Hierarchical cortical transcriptome disorganization in autism

Abstract

Background: Autism spectrum disorders (ASD) are etiologically heterogeneous and complex. Functional genomics work has begun to identify a diverse array of dysregulated transcriptomic programs (e.g., synaptic, immune, cell cycle, DNA damage, WNT signaling, cortical patterning and differentiation) potentially involved in ASD brain abnormalities during childhood and adulthood. However, it remains unclear whether such diverse dysregulated pathways are independent of each other or instead reflect coordinated hierarchical systems-level pathology.

Methods: Two ASD cortical transcriptome datasets were re-analyzed using consensus weighted gene co-expression network analysis (WGCNA) to identify common co-expression modules across datasets. Linear mixed-effect models and Bayesian replication statistics were used to identify replicable differentially expressed modules. Eigengene network analysis was then utilized to identify between-group differences in how co-expression modules interact and cluster into hierarchical meta-modular organization. Protein-protein interaction analyses were also used to determine whether dysregulated co-expression modules show enhanced interactions.

Results: We find replicable evidence for 10 gene co-expression modules that are differentially expressed in ASD cortex. Rather than being independent non-interacting sources of pathology, these dysregulated co-expression modules work in synergy and physically interact at the protein level. These systems-level transcriptional signals are characterized by downregulation of synaptic processes coordinated with upregulation of immune/inflammation, response to other organism, catabolism, viral processes, translation, protein targeting and localization, cell proliferation, and vasculature development. Hierarchical organization of meta-modules (clusters of highly correlated modules) is also highly affected in ASD.

Conclusions: These findings highlight that dysregulation of the ASD cortical transcriptome is characterized by the dysregulation of multiple coordinated transcriptional programs producing synergistic systems-level effects that cannot be fully appreciated by studying the individual component biological processes in isolation.

Keywords: Autism; Gene co-expression networks; Immune; Synapse; Systems biology; Transcriptome; Translation.

Fig. 1

Upregulated gene co-expression modules in ASD. This figure shows gene co-expression modules that were on-average elevated in ME expression in ASD and in a replicable manner across datasets. Each module has a scatter-boxplot whereby each individual is represented by a dot and the central tendency (median) and dispersion (interquartile range) is shown with the boxplot. Next to each scatter-boxplot are the process-level enrichment terms passing FDR q < 0.05 (limited to the top 10 terms) from MetaCore GeneGO. The vertical black line on the enrichment bar plots represents the p value where FDR q < 0.05. For each module, the replication Bayes Factor statistic (repBF) is cited above the scatter-boxplot (repBF >10 indicates strong evidence for replication). In the bottom right corner of this figure is a Venn diagram summarizing the common overlap between ASD-upregulated genes across both datasets

Fig. 2

Downregulated gene co-expression modules in ASD. This figure shows gene co-expression modules that were on-average decreased in ME expression in ASD and in a replicable manner across datasets. Each module has a scatter-boxplot whereby each individual is represented by a dot and the central tendency (median) and dispersion (interquartile range) is shown with the boxplot. Next to each scatter-boxplot are the process-level enrichment terms passing FDR q < 0.05 from MetaCore GeneGO. The vertical black line on the enrichment bar plots represents the p value where FDR q < 0.05. For each module, the replication Bayes Factor statistic (repBF) is cited above the scatter-boxplot (repBF >10 indicates strong evidence for replication). In the bottom right corner of this figure is a Venn diagram summarizing the common overlap between ASD-downregulated genes across both datasets

Fig. 3

Cell type/cellular compartment enrichments for dysregulated modules. This figure shows enrichments in a variety of cell types and cellular components for the modules that are replicably dysregulated in ASD. The left panel shows enrichments for downregulated modules, while the right panel shows enrichments for the upregulated modules. The coloring of the bars denote which specific module shows the enrichment and the color legend is shown in the bottom right box for each panel. The x-axis plots the –log₁₀ p values while the y-axis indicates the specific cell type or cellular compartment. Next to each bar, we indicate the enrichment odds ratio (OR)

Fig. 4

Correlations between dysregulated modules. a, b Correlations between differentially expressed modules in the Voineagu Control (a) or ASD (b) datasets. c, d Correlations between these same modules in the Gupta Control (a) or ASD (b) datasets

Fig. 5

Protein-protein interactions between dysregulated modules. This figure plots the number of protein-protein interactions (on log₁₀ scale) between seed modules and downregulated (left) or upregulated (right) modules (y-axis) as a function of module size (number of genes in the module; x-axis). Seed modules that are downregulated are colored in blue, while upregulated seed modules are colored in red. Non-dysregulated seed modules are colored in green. For dysregulated seed modules, the number of connections reflects the number of protein connections with other dysregulated modules, not counting self-connections (e.g., connections between genes of the same co-expression module). This figure clearly shows that seed modules that are dysregulated (red or blue) possess a far greater number of connections with other dysregulated modules compared to non-dysregulated modules (green) of a similar size

Fig. 6

GO biological process enrichments for collections of downregulated or upregulated modules. This plot shows GO biological process enrichment terms for the combination of all downregulated (a) or upregulated (b) modules. The top 50 GO terms ranked by fold enrichment were input into REVIGO [30] in order to cluster GO terms by semantic similarity. These clusters are shown in different colors along with a descriptive label for each cluster. Plotted on the x-axis of each plot is the Bonferroni-corrected –log₁₀ p value for each term

Fig. 7

Eigengene network topology and connectivity differences within the Voineagu dataset. a, b Eigengene networks as weighted graphs in a spring embedded layout for the Voineagu Control (a) or ASD (b) groups. The spring-embedded layout places modules (nodes within the graphs) that are highly connected as much closer in space whereas modules that are less highly connected are repelled away from each other. The thickness of the connections (i.e., edges) between modules are scaled to connection strength whereby the thinnest line represents a correlation of r = −1 and the thickest line represents a correlation of r = 1. The color of each module node represents the ASD meta-module it belongs to. This was done to represent where the ASD meta-modules are located within the Control graph. The color-filled outlines around collections of modules represent the meta-module boundaries. Modules with a solid red or blue circle around it are modules that were identified in Figs. 1 and 2 as being replicably dysregulated in ASD across both datasets (blue = ASD-downregulated; red = ASD-upregulated). The dotted circles represent differentially expressed modules (FDR q < 0.05) present only within that specific dataset (see Additional file 4: Table S3). c Within (c) and outside (d) normative meta-module connectivity strength for each seed module depicted on the y-axis. The normative (Control-defined) meta-modules are denoted by the color of the rectangular outlines on the y-axis. Connectivity strength is depicted on the x-axis and for within meta-module connectivity is defined as the sum of connection strength between the seed module and all other modules within the seed module’s normative meta-module. Outside meta-module connectivity strength is defined as the sum of connection strength between the seed module and all other modules outside of the seed module’s normative meta-module. Turquoise bars indicated Controls, and salmon-colored bars indicate ASD. The stars next to specific modules indicate a significant between-group difference in connectivity strength. d Eigengene networks as robust ME partial correlation matrices. Red coloring within the matrices indicates increasing positive correlation strength, while blue coloring indicates increasing negative correlation strength; see color bar for key indicating how color corresponds to correlation strength. Matrices have rows and columns ordered by hierarchical clustering based on the Control group, and the individual module numbers as well as meta-module colors are shown. Normative (Control-defined) meta-module boundaries are also clearly delineated by the black outlines over cells in the correlation matrices. Any cells with green outlines are those specific between-module connectivity comparisons that differed between-groups

Fig. 8

Eigengene network topology and connectivity differences within the Gupta dataset. a, b Eigengene networks as weighted graphs in a spring embedded layout for the Gupta Control (a) or ASD (b) groups. The spring-embedded layout places modules (nodes within the graphs) that are highly connected as much closer in space whereas modules that are less highly connected are repelled away from each other. The thickness of the connections (i.e., edges) between modules are scaled to connection strength whereby the thinnest line represents a correlation of r = −1 and the thickest line represents a correlation of r = 1. The color of each module node represents the ASD meta-module it belongs to. This was done to represent where the ASD meta-modules are located within the Control graph. The color-filled outlines around collections of modules represent the meta-module boundaries. Modules with a solid red or blue circle around it are modules that were identified in Figs. 1 and 2 as being replicably dysregulated in ASD across both datasets (blue = ASD-downregulated; red = ASD-upregulated). The dotted circles represent differentially expressed modules (FDR q < 0.05) present only within that specific dataset (see Additional file 4: Table S3). c Within (c) and outside (D) normative meta-module connectivity strength for each seed module depicted on the y-axis. The normative (Control-defined) meta-modules are denoted by the color of the rectangular outlines on the y-axis. Connectivity strength is depicted on the x-axis and for within meta-module connectivity is defined as the sum of connection strength between the seed module and all other modules within the seed module’s normative meta-module. Outside meta-module connectivity strength is defined as the sum of connection strength between the seed module and all other modules outside of the seed module’s normative meta-module. Turquoise bars indicated Controls and salmon-colored bars indicate ASD. The stars next to specific modules indicate a significant between-group difference in connectivity strength. d Eigengene networks as robust ME partial correlation matrices. Red coloring within the matrices indicates increasing positive correlation strength, while blue coloring indicates increasing negative correlation strength; see color bar for key indicating how color corresponds to correlation strength. Matrices have rows and columns ordered by hierarchical clustering based on the Control group, and the individual module numbers as well as meta-module colors are shown. Normative (Control-defined) meta-module boundaries are also clearly delineated by the black outlines over cells in the correlation matrices. Any cells with green outlines are those specific between-module connectivity comparisons that differed between-groups