(100) Artificial intelligence (machine learning) and molecular classification of sibling twin (TW) placentas in preeclampsia (PE).

TW pregnancies are at increased risk for PE. Glycolytic metabolic shift is a key transcriptomic signature in PE placenta. Here, we explored transcriptomic features of the TW placenta using artificial intelligence (AI) tools to understand its contribution to the pathobiology of PE.

Study Design: RNAseq was performed on villous tissue of TW pregnancies grouped by chorionicity and PE diagnosis: 1) dichorionic (dc) PE (n=5, GA 33±2 w); 2) dc non-PE (n=7, GA 32±1 w); 3) monochorionic (mc) PE (n=2, GA 33±3 w); 4) mc non-PE (n=2, GA 32±1 w). Placentas from mc cases were sampled twice. A 3-cohort early-onset PE dataset from 127 singleton PE and non-PE placentas (GSE114691, GSE148241, GSE203507) was mined using machine learning (DaMiRseq) and used to train support vector machine (SVM) classifiers (caret R package). A gene-signature scoring (singscore) algorithm quantified molecular phenotypes along a continuum.

Results:

In the learning phase, SVM had accuracies of 94.6% and 89.2% for PE and glycolytic metabolic shift, respectively (Fig. A), segregating singleton data in 3 clusters: normal (baseline), disease group 1 (PE, no metabolic shift), and disease group 2 (PE with metabolic shift). In twin pairs, 56% (9/16; dc n=7, mc n=2) had inter-TW placental transcriptional discordance. Individually, 27% (5/18) of clinically non-PE TW placentas were reclassified to a diseased molecular class (group 1 or 2), while 35% (5/14; dc n=4, mc n=1) of clinically PE TW placentas were categorized as baseline (Fig. B). The TW PE placentas had higher (more PE-like) singscores (p< 0.001, ANOVA). Non-PE TW placentas also had elevated scores compared to non-PE singleton placentas (p< 0.01, ANOVA; Fig. C). Among dc pregnancies, only 42% (5/12) of cases had both placentas displaying the same molecular signature simultaneously.

Conclusion: Our novel AI approach demonstrates that placental transcriptional metabolic stress and discrepancy between clinical and molecular classification is common in TW. In dc TW one placenta carrying the molecular PE signature may be sufficient to trigger the hypertensive phenotype.