Placental genomics mediates genetic associations with complex health traits and disease

Abstract

As the master regulator in utero, the placenta is core to the Developmental Origins of Health and Disease (DOHaD) hypothesis but is historically understudied. To identify placental gene-trait associations (GTAs) across the life course, we perform distal mediator-enriched transcriptome-wide association studies (TWAS) for 40 traits, integrating placental multi-omics from the Extremely Low Gestational Age Newborn Study. At [Formula: see text], we detect 248 GTAs, mostly for neonatal and metabolic traits, across 176 genes, enriched for cell growth and immunological pathways. In aggregate, genetic effects mediated by placental expression significantly explain 4 early-life traits but no later-in-life traits. 89 GTAs show significant mediation through distal genetic variants, identifying hypotheses for distal regulation of GTAs. Investigation of one hypothesis in human placenta-derived choriocarcinoma cells reveal that knockdown of mediator gene EPS15 upregulates predicted targets SPATA13 and FAM214A, both associated with waist-hip ratio in TWAS, and multiple genes involved in metabolic pathways. These results suggest profound health impacts of placental genomic regulation in developmental programming across the life course.

Fig. 1. Overview of the DOHaD hypothesis.

The placenta facilitates important functions in utero, including nutrient transfer, metabolism, gas exchange, neuroendocrine signaling, growth hormone production, and immunologic control. Accordingly, it is a master regulator of the intrauterine environment and is core to the Developmental Origins of Health and Disease (DOHaD) hypothesis. Placental genomic regulation is influenced by both genetic and environmental factors and affects placental developmental programming. In turn, this programming has been shown to have profound impacts on a variety of disorders and traits, both early- and later-in-life.

Fig. 2. Overview of analytic pipeline.

a Predictive models of placental expression are trained from germline genetics, enriched for mediating biomarkers using MOSTWAS and externally validated in RICHS eQTL data. b Predictive models are integrated with GWAS for 40 traits to detect placental gene-trait associations (GTAs). c GTAs are followed up with gene ontology analyses, probabilistic fine-mapping, and phenome-wide scans of genes with multiple GTAs. Relationships between identified distal mediators and TWAS genes are investigated further in RICHS and ENCODE. Expression-mediated genetic heritability of and correlations between traits are estimated. d In vitro validation of prioritized regulatory protein-TWAS gene pairs are conducted using placenta-derived choriocarcinoma cells by gene silencing, qRT-PCR, and RNA-seq to measure transcriptomic impacts of the regulatory protein.

Fig. 3. Placental MOSTWAS prediction and association test results.

a Overview of TWAS association testing pipeline with number of gene-trait associations (GTAs) across unique genes over various levels of TWAS tests. b Kernel density plots of in- (through cross-validation in ELGAN, red) and out-sample (external validation in RICHS, blue) McNemar’s adjusted $R^2$ between predicted and observed expression. Dotted and solid lines represent the mean and median of the respective distribution, respectively. c Bar graph of numbers of TWAS GTAs at overall TWAS $P<2.5\times {10}^{-6}$ (for two-sided weighted burden Z-test) and two-sided permutation test FDR-adjusted $P<0.05$ (x-axis) across traits (y-axis). The total number of GTAs per trait are labeled, colored by the category of each trait. The bar is broken down by numbers of GTAs with (orange) and without (green) significant distal expression-mediated associations, as indicated by FDR-adjusted $P<0.05$ for the distal-SNPs added-last test. d Enrichment plot of over-representation in 176 TWAS genes of PANTHER pathways (y-axis) with -log₁₀ FDR-adjusted P-value (x-axis) of one-sided Fisher’s exact test. The size of the point gives the relative enrichment ratio for the given pathway.

Fig. 4. Trait genetic heritability and correlations mediated by placental expression.

a Box-plot of expression-mediated trait heritability ($h_{GE}^2$), estimated with LD-score regression, (y-axis) by category (x-axis), with labels if ${\widehat{h}}_{GE}^2$ is significantly greater than 0 using two-sided jack-knife test of significance. Boxplots are b Box-plot of expression-mediated trait heritability ($h_{GE}^2$) standardized by SNP heritability ($h^2$) (y-axis) by category (x-axis), with labels if ${\widehat{h}}_{GE}^2$ is significantly greater than 0 using two-sided jack-knife test of significance. c Forest plot of significant placenta expression-mediated genetic correlations (shown as points) and 95% FDR-adjusted two-sided Wald-type confidence intervals between traits from different categories. Boxplots in a, b provide, in order of bottom to top, the lower extreme, lower quartile, median, upper quartile, and upper extreme. Sample sizes used to derive these statistics across a–c are provided in Supplementary Data 1.

Fig. 5. Computational follow-up analyses of TWAS-prioritized genes.

a Box-plot of -log10 FDR-adjusted P-value (two-sided weighted burden Z-tests) of multi-trait scans of GTAs in UKBB, grouped by eight groups of traits (grouped generally around ICD code blocks or organ systems) across nine genes with multiple TWAS GTAs across different trait categories. The red dotted line represents FDR-adjusted P = 0.05. Sample sizes vary for each trait. Boxplots provide, in order of bottom to top, the lower extreme, lower quartile, median, upper quartile, and upper extreme. b Forest plot of GTA association estimates (shown as points) and 95% FDR-adjusted two-sided Wald-type confidence intervals for six neonatal traits in ELGAN for nine genes with multiple TWAS GTAs across categories. The red line shows a null effect size of 0, and associations are colored blue for associations at FDR-adjusted P < 0.05. Sample size of 729 individuals from ELGAN were used to derived these estimates. c Follow-up GBAT and Mendelian randomization (MR) analysis results using RICHS data. On the left, effect size (shown as points) and 95% adjusted confidence intervals from GBAT (x-axis; two-sided t-test, df = 148) between GReX of RP-encoding genes and TWAS gene associations (pairs given on y-axis). On the right, MR effect size and 95% adjusted confidence interval (x-axis; two-sided Wald-type test) of RP-gene on TWAS gene (pairs on y-axis). The red line shows a null effect size of 0, and associations are colored blue for associations at FDR-adjusted p < 0.05. Sample size of 149 individuals from RICHS were used to derive these estimates.

Fig. 6. In vitro experiments in EPS15-knockdown human placenta-derived choriocarcinoma epithelial cells.

a Bar graph of the mean gene expression fold-changes with error bars of 1 standard deviation from the qRT-PCR from JEG-3 RNA. Nominal P-values of pairwise t-tests are shown, with an asterisk if Benjamini–Hochberg FDR-adjusted $P<0.10$ (n = 2 cells examined over two independent experiments). Note differences in y-axis scales. b Volcano plot of log₂ fold-change (x-axis) of differential expression across EP15 knockdown cells and scramble oligonucleotide against –log₁₀ FDR-adjusted P-value (y-axis; two-sided Wald-type tests from negative binominal regression). Upregulated genes are in red and downregulated genes in blue. Top up- and downregulated genes by P-value are labeled, as well as EPS15, SPATA13, and FAM214A. c Enrichment plot of over-representation of down- (blue) and upregulated (red) genes of PANTHER and KEGG pathways (y-axis) with –log₁₀ FDR-adjusted P-value (x-axis; one-sided Fisher’s exact test). The size of the point gives the enrichment ratio.