. Model performance was evaluated on the hold-out test set. Analysis of variance (ANOVA) assessed contributions of clinical variables (BMI, maternal age, race) vs cfRNA transcripts; clinical variables were also included in alternative models to test for added value.
• Tissue and pathway signal attribution: Gene set enrichment and attribution analyses used predefined sets from Gene Ontology and the Molecular Signatures Database, integrating with adult and fetal organ expression references (Supplementary Table 2). Longitudinal data from cohort H (93 women sampled four times) and two additional cohorts with longitudinal sampling (total n=351 women) were used to validate temporal trends across pregnancy in gene sets (e.g., placenta, fetal heart development, collagen/extracellular matrix reflecting uterine/cervical remodeling). Randomized label permutation compared against observed GA labels to confirm association with pregnancy progression.
• Pre-eclampsia (PE) prediction: A case-control study included 72 PE cases and 452 controls from two cohorts (controls included normotensive, chronic hypertension, gestational hypertension, and spontaneous preterm births). Blood draws occurred in the second trimester (16–27 weeks), on average 14.5 weeks before delivery; an additional analysis referenced an early pregnancy dataset (4.5–6 weeks post-conception). Cohort correction was applied before modeling. Feature selection used Spearman correlation in cross-validation, retaining features with adjusted p≤0.05, repeatedly identifying genes including PAPPA2, CLDN7, FABP1, SNORD114A, PLEKHH1, MAGEA10, among others. A logistic regression classifier with leave-one-out cross-validation estimated PE risk probabilities and supported learning curve analysis. Pathway analysis characterized up- and downregulated biological processes.
• Statistical analyses and outcomes: Performance metrics included mean absolute error (MAE) for GA, sensitivity, specificity, positive predictive value (PPV), and area under the receiver operating characteristic curve (AUC) for PE classification. Additional analyses evaluated contributions of clinical variables, delivery timing differences between test-positive and test-negative individuals, and overlap/independence of signatures relative to chronic/gestational hypertension.

• Gestational age prediction: The cfRNA-based Lasso model achieved a test-set mean absolute error of 1.4 days relative to first-trimester ultrasound dating, comparable to second-trimester ultrasound and superior to third-trimester ultrasound.
• Independence from clinical factors: ANOVA showed the model was driven almost entirely by cfRNA transcripts; BMI, maternal age, and race each accounted for less than 1% of variance. Incorporating these clinical variables did not improve GA accuracy.
• Maternal–fetal tissue signals: Temporal cfRNA profiles captured dynamic changes across pregnancy in gene sets from placenta, developing fetal heart, and collagen/extracellular matrix (uterine/cervical remodeling). Of 793 predefined gene sets tested (384 adult, 409 fetal), 129 (56 fetal) correlated significantly with GA; 99 (40 fetal) increased and 30 (15 fetal) decreased with advancing GA in cohort H and were confirmed in at least two other longitudinal cohorts.
• Pre-eclampsia prediction (early dataset): In first-trimester sampling (4.5–6 weeks post-conception), cfRNA signatures yielded diagnostic sensitivity and specificity of 82.3% (s.d., 3%), outperforming conventional methods (56.0%, s.d., 9%).
• Pre-eclampsia prediction (second trimester analysis): Logistic regression with leave-one-out cross-validation achieved sensitivity of 73%, PPV of 32.3% at 13.7% prevalence, and AUC of 0.82 (95% CI, 0.76). Test-positive individuals delivered significantly earlier than test-negative individuals (P < 2 × 10^-4), correctly identifying 73% of those destined for medically indicated preterm birth over three months prior to symptom onset or delivery.
• Pathway insights: Upregulated processes included placental blood vessel development, arterial remodeling, and maternal physiological development; downregulated processes were largely immune-related, aligning with accepted mechanisms of PE pathophysiology.
• Specificity to PE vs chronic hypertension: No genes were differentially expressed between chronic/gestational hypertension and normotensive controls, and overlaps with PE genes suggest the PE signal is specific to a placental disorder rather than general hypertension, supporting clinical differentiation between superimposed PE and exacerbation of chronic hypertension.

The findings demonstrate that cfRNA profiling from a single maternal blood sample can accurately track pregnancy progression and predict pre-eclampsia risk months before clinical presentation. The minimal influence of clinical variables (e.g., race, BMI, age) on model performance underscores the biological specificity and potential for equitable, bias-resistant risk assessment. Temporal cfRNA signatures provide a non-invasive window into maternal–placental–fetal development, revealing coordinated changes across placental, fetal organ, and maternal tissue pathways. For PE, the molecular classifier detects pathophysiological changes well in advance of symptoms and delivery, highlighting opportunities for early surveillance and intervention planning. Pathway analyses corroborate known mechanisms of PE involving vascular remodeling and immune dysregulation. The specificity of cfRNA signals to PE versus chronic hypertension suggests utility in challenging clinical scenarios such as distinguishing superimposed PE. Overall, cfRNA-based assessments can enable personalized, mechanistically grounded prenatal care and may guide targeted therapeutics for molecular subtypes of PE and related complications.

This study provides a comprehensive, non-invasive cfRNA framework that (1) models gestational age with near-daily precision independent of clinical covariates and (2) predicts pre-eclampsia risk months before clinical onset, outperforming conventional approaches. The approach generalizes across diverse populations and captures coordinated maternal–placental–fetal biology over gestation. Future work should focus on validating predictors in additional prospective clinical settings, elucidating causal drivers of the identified pathways, refining molecular subtyping of pre-eclampsia, and testing precision interventions timed to the earliest detectable pathophysiological changes.

Although multi-cohort and diverse, the study relies on predefined gene sets and case–control analyses for PE; broader prospective validation in routine clinical workflows is needed. Directionality of certain gene set trends, while confirmed across available longitudinal cohorts, warrants further replication. The work highlights pathways associated with PE but does not establish causal mechanisms; further studies are required to identify upstream drivers and evaluate intervention efficacy. Performance metrics such as PPV depend on prevalence and may vary in different populations.