Divergent landscapes of A-to-I editing in postmortem and living human brain

Abstract

Adenosine-to-inosine (A-to-I) editing is a prevalent post-transcriptional RNA modification within the brain. Yet, most research has relied on postmortem samples, assuming it is an accurate representation of RNA biology in the living brain. We challenge this assumption by comparing A-to-I editing between postmortem and living prefrontal cortical tissues. Major differences were found, with over 70,000 A-to-I sites showing higher editing levels in postmortem tissues. Increased A-to-I editing in postmortem tissues is linked to higher ADAR1 and ADARB1 expression, is more pronounced in non-neuronal cells, and indicative of postmortem activation of inflammation and hypoxia. Higher A-to-I editing in living tissues marks sites that are evolutionarily preserved, synaptic, developmentally timed, and disrupted in neurological conditions. Common genetic variants were also found to differentially affect A-to-I editing levels in living versus postmortem tissues. Collectively, these discoveries illuminate the nuanced functions and intricate regulatory mechanisms of RNA editing within the human brain.

Figure 1.. Overview of study design and multi-omic utilization.

This study leverages a comprehensive set of multi-omic data from the Living Brain Project, including: (i) Bulk RNA-sequencing data from 164 living and 233 postmortem dorsolateral prefrontal cortex (DLPFC) samples; (ii) Single-nuclei RNA-sequencing data from an independent subset of 31 living and 21 postmortem DLPFC samples, ensuring no participant overlap with the bulk sequencing cohort; and (iii) Paired whole-genome sequencing (WGS) data from 155 living and 195 postmortem DLPFC samples. Detailed cohort demographics are detailed in Supplemental Data 1. Comprehensive analyses were conducted to quantify global Alu editing levels and individual A-to-I editing sites, with subsequent investigations encompassing bulk tissue comparisons, cell type-specific editing patterns, pathway-driven predictors of editing, and the genetic influences on A-to-I editing dynamics. We propose a mechanistic model to frame the interpretation of our overall findings.

Figure 2.. Global Alu editing across living and postmortem DLPFC.

(A) Alu editing index (AEI; y-axis) computed on bulk RNA-seq from living and postmortem (PM) DLPFC. Two-sided linear regression was used to test for significance. (B) ADAR, ADARB1 and ADARB2 normalized expression profiles on bulk including RNA-seq between living and PM. All boxplots show the medians (horizontal lines), upper and lower quartiles (inner box edges), and 1.5× the interquartile range (whiskers). Reported BH adjusted p-values were derived from a moderated t-test comparing transcriptome-wide gene expression between living and postmortem tissue. (C) Linear mixed model explaining AEI variance by eleven known biological and technical factors. (D) UMAP dimension reduction analysis of snRNA-seq classified nine unique cell populations. Values in brackets indicate the number of cells per sub-population: excitatory (EXC) and inhibitory (INT) neurons, astrocytes (AST), microglia (MG), oligodendrocytes (OLI), OLI precursor cells (OPCs), endothelial cells (ENDO). (E) The mean frequencies for each cell population quantified between living and PM. Two-sided linear regression was used to test for significance. (F) Hierarchical clustering of scaled ADAR, ADARB1 and ADARB2 expression across all cell populations. (G) Cell type-specific ADAR, ADARB1 and ADARB2 expression for living and PM. (H) Alu editing index computed for each cell population for each donor and compared across living and PM samples (bottom). PM-induced effect sizes calculated by Cohen’s d for each cell population (top). (G-H) Two-sided linear regression was used to test for significance (*denotes p< 0.05). All boxplots show the medians (horizontal lines), upper and lower quartiles (inner box edges), and 1.5× the interquartile range (whiskers). Two-sided linear regression was used to test for significance. (I) Pearson’s correlation coefficient between the mean Alu editing index and mean normalized ADARB1 expression for each cell population according to living and PM samples. Standard error bars capture group-wise variance within living and postmortem tissues, respectively. RNA-seq analysis encompassed 164 and 233 biologically independent samples from living and postmortem sources, respectively. Single-nucleus RNA-seq was conducted on 31 living and 21 PM biologically independent samples. All box plots in this figure show the medians (horizontal lines), upper and lower quartiles (inner box edges), and 1.5 × the interquartile range (whiskers).

Figure 3.. Dynamically regulated A-to-I sites between living and postmortem DLPFC.

(A) Principal component analysis of editing levels for 54,825 high-confidence sites detected across all samples in the current study. (B) Differential editing analysis compare delta editing levels (%; x-axis) and strength of significance (−log₁₀ adjusted P, y-axis) for each site between living and postmortem (PM) DLPFC. Sites are colored by novelty (i.e. detection in REDIportal) and shaped uniquely by genic region. Reported BH adjusted p-values were derived from a moderated t-test comparing transcriptome-wide A-to-I editing levels between living and postmortem tissue. (C) Frequency distribution of mean editing levels in living DLPFC (x-axis) based on PM biased sites (y-axis). (D) The fraction of genic regions for all living biased and PM biased sites. (E) Frequency distributions of Pearson’s correlation coefficients (x-axis) between the expression for ADAR, ADARB1, ADARB2 relative to editing levels for 54,825 sites. An additional analysis modeled ADAR+ADARB1-ADARB2 to capture ADAR and ADARB1 effects. The total number of sites with significant correlations are listed in the top right corner of each histogram. (F) Dynamic recoding sites: 27 living-biased recoding sites and 31 postmortem-biased recoding sites ordered by their effect size differences (y-axis, lower) and plotted alongside with the mean editing levels (y-axis middle). The strength of evolutionary conservation (phastCons) was measured for each site and the probability of being loss of function intolerant (pLI) was measured for each gene (top). (G) Frequency distributions demonstrating that living biased recoding sites are often more strongly evolutionarily conserved and map to genes with higher pLI relative to PM biased recoding sites. Mann Whitney U-test was used to test for significance. Living Brain Project data encompassed 164 and 233 biologically independent samples from living and postmortem sources, respectively.

Figure 4.. Annotating dynamically regulated sites between living and postmortem DLPFC.

(A) CAMERA enrichment scores (y-axes) for three candidate sets of A-to-I editing sites along a ranked list of differentially edited sites (t-statistics; x-axis) between living and postmortem (PM) DLPFC, from highly living-biased (right; pink) to highly PM-biased (left; blue) (x-axes). Enrichment plots for non-neuronal sites (top, PM biased), sites disrupted in autism spectrum disorder (ASD) cortex (middle, living biased) and those disrupted in schizophrenia (SCZ) ACC (bottom, living biased). (B) Summary of all multiple test corrected p-values (−log₁₀) for all sets of RNA editing sites across cell types, neurodevelopmental disorders, and brain development. (A-B) CAMERA gene set enrichment p-values, quantifying the statistical significance of overrepresented A-to-I sites within the ranked living versus postmortem data. (C) Pearson’s correlation and scatterplot of delta editing rates for cell-specific recoding sites (y-axis) versus delta editing rates for living/PM differences (x-axis). Y-axis description: Fluorescence activated nuclei sorted (FANS) neurons and non-neuronal cell populations were collected from 10 postmortem donors across five cortical regions (see Supplemental Note 1). (D) SynGO synaptic enrichment (−log10 q-value) for genes harboring living-biased editing sites (top) and genes harboring postmortem-biased sites (bottom). RNA-seq analysis encompassed 164 and 233 biologically independent samples from living and postmortem sources, respectively.

Figure 5.. Biological processes that explain differences in Alu editing.

(A) Single-sample scores were generated for 10,493 Gene Ontology Biological Processes. These pathway scores were regressed onto the AEI while covarying for known biological and technical factors. Reported BH adjusted p-values and t-statistics were derived from a moderated t-test comparing pathway scores between living and postmortem tissue. The t-statistics (x-axis) for each biological process relative to strength of association with the AEI (y-axis; −log₁₀ adjusted p-value). Pathways with an absolute t-statistic > 7 and FDR adjusted p-value < 0.05 were deemed significant (blue, negative association; red, positive association). (B) A density distribution of t-statistics illustrates most pathways are positive predictors of the AEI (top) and REVIGO semantic similarity representation of the top positive 1688 pathways (bottom, ). Multiple broad groupings emerge that map to intracellular signaling, apoptosis, hypoxia, cellular metabolism and innate immune/inflammatory responses. Colors indicate the Adjusted P-value of the enriched GO terms. The size of each bubble shows the GO terms with more significant P-values. (C) Single sample scores that represent top pathways from each cluster also predict differences between living and postmortem (PM) tissues. (D) Pearson’s correlation coefficient illustrates regressions of single-sample pathway score onto the AEI for the top six candidate pathways. All Living Brain Project data encompassed 164 and 233 biologically independent samples from living and postmortem sources, respectively. (E) Validating the interaction between interferon-γ and the AEI. Two-dimensional (2D) hiPSC-derived neural progenitor cells (NPCs; day 18) and mature neurons (day 30) treated with interferon-gamma (IFN-γ) (PMID: 32875100). Data was generated by bulk RNA-seq from n=3 biological replicates. (F) Validating the interaction between hypoxia and the AEI. Three-dimensional (3D) model of human cortical spheroids (hCS) exposed to differing degrees of hypoxia (PMID: 31061540). Data was generated by bulk RNA-seq from n=8 biological replicates. (E-F) Two-sided linear regression was used to test for significance. All boxplots in this figure show the medians (horizontal lines), upper and lower quartiles (inner box edges), and 1.5× the interquartile range (whiskers).

Figure 6.. Context-dependent cis-edQTLs between living and postmortem DLPFC.

(A) Distribution of the cis-edQTL associations evaluating the distance between eSites and common variants. The gray box indicates ±200 kb relative to the editing site. Inset Venn diagram depicts the overlap of eSites between the primary and interaction analyses. (B) eSite discovery (y-axis) according to genic regions (x-axis) for primary (top) and interaction (bottom) cis-edQTL analyses. (C) Editing level variance within living and postmortem DLPFC parsed by sites with primary and interaction cis-edQTLs. (D) The fraction of eSites from the primary (top) and interaction (bottom) analyses that are either neuronal or non-neuronal cell type specific. (E) Two examples of primary cis-edQTLs, in which editing levels (y-axes) are associated with common genotypes (x-axes). (F) Two examples of interaction cis-edQTLs, in which different common genotypes (x-axis) are associated with differing editing levels between living and postmortem DLPFC (y-axes). (C-E) Two-sided linear regression was used to test for significance. Analyses encompassed 164 and 233 biologically independent samples from living and postmortem sources, respectively. All boxplots in this figure show the medians (horizontal lines), upper and lower quartiles (inner box edges), and 1.5× the interquartile range (whiskers).