A study of gene expression in the living human brain

Abstract

A goal of psychiatric research is to determine the molecular basis of human brain health and illness. One way to achieve this goal is through studies of gene expression in human brain tissue. Due to the unavailability of brain tissue from living people, most such studies are performed using tissue from postmortem brain donors. An assumption underlying this practice is that gene expression in the postmortem human brain is an accurate representation of gene expression in the living human brain. This assumption - which, until now, had not been adequately tested - was tested by comparing human prefrontal cortex gene expression between 275 living samples and 243 postmortem samples. Expression levels differed significantly for nearly 80% of genes, and a systematic examination of alternative explanations for this observation determined that these differences are not explained by cell type composition, RNA quality, postmortem interval, age, medication, morbidity, symptom severity, tissue pathology, sample handling, batch effects, or computational methods utilized. Using gene expression data from two independent cohorts, the differences identified between living and postmortem samples were replicated and shown to be present in all brain cell types. Analyses integrating the data generated for this study with data from earlier studies that used tissue from postmortem brain donors showed that postmortem brain gene expression signatures of psychiatric and neurological illnesses, as well as of normal traits such as aging, may not always be accurate representations of these gene expression signatures in the living brain. By using tissue safely obtained from large cohorts of living people, future studies of the human brain have the potential to (1) determine the biomedical research questions that can be addressed using postmortem tissue as a proxy for living tissue and (2) expand the scope of medical research to include questions about the molecular basis of human brain health and illness that can only be addressed in living people (e.g., "What happens in the brain at the molecular level as a person experiences an emotion?").

Fig. 1. The Living Brain Project (LBP).

A Study overview. The pink and blue colors represent LIV samples and PM samples, respectively, throughout all subsequent figures. B Sampling the living human brain. The schematic illustrates that the standard clinical procedure and the modified study procedure amount to the same effective PFC volume loss. In (1) the state of the PFC at baseline is depicted. In (2) the final state of the PFC is depicted (i.e., the inner cylinder of tissue removed). The arrows and associated labels indicate how the final state of preparation is achieved in the standard clinical procedure (i.e., cauterization only) and the LBP procedure (i.e., biopsy followed by cauterization). Shown in (3) is a photograph of a PFC biopsy obtained for the LBP that is representative of the samples studied in this report. The top of the image is the anterior surface of the PFC. C Living Brain Project cohort. Numbers refer to sample size (i.e., individuals or samples) except for age. The individual and sample numbers are identical for the PM cohort because only one sample was obtained per individual. Sample sizes inside and outside of the square brackets indicate counts prior to and after the application of quality control filters, respectively. [PFC prefrontal cortex, QC quality control, HBTRC (PM1) Harvard Brain and Tissue Resource Center, NYBB (PM2) New York Brain Bank at Columbia University, UMBEB (PM3) University of Miami Brain Endowment Bank, MDD major depressive disorder, OCD obsessive-compulsive disorder].

Fig. 2. Identifying and interrogating the primary LIV-PM DE signature.

A LIV-PM DE results in the full LBP cohort. The scatterplot has one point for each of the 21,635 genes expressed in the full LBP cohort (275 LIV samples and 243 PM samples). The x-axis of the scatterplot is the average normalized expression level for each gene. The y-axis of the scatterplot is the LIV-PM DE regression model beta (by convention, the “logFC”) for each gene. Blue points represent DEGs upregulated in PM samples relative to LIV samples and pink points represent DEGs upregulated in LIV samples relative to PM samples. Grey points represent genes that were not differentially expressed between LIV samples and PM samples. The pie chart shows the fraction of scatterplot points that are blue, pink, and grey. The numbers in the pie chart indicate the number of LIV DEGs, PM DEGs, and Not DEGs. B LIV-PM DE signature comparisons. Each of the four scatterplots depicts the relationship between LIV-PM DE signatures generated from independent subsets of LBP samples. The x-axis and y-axis are the logFC values of the two LIV-PM DE signatures being compared in the plot. The Spearman’s correlation coefficient (ρ) between the two LIV-PM DE signatures is shown. The LIV-PM DE signatures compared in each plot are indicated by the plot title and fully described in the main text. For the “Low PMI vs. High PMI” panel, only the PM1 subset is depicted. C Gene set enrichment in LIV-PM DEGs. Each circle or triangle is a KEGG gene set tested for enrichment in LIV DEGs (pink circles and triangles) and PM DEGs (blue circles and triangles). The x-axis is the parent category of the gene set in the KEGG database. The y-axis is the Fisher’s exact test odds ratio for enrichment of a KEGG gene set in LIV DEGs and PM DEGs. Only parent categories with >=1 significant enrichment test results in either LIV DEGs or PM DEGs are included in the figure. For a given parent category, the KEGG gene set with the smallest p-value in LIV DEGs and PM DEGs is plotted. Triangles indicate that the enrichment test result is significant after p-value adjustment and circles indicate that the enrichment test result is not significant. The size of the triangles and circles corresponds to the -log₁₀(p-value) of the Fisher’s exact test, meaning larger sizes are more statistically significant. The full KEGG gene set enrichment results for LIV DEGs and PM DEGs are provided in Supplementary Table 2 with a column indicating the gene sets depicted in this figure. D Enrichment of postmortem actively regulated genes in LIV-PM DEGs. Each of the 3852 actively regulated genes expressed in the full LBP cohort was assigned to one of five sequential “enhancer bins” based on the number of high-confidence active enhancers linked to the gene (i.e., 1-5, 6-10, 11-15, 16-20, or 21-83). The x-axis is the Fisher’s exact test odds ratio for the overlap of enhancer bin genes with LIV DEGs (pink bars) or PM DEGs (blue bars). The y-axis indicates the enhancer bin.

Fig. 3. Dissecting the LIV-PM DE signature using single-nucleus RNA sequencing (snRNA-seq).

A The top and bottom bar plots show effects of LIV-PM status on gene expression for 10,770 genes that were LIV-PM DEGs in the cell type-uncorrected bulk LIV-PM DE signature and in at least one of the nine snRNA-seq LIV-PM DE signatures (“bulk/snRNA-seq LIV-PM DEGs”). In both the top and bottom bar plots, each bar represents one gene (i.e., one of the 10,770 bulk/snRNA-seq LIV-PM DEGs) and genes are plotted sequentially along the x-axis in the same order (determined by ordering the y-axis values in the top plot). Top plot – For a given gene plotted on the x-axis, the y-axis shows the net effect of LIV-PM status on the gene expression level based on the logFC values for the gene in the snRNA-seq LIV-PM DE signatures where the gene was found to be a LIV-PM DEG. Each bar is segmented by colors, which represent the snRNA-seq LIV-PM DE signatures where the gene was a LIV-PM DEG; the length of each colored segment represents the absolute value of the logFC for the gene in the corresponding snRNA-seq LIV-PM DE signature. A white horizontal line is plotted that intersects the y-axis at zero. For a given gene, the segments that appear below the white horizontal line represent instances in which the gene was a LIV DEG in a snRNA-seq LIV-PM DE signature and the segments that appear above the white horizontal line represent instances in which the gene was PM DEG in a snRNA-seq LIV-PM DE signature. Bottom plot – For a given gene plotted on the x-axis, the y-axis shows the effect of LIV-PM status on the gene expression level based on the logFC value for the gene in the cell type-uncorrected bulk RNA-seq LIV-PM DE signature. A white horizontal line is plotted that intersects the y-axis at zero. Bars that appear below the white horizontal line represent LIV DEGs and bars that appear above the white horizontal line represent PM DEGs in the cell type-uncorrected bulk RNA-seq LIV-PM DE signature. B Each of the 10,770 genes plotted in A was assigned to one of nine bins (x-axis) based on the number of snRNA-seq LIV-PM DE signatures for which the gene was a LIV-PM DEG (i.e., the number of segments in the bar for the gene in the plots in A). For each bin there is a point, and the y-axis value for the point is the median of the logFC magnitudes in the cell type-uncorrected bulk RNA-seq LIV-PM DE signature for the genes in the bin. Error bars extending from each point show the 1st and 3rd quartile boundaries for the logFC magnitudes in the cell type-uncorrected bulk RNA-seq LIV-PM DE signature for the genes in the bin.

Fig. 4. Gene-gene relationships in the full LBP cohort.

A Gene-gene correlation difference matrix means. The y-axis is the mean of each correlation difference matrix. Each of the 10,000 green points represents the mean of a “null correlation difference matrix” generated by comparing two random sample subsets. The overlaying boxplot summarizes the distribution of green points. The dotted red line represents the mean of the “LIV-PM correlation difference matrix” generated by comparing LIV samples to PM samples. B Conserved structures in LIV and PM networks. In both panels, the x-axis depicts the three conserved network structures identified across the six LBP networks that were enriched for LIV DEGs. The six modules of each conserved network structure are represented by bars. Each module is a network-specific representation of the conserved network structure, and the order of the bars for each conserved network structure is PM2 non-PD, PM2 PD, PM1 non-PD, PM1 PD, LIV non-PD, and LIV PD. The y-axis in the top panel is the odds ratio from the Fisher’s exact test of the enrichment of genes in the module for LIV DEGs. Purple bars indicate the enrichment test was significant after multiple testing correction. The y-axis in the bottom panel is the odds ratio from the Fisher’s exact test of the enrichment of genes in the module for the KEGG gene set most strongly associated with the module, and the color of the bars indicates the name of the KEGG gene set (all of the enrichment tests in the bottom panel were significant after multiple testing correction).

Fig. 5. Effect of LIV-PM status on DE signatures and co-expression networks.

A Overlap between LIV-PM DE signature and ALZ, SCZ, PD, and Age DE signatures. Each scatterplot depicts the results of DE analysis for either ALZ (top row, columns 1 and 2), SCZ (middle row, columns 1 and 2), PD (bottom row, columns 1 and 2), or Age (column 3, highlighted in yellow). The x-axis of all nine scatterplots is the DE regression model beta (by convention, the “logFC”) for each gene. Positive logFC values indicate higher expression levels in cases (or with higher age) and negative logFC values indicate higher expression in controls (or with lower age). The y-axis is the -log₁₀(p-value) for the DE test and the dotted red horizontal lines indicate the threshold for defining DEGs after multiple testing correction. ALZ, SCZ, PD, and PM Age DE signatures are derived from postmortem PFC samples. The LIV Age signature in the bottom row is from living PFC samples. The colored bar beneath every point in each scatterplot indicates whether the gene represented by the point is a LIV DEG (pink bar), PM DEG (blue bar), or not differentially expressed between LIV samples and PM samples (grey bars). B Stability of LIV and PM Age DE Signatures. The six violin plots depict distributions of correlations between Age DE signatures. The x-axis indicates the source of the samples used to generate the Age DE signatures compared. The colors indicate whether both sources are LIV samples (pink), both sources are PM samples (blue), or one source is LIV samples and the other is PM samples (purple). The Spearman’s correlation coefficient (ρ) between Age DE signatures is shown on the y-axis. C Enrichment of PD structures for PD GWAS genes. The y-axis is the Fisher’s exact test odds ratio for enrichment of PD GWAS genes in network-specific representations of each PD-specific conserved network structure on the x-axis (i.e., PD structure). The three classes of PD structures are shown: LIV PD structures (structures only seen in the LIV PD network; pink), PM PD structures (structures only seen in the two PM PD networks; blue), and LIV-PM PD structures (structures seen in LIV and PM PD networks; purple). For each structure, the network-specific representation with the highest odds ratio is shown. Only LIV PD structures that were enriched for PD GWAS genes (at the unadjusted or adjusted p-value threshold) are shown for that category. Since no PM PD structures and no LIV-PM PD structures were enriched for PD GWAS genes at the unadjusted p-value threshold, structures are shown for which all of the network-specific representations had an odds ratio > 0. The size of the circles corresponds to the significance of the enrichment test as indicated in the figure legend.