A study of RNA splicing and protein expression in the living human brain

Abstract

Due to the unavailability of living human brain tissue for molecular research, postmortem brain samples are currently the standard tissue source for molecular studies of the human brain. The Living Brain Project (LBP) was designed to test the assumption that the postmortem brain is an accurate molecular representation of in the living brain on multiple levels of molecular biology. Findings from previous LBP reports suggest that this assumption does not hold with respect to RNA transcript expression levels. Here, molecular differences between living and postmortem human prefrontal cortex tissues obtained for the LBP are corroborated through analyses of RNA splicing and protein expression data. Significant differences were observed with respect to (1) the expression of most primary RNA transcripts, mature RNA transcripts, and proteins, (2) the splicing of most primary RNA transcripts into mature RNA transcripts, and (3) the patterns of co-expression between RNA transcripts and proteins. Taken together, this report corroborates the presence of widespread molecular differences between living and postmortem human brain tissues. These observations should be considered when designing and interpreting studies of human brain biology.

Fig 1. (A) Report overview.

Schematic illustrating the study objective and how it was achieved. (B) Study cohort. Numbers refer to sample size (i.e., individuals or samples) except for age. Sample sizes are shown for most of the analyses of LBP data presented in the report. Sample sizes inside and outside of the square brackets indicate counts for bulk RNA-seq and bulk LC-MS analyses, respectively.

Fig 2. (A) Identification of LIV-PM DE signatures.

A matrix of nine plots is presented that altogether summarize the differences in RNA and protein expression levels between LIV samples and PM samples. Each row of the matrix is a different analysis, and each column is a different data type (primary RNA, mature RNA, protein). The off-white background is to differentiate columns from one another. Top Row: scatter plots showing the results of dimensionality reduction performed on primary RNA, mature RNA, and protein expression data using the UMAP algorithm. Each point is a sample, and colors differentiate LIV samples (pink) from PM samples (blue). The horizontal axis is the first UMAP dimension (“UMAP 1”), the vertical axis is the second UMAP dimension (“UMAP 2”). Middle Row: boxplots showing distributions of UMAP 1 values (horizontal axis) stratified on the vertical axis both by LIV-PM status and Parkinson’s disease (PD) status of the samples (LIV samples – pink, PM samples – blue; samples from individuals with PD – A [“Affected”], samples from individuals without PD – U [“Unaffected”]). For each boxplot: the colored line inside the box is the media; the left and right edges of the box are the first and third quartiles (the 25th and 75th percentiles); the right whisker extends from the right edge of the box to the largest y-axis value no further than 1.5 times the interquartile range away from the right edge of the box; the left whisker extends from the left edge of the box to the smallest y-axis value no further than 1.5 times the interquartile range away from the left edge of the box. Bottom Row: scatter plots showing the results of differential expression analysis comparing LIV samples to PM samples for each primary RNA transcript, mature RNA transcript, and protein feature quantified for the bulk RNA-seq and bulk LC-MS analyses in the current report. Each point is a feauture (i.e., RNA transcript or protein). The horizontal axis shows the average normalized RNA transcript or protein expression level. The vertical axis shows the logFC values, and the range of values is −4.95 to 5.29 for primary RNA transcripts, −7.71 to 5.40 for mature RNA transcripts, and −1.71 to 1.27 for protein. Positive logFC values are indicative of higher levels in PM samples compared to LIV samples, and negative logFC values are indicative of higher levels in LIV samples compared to PM samples. Colors differentiate features with levels that were significantly higher in PM compared to LIV (blue), significantly higher in LIV compared to PM (pink), or not significantly different between LIV and PM (gray). (B) RNA and protein LIV-PM DE signature concordance. Scatter plot showing the concordance of RNA transcript and protein LIV-PM DE signatures. Each point is a feature, and only features present in all 3 LIV-PM DE signatures (i.e., primary RNA, mature RNA, protein) are plotted. The x-axis shows the logFC values from the primary RNA transcript LIV-PM DE analysis. The y-axis shows the logFC values from the mature RNA transcript LIV-PM DE analysis. The color of points indicates the logFC values from the protein LIV-PM DE signature. The colors range from pink (greater expression in LIV samples compared to PM samples) to white (no difference in expression between PM samples and LIV samples) to blue (expression greater in PM samples relative to LIV samples). (C) LIV-PM differential splicing rate analysis. An alluvial plot showing changes in the RNA transcript LIV-PM DE signature across three variables (represented by three “pillars” on the horizontal axis) that were tested for differences between LIV and PM samples on the same set of features: levels of primary RNA transcript expression (left pillar), the rate of conversion of primary RNA to mature RNA (middle pillar), and levels of mature RNA expression (right pillar). There are three categories, called “stratum”, within each of the three pillars: RNA features tested that were greater in LIV samples compared to PM samples (pink), greater in PM samples compared to LIV samples (blue), or not significantly different between LIV samples and PM samples (gray). Each “alluvial fan” (i.e., a set of thick wavy lines connecting pillars pillars) shows change of a set of features across the three variables and is comprised of two “flows”, which are the segments connecting adjacent pillars. Flows are colored by the DE status of the stratum from which it originated, and fans can therefore be comprised of flows of multiple colors. The pie charts above each pillar represent the cumulative proportion of the transcriptome that is different between LIV samples and PM samples in at least one of the three levels tested (totaling over 95% as indicated by the proportion of the pie that is colored purple in the last pie chart). (D) LIV-PM differential intron usage example. Differential intron usage between LIV samples and PM samples is shown for the RSRP1 intron cluster. Exons are represented by the black segments beneath the continuous black line (which represents the full gene). Introns in the cluster are represented by curves that connect two exon ends. The thickness of each curve represents the intron usage ratio. To calculate the intron usage ratio, the mean intron usage was calculated separately for LIV and PM samples and the larger mean intron usage was divided by the smaller mean intron usage. Curves colored pink are introns with greater usage in LIV compared to PM and curves colored blue are introns with greater usage in PM compared to LIV.

Fig 3. (A) Correlating mature RNA to protein of same gene.

(I) Positive Correlations Enriched. Left – density plots of the same-gene RNA-protein correlations in LIV samples (pink) and PM samples (blue). Right – Bar plot showing the results of tests of whether the same-gene RNA-protein correlations were significant and positive more than expected by chance in LIV samples and PM samples (vertical axis). The summary statistic of this test is the odds ratio, presented on the horizontal axis. Error bars are the 95% confidence interval of the odds ratio from Fisher’s exact test, and error bars that do not cross the dotted vertical line intersecting the horizontal axis at 1 indicate statistical significance. (II) Effect of LIV-PM status. Left – scatter plot showing relationship of same-gene RNA-protein correlations in PM samples (x-axis) to same-gene RNA-protein correlations in LIV samples (y-axis). To color and shade the points, the mature RNA logFC and the protein logFC were summed; if the sum was greater than 0 the point was colored blue, if the sum was less than 0 the point was colored pink, and the shade of the points was set to reflect the absolute value of the sum. Right – To determine if LIV-PM status was driving same-gene mature RNA-protein correlations, betas (y-axis) and associated p-values were obtained using the following linear model applied separately to LIV samples and PM samples (x-axis): same-gene correlation coefficient ~ Mature RNA LIV-PM logFC (resulting beta is “R” on x-axis) + Protein LIV-PM logFC (resulting beta is “P” on x-axis). (B) Correlating all mature RNAs to all proteins. For the two KEGG sets most enriched for differentially correlated RNA-protein pairs (transcription factors; spliceosome) the RNA-protein correlations are shown between the proteins in the set (heatmap rows) and all mature RNA transcripts (heatmap columns). Positive correlations are in red, negative correlations are in blue. “LIV Data” and “PM Data” describe the set of samples (i.e., LIV samples or PM samples) used to generate the correlations in the corresponding row of heatmaps. “LIV Order” and “PM Order” describe the data that was used to order the proteins and mature RNA transcripts in the corresponding column of heatmaps.