Cell cycle networks link gene expression dysregulation, mutation, and brain maldevelopment in autistic toddlers

Abstract

Genetic mechanisms underlying abnormal early neural development in toddlers with Autism Spectrum Disorder (ASD) remain uncertain due to the impossibility of direct brain gene expression measurement during critical periods of early development. Recent findings from a multi-tissue study demonstrated high expression of many of the same gene networks between blood and brain tissues, in particular with cell cycle functions. We explored relationships between blood gene expression and total brain volume (TBV) in 142 ASD and control male toddlers. In control toddlers, TBV variation significantly correlated with cell cycle and protein folding gene networks, potentially impacting neuron number and synapse development. In ASD toddlers, their correlations with brain size were lost as a result of considerable changes in network organization, while cell adhesion gene networks significantly correlated with TBV variation. Cell cycle networks detected in blood are highly preserved in the human brain and are upregulated during prenatal states of development. Overall, alterations were more pronounced in bigger brains. We identified 23 candidate genes for brain maldevelopment linked to 32 genes frequently mutated in ASD. The integrated network includes genes that are dysregulated in leukocyte and/or postmortem brain tissue of ASD subjects and belong to signaling pathways regulating cell cycle G1/S and G2/M phase transition. Finally, analyses of the CHD8 subnetwork and altered transcript levels from an independent study of CHD8 suppression further confirmed the central role of genes regulating neurogenesis and cell adhesion processes in ASD brain maldevelopment.

Keywords: Autism Spectrum Disorder; brain development; co‐expression; gene networks.

Figure 1. Schematic of the approach used in the study

1. Blood gene expression was analyzed in relationship with neuroanatomic measures using a co‐expression network‐based approach (WGCNA). The distribution of neuroanatomic measure was normal and not significantly different between ASD and control toddlers. The analysis of co‐expression was combined with all available samples.

2. Data from the combined network‐based analysis was further investigated in each ASD and control group separately using a linear model.

3. Network features, calculated from the WGCNA co‐expression analysis in relationship to brain size, were used to dissect alterations of network patterns in ASD brains.

4. Network features were also used to characterize smaller and bigger brains in each study group.

Figure 2. Analysis of gene networks associated with variation in brain size in ASD and control toddlers

1. A

   Distributions of brain size as indexed by total brain volume (TBV) in ASD and control toddlers used in the co‐expression analysis (WGCNA). T, value from t‐test; F, value from Levene's test.

2. B

   Module eigengenes (MEs) from the combined WGCNA are linearly correlated with TBV measures in all brains, ASD and control groups. P‐value is in parenthesis and adjusted P‐value (q‐value) is < 0.05 for all seven modules. Significant associations after 10,000 permutation tests are provided in Appendix Figs S4 and S5.

3. C

   Metacore enrichment scores of the seven (7) modules initially related to brain size variation across all subjects. Each module is called by its assigned color and represents the top process network obtained by the enrichment analysis in Metacore GeneGO (see also Dataset EV1).

4. D–F

   (i) Linear modeling of module eigengenes (MEs) by TBV measures in control (blue) and ASD (red) toddlers. See also Fig 2B for cor and P‐values. (ii) Linear modeling of GS by GC to display changes in network organization relevant to brain size. (iii) The top 30 genes with highest values for GS and GC were compared between ASD and control. Purple indicates the number of genes that moved away from the top 30 rank position between the two groups (Different genes), and grey indicates the number of genes that did not (Common genes). Significance codes: ***P‐value < 0.001; **P‐value < 0.01; cor, correlation coefficient; ns, not significant.

Figure 3. BrainSpan preservation analysis and cell cycle developmental trajectory

1. Preservation analysis between BrainSpan dorsolateral prefrontal cortex gene expression data and the ASD + control blood data. Zsummary statistic (e.g., Zsummary > 10 means highly preserved, Zsummary in between 2 and 10 is weak to moderate preservation, Zsummary < 2 is little to no preservation and median rank (modules with lowest rank are highly preserved)). Median Rank and Zsummary values indicate high module preservation between the two datasets.

2. Boxplot showing module eigengene (ME) values the BrainSpan cell cycle module for fetal versus postnatal time points (15 fetal versus 16 postnatal time points). The box refers to the interquartile range (IQR), which we refer to as Q1 (25^th percentile) and Q3 (75^th percentile). The upper whisker represents Q3 + 1.5*IQR, while the lower whisker represents Q1 − 1.5*IQR. The line in the middle of the box represents the median.

3. Scatterplot indicating the BrainSpan cell cycle module trajectory across development (vertical line indicates birth; time points to the left of the line are fetal time points, while time points to the right of the line are postnatal time points; best‐fit curve indicates a 4^th order polynomial fit).

Figure 4. Topological analysis of cell cycle hub‐genes in a protein–protein interaction (PPI) network

Mapping of the top 30 hub‐genes relevant to TBV measures in smaller and bigger ASD and control brains.

1. Hub‐genes that are normally active in control toddlers.

2. Hub‐genes that are active in both control and ASD toddlers.

3. Hub‐genes that are abnormally active in ASD toddlers.

Data information: Node size (circle) is proportional to the number of actual biologically driven PPIs. Blue and red nodes are hub‐genes that displayed PPIs and were in the top 30 lists in control and ASD, respectively. Grey nodes are genes in the cell cycle PPIs network that were not ranked within the top 30 genes. Black nodes are genes that fall out from the top 30‐gene list in ASD compared to control.

Figure 5. High‐confidence (Hc) network

A Hc network was generated following the approach represented in Fig EV1. Big nodes represent Hc genes (grey or colored circles). Small grey or colored circles represent direct downstream targets of the Hc genes. Colors code for different categories of genes mapped into the Hc network. Red indicates genes that are differentially expressed (DE) in blood of the same subjects described in this study. Cyan indicates genes that are upstream and regulate any of the 23 brain‐relevant genes identified in this study. Green indicates genes that are both DE and regulators of the 23 brain‐relevant genes. Big cyan diamond shapes are brain‐relevant genes that mapped into the Hc network. Bold circle outlines represent genes that are DE in postmortem brain tissue of ASD donors.

Figure EV1. Schematic of the approach used to generate the High‐confidence (Hc) network

Thirty‐two (32) Hc genes (blue circles) were mapped onto Metacore GeneGo to construct a Hc network. Each Hc gene was used as node to identify direct downstream targets (diamond shapes). Red lightning bolts represent the possible presence of mutations affecting the Hc genes. Similarly, twenty‐three (23) brain‐relevant genes (purple squares) were used in Metacore to look for direct upstream regulatory genes (yellow, cyan, green diamond shapes). Downstream and upstream targets were integrated into a final Hc network.

Figure 6. CHD8 subnetwork analysis in relationship to brain size and downstream CHD8‐knockdown effects in vitro

1. CHD8 subnetwork analysis included all CHD8‐targets from the Hc network. Network legend is the same as in Fig 5: grey (downstream CHD8 target), cyan (upstream regulatory gene of brain‐size relevant genes altered in ASD), red (differentially expressed in blood), green (cyan and red), diamond shape (brain‐size relevant gene altered in ASD), and black circle (differentially expressed in ASD cortex from Voineagu et al, 2011).

2. Linear correlation analysis of gene expression levels with TBV measures in ASD and control toddlers. See permutation analysis in Appendix Table S6. **P < 0.001.

3. Pathway analysis in Metacore. FDR < 0.05.

Figure 7. Schematic representations of mechanisms that may underlie brain maldevelopment in ASD

In control brains (top panel), both smaller and bigger brains develop and function normally. At the cellular level, there are no significant alterations of cell division phases and the number of cells produced is within normal variation. At the molecular level, there are no genetic alterations and changes in gene expression are within normal variation with no significant alterations of network structure and function. In ASD brains (bottom panel), brain development is abnormal and smaller and bigger brains represent two different anatomical and functional outcomes. In smaller ASD brains, the cellular characteristics are currently less clear compared to bigger brains. We hypothesize that G1/S phase transitions may be longer and/or checkpoints may stall/delay the timing of cell divisions leading to a reduced number of cells. Alternatively, there are an increased number of apoptotic cells. These cellular phenotypes may be related to genetic mutations (red lightning bolt) of Hc genes that, for instance, regulate chromatin modification as in case of CHD8 (blue). The mutation leads to altered regulation (red arrows) of downstream transcription factors or regulatory elements (yellow, cyan, green diamond shapes) that in turn regulate the expression of brain‐relevant genes (purple square). Mutated CHD8 can also alter directly the expression of brain‐relevant genes, as in case of CCNE2. Gene expression and network functions are altered, but closer to normal brains. In bigger brains, cellular and molecular phenotypes are more pronounced compared to smaller ASD brains. Cellular evidence suggests that the increased number of neurons may be due to the shortening of G1/S phases. At the molecular level, this may be related to mutations and changes in gene expression that lead to a reorganization of networks controlling neuroprogenitor cell divisions. In addition to the downstream effects of Hc mutations (i.e. CHD8, red arrows), pronounced gene expression changes cause a substantial reorganization of the network with the activation of new regulatory genes (i.e., TOP2A, TYMS, triangle shapes) and new interactions (green arrows), thus leading to altered or different network functions.