Alterations in oligodendrocyte transcriptional networks reveal region-specific vulnerabilities to neurological disease

Abstract

Neurological disease is characterized the by dysfunction of specific neuroanatomical regions. To determine whether region-specific vulnerabilities have a transcriptional basis at cell-type-specific resolution, we analyzed gene expression in mouse oligodendrocytes across various brain regions. Oligodendrocyte transcriptomes cluster in an anatomical arrangement along the rostrocaudal axis. Moreover, regional oligodendrocyte populations preferentially regulate genes implicated in diseases that target their region of origin. Systems-level analyses identify five region-specific co-expression networks representing distinct molecular pathways in oligodendrocytes. The cortical network exhibits alterations in mouse models of intellectual disability and epilepsy, the cerebellar network in ataxia, and the spinal network in multiple sclerosis. Bioinformatic analyses reveal potential molecular regulators of these networks, which were confirmed to modulate network expression in vitro in human oligodendroglioma cells, including reversal of the disease-associated transcriptional effects of a pathogenic Spinocerebellar ataxia type 1 allele. These findings identify targetable region-specific vulnerabilities to neurological disease mediated by oligodendrocytes.

Keywords: Biological sciences; Cellular neuroscience; Neuroscience.

Graphical abstract

Figure 1

Analysis of OL-specific gene expression across various regions of the mouse brain (A) Bar plots showing scaled expression of regional OL populations for nine region-specific markers derived from the Human Protein Atlas. (B) Scatterplots of normalized counts on log2 scale for the six between-region comparisons and the four between-replicate comparisons. Colored dots represent differentially expressed genes (DEGs) between the two regions. Total number of DEGs is shown in each panel along with the Pearson correlation between normalized, log-transformed counts. Number of replicates for each comparison are shown in parentheses. Frontal cortex, FC (n = 7); corpus callosum, CC (n = 6); cerebellum, CBL (n = 6); spinal cord, SC (n = 5). (C) Unsupervised hierarchical clustering from pairwise Pearson correlations of regional OL transcriptomes. (D) Principal component analysis (PCA) of regional OL transcriptomes from the four anatomical regions studied. Anatomy from surgical isolation is shown in the background. Percentage of variance explained is shown in parentheses.

Figure 2

Differential expression, stability, and splicing of disease-associated genes reveals potential regional susceptibilities to disease mediated by OLs (A) Gene set enrichment analysis (GSEA) using normalized counts to assess disease-associated genes compiled from the online mendelian inheritance in man database. (B) As in panel A but using unbiased relative mRNA stability estimates to assess disease genes. (C) As in panel A but using enrichment of differentially spliced genes for disease-associated genes. Note that, in general, more spinal cord-specific RNA splicing was observed overall relative to the other tissues. Dot sizes correspond to false discovery rates (FDR). NES, normalized enrichment score.

Figure 3

Heatmap and top representative enrichment results for the key regional oligodendrocyte modules Left) Heatmap showing scaled expression of the region-specific OL co-expression networks, aligned from highest expression in cortex to highest expression in spinal cord. Red indicates high expression and blue indicates low expression. Right) Table showing enrichment for GO terms and disease-associated genes compiled from the online mendelian inheritance in man database for each module on the left. Values shown are false discovery rates (FDR) with fold enrichment in parentheses. ALS, amyotrophic lateral sclerosis; ID, intellectual disability; MS, multiple sclerosis.

Figure 4

Modules profiled by differential expression, mRNA stability, splicing, and putative regulation (A) Pie-donut charts of the region-specific OL co-expression networks. The pie shows the percentage of differentially expressed genes (up or down) and differentially stabilized genes (stabilized or destabilized). The donuts show the percentage of module genes that are spliced in their corresponding region, or predicted targets of their putative regulator, or both (marked by ∩). (B) Silhouettes of module genes ranked by intramodular connectivity (kWithin), where the genes belonging to each y axis category are colored in to show where they lie along the connectivity ranking. Each panel of silhouettes corresponds to the pie chart above it. The p value shown is from a one-tailed t-test comparing the connectivity of genes belonging to each subset to the connectivity of all other module genes. The median connectivity is highlighted for convenience.

Figure 5

Overlaps between OL-derived networks and disease-associated networks (A) Gene overlaps between region-specific OL co-expression networks and various ID-associated modules. On the right, we show module eigengene correlation to the cortex and enrichment for ID-associated genes. (B) Gene overlaps between regional OL modules and epilepsy-associated modules. On the right, we show enrichment for epilepsy-associated genes. (C) Gene overlaps between regional OL modules and cuprizone-induced remyelination-associated modules. On the right, we show enrichment for multiple sclerosis-associated genes. (D) Gene overlaps between regional OL modules and Pelizaeus-Merzbacher disease (PMD)-associated modules. On the right, we show enrichment for leukodystrophy-associated genes. (E) Gene overlaps between regional OL modules and ataxia with oculomotor apraxia type 2 (AOA2)-associated modules or spinocerebellar ataxia type 1 (SCA1)-associated modules. On the right, we show module eigengene correlation to the cerebellum and enrichment for ataxia-associated genes. (F) Gene overlaps between regional OL modules and experimentally induced autoimmune encephalomyelitis (EAE)-associated modules from two independent datasets. On the right, we show OL module eigengene correlation to the spinal cord and enrichment for multiple sclerosis (MS)-associated genes. Color corresponds to -log10 FDR from hypergeometric probability of overlap shown in color legend and the number of overlapping genes is shown in each box. Significance for the bar plots is shown by a dashed line.

Figure 6

Assessment of network perturbations in human oligodendroglioma cells (A) Schematic representation of cuprizone perturbation experiment. (B) Schematic representation of antisensense oligonucleotide (ASO) experiments (Plp1, miR-124, miR-506). (C) Schematic representation of drug-induced inhibition of PRC1 experiment. (D) Scatterplots showing the module-trait correlation (-log10 p value, Pearson correlation test) for each module identified by WGCNA from the experiments in A–C. (E) Table summary of significant overlap (hypergeometric test) with the key OL modules being targeted. (F) Heatmaps of significant human oligodendroglioma cell modules from E. For each module, the expression heatmap is shown above while a bar graph of the module eigengene is shown below. All experiments were performed in replicate and all replicates were handled identically: cuprizone treatment (n = 3 for each group), PLP1 knockdown (n = 3 to 4 for each group), miR-124 knockdown (n = 3 to 4 for each group), miR-506 knockdown (n = 3 to 4 for each group), and the PRC1 inhibition experiment (n = 3 for each group).

Figure 7

Identification of regional OL module regulators (A) Transcription factor target enrichment for M20 genes versus correlation between transcription factor expression and the module eigengene. Red suggests gene activation and blue indicates gene repression. Dashed lines mark statistical significance (p < 0.05, Pearson correlation test). (B) Heatmap and boxplot of M20 expression (top 50 most connected genes) in wildtype, heterozygous, and homozygous mutant Chd7 cerebellar granule neural progenitors (GNPs) from Feng et al., 2017. (C) Module overlaps between the cerebellar M20 module and two Chd7 deficiency-associated modules (hypergeometric test). (D) Transcription factor target enrichment for M20 genes and target enrichment for differentially expressed genes within the module. (E) Heatmap and boxplot for mouse embryonic stem cells (ESCs) with endogeous (end) Ring1b, loss of endogeous Ring1b, exogenous (exo) expression of wildtype Ring1b, or exogenous expression of catalytically inactive Ring1b I53S. Data from Tamburri et al., 2020. (F) Module overlap between M20, the Chd7 deficiency-associated module (Fe10), and the PRC1 loss-of-function-associated module (Ta3). (G) ChIP-seq coverage of H2AK119ub1 at the Chd7 locus showing peaks in samples with endogenous or exogenous wild-type Ring1b but not in the conditional knockout (cKO) or the Ring1b I53S inactive mutant data. Data from Tamburri et al., 2020. (H) Chd7 expression during various Ring1b perturbations in ESCs. n.s. not significant, ∗ FDR<0.05, ∗∗ FDR<0.01, DESeq2 Wald test. Data shown as mean ± range of two biological replicates. (I) miRNA target enrichment for M9 genes and target enrichment for differentially expressed genes within the module. (J) Heatmap and boxplot for M9 expression (top 100 connected genes) in mouse hippocampi with or without miR-124-3p knockdown from Malmevik et al., 2015. (K) Three-way module overlap between the OL-derived M9 module, miR-124-3p inhibition module (Ma18), and multiple sclerosis (MS) remyelination module (Vo1).