First, we analyzed genome-wide expression of 57 frozen samples of dorsolateral prefrontal cortex (DLPFC, BA 9/46; Table S1
) using a DASL-based platform on the Illumina Human-Ref8 v3 microarray 
. Thirty-three samples in total passed quality control measures (Text S1
): 16 of these samples were from young postmortem males (2–14 y; autism
7) and 17 were from adult males (15–56 y, autism
11). Though RNA Integrity Numbers (RIN) are not predictive of array quality 
, multiple RIN assessments were used to determine RNA quality (Table S1
). We found no significant differences between the RINs of the autism and control groups (t
Two-way analysis of variance (ANOVA) and post-hoc pair-wise ANOVA-based comparisons were performed to identify genes exhibiting expression differences in young autistic vs. young control DLPFC compared to adult autistic vs. adult control DLPFC. An overall ANOVA-based F-test p-value<0.05 [empirical false discovery rate (FDR), not threshold, FDR
0.27] was used to identify genes showing an effect of interaction between age and diagnosis for further analyses. From the two-way ANOVA results, we attempted to tease out expression differences between: 1) young autism, 2) young control, 3) adult autism and 4) adult control groups.
Of the genes exhibiting an effect of interaction from ANOVA-based F-tests with p-values<0.05, 102 genes resulted in ANOVA-based post-hoc contrasts between young autistic and control groups with p-value<0.05 (Figure S1
). Seven hundred thirty six genes resulted in an ANOVA-based post-hoc contrast between adult autistic and control cases (p-value<0.05). Finally, an overall ANOVA-based F-test p-value<0.05 (empirical FDR
0.13) was performed to assess the main effect of diagnosis across all autistic and control cases independent of age. Two thousand seventeen genes resulted in this contrast with p-value<0.05. Genes common between the three comparisons are shown in Figure S1
Genes exhibiting a diagnosis main effect and a post-hoc pairwise contrast (e.g., autism vs. control, adult vs. young; p-value<0.05) were further subjected to enrichment analyses in the MetaCore software suite (MetaCore from GeneGo Inc.) as well as the online enrichment program DAVID 
. By focusing on genes exhibiting a diagnosis main effect, we explored pathological mRNA expression in autism irrespective of age. qPCR validation verified expression differences on a subset of the differentially expressed genes (Figure S2
Genes regulating cell number, genetic integrity, and neural patterning are dysregulated in young autistic cases
Since early brain overgrowth and the clinical onset of autistic symptoms occur at young ages in autism, we focused on unique gene expression differences between autistic and control cases below the age of 14 years to identify genes that may be dysregulated early in autism pathogenesis. One hundred two genes were differentially expressed in young autistic cases compared with the young control group and exhibited a diagnosis x age group interaction effect (p-value<0.05, empirical FDR
0.27; Table S2
). Many of these genes were previously identified autism candidate loci based on the literature (Table S3
Pathway enrichment analyses of these 102 genes via the MetaCore software suite (defining an enriched pathway as having an enrichment p-value<0.05 and empirical FDR<0.1) suggested that DNA damage-response, cell cycle and apoptosis-related MetaCore pathways (‘M-Pathways’) were significantly altered (, Young autism versus young control map folders). Key players such as BRCA1 and CHK2 were downregulated. Most significantly dysregulated M-Pathways in this category included the DNA-damage-induced response and role of NFBD1 in DNA damage response. A Development-Neurogenesis Process Network, Development-Neurogenesis (p
1.09E-03, 6/192 network objects), was also significantly altered.
MetaCore Pathway Map Folders (left) and M-Pathways (right) in three ANOVA-based analyses.
We additionally annotated all 102 genes using DAVID and identified overlapping sets of 12, 19, 7 and 16 genes involved in DNA damage/cell cycle, apoptosis, immune signaling and neurogenesis and neural development (), respectively. Of the 7 immune response genes, 4 (FAS, BCL3, GREM1 and FOSL2) were also found to contribute to apoptosis. Neurogenesis and neural development pathways included the WNT pathway and were driven by dysregulation of the WNT3 gene. WNT pathway genes are known to regulate cell proliferation, cell fate and patterning during embryogenesis 
. Downstream components of the WNT pathway, such as Dvl1, are also known to regulate social behavior in mouse models 
. In addition, we found significant downregulation of genes involved in neural patterning and differentiation, such as FGF1, HOXD1, NDE1, NODAL, PCSK6 and GREM1 (, Table S2
Dysregulated gene expression in developmental M-Pathways and Process Networks in young autistic prefrontal cortex.
Expression anomalies of genes involved in DNA damage, cell cycle and apoptosis may contribute to abnormal brain growth by increasing production or reducing elimination of neurons during development. Dysregulated neural patterning and differentiation genes may lead to abnormal cellular organization and cytoarchitecture. For example, HOX and DLX family genes play important roles in vertebrate patterning and are important for neuronal subtype differentiation 
. Moreover, NODAL controls dorsal mesoderm induction, anterior patterning and initiation of left-right asymmetry during gastrulation 
. These findings suggest that in the young autistic brain, genetic regulation of cell number, genetic integrity and neural patterning is disturbed.
Genes regulating signaling, repair, and response pathways are dysregulated in adult autistic cases
As a complement to the analyses of the young autistic brain, we compared the genes dysregulated in young autistic cases to those dysregulated in adult autistic cases. We identified genes using the same ANOVA diagnosis x age interaction effect p-value criterion but with emphasis on genes differentially expressed in adult autistic brains relative to adult control brains (defined as cases ≥15 years of age). Seven hundred thirty six genes were differentially expressed based on this analysis (Figure S1
, Table S4
). These genes were also analyzed with MetaCore for functional enrichment. The 3 most significant MetaCore Map Folders (‘Map Folders’: a label MetaCore provides to sets of genes with an overarching function of pathway participation) included cell differentiation, mitogenic signaling and apoptosis genes (, Adult autism versus adult control map folders, ; p<0.05, FDR<0.1). Dysregulated pathways specific for the adult brain included multiple signaling and remodeling functions in neurons and glia (, Adult autism versus adult control map folders). M-Pathway categories that were dysregulated in adults but not young cases included development, signaling and oxidative stress pathways (, ).
Dysregulated gene expression in top three Map Folders in adult autistic prefrontal cortex and differentially affected M-Pathways of young and adult autistic cases.
The Cell Differentiation Map Folder included significantly dysregulated genes RELN, BTRC, BMP4, MAPK10 and NTRK3 (). This Map Folder also included suggested dysregulation of the ‘Activin A in Cell Differentiation and Proliferation’ pathway, which involved the genes LHB, NODAL, STAR, CDKN1A, PRKAR1A and ADCY6. Genes playing multiple functions in all three top map folders included MAPK12, CDKN1A, NTRK3, PRKAR1A, PIK3CA, CASP9, MAPK10, ADCY6 and MAGED1 (). Notably, Tissue Remodeling and Wound Repair-related genes also exhibited dysregulation in adult autistic cases (, Adult autism versus adult control map folders).
These analyses suggest that in the adult autistic brain, cell differentiation, mitogenic, apoptotic and remodeling and repair functions could be components of recovery responses, in accordance with previous reports 
. They could, however, also be signatures of ongoing reparatory neurogenesis processes 
. For example, the Activin A signaling pathway is expressed by neurons following injury and is essential for adult neurogenesis 
. BTRC inhibits the beta-catenin (CTNNB1) pathway 
, which in adult animals is upregulated after insults such as seizures 
and promotes adult neurogenesis 
. Furthermore, BMP4 expressed in adult subventricular zones serves to inhibit neurogenesis 
. Thus, it appears that aberrant signaling and repair processes may distinguish adult autistic cases from young cases.
Genes important in development were dysregulated across both young and adult autistic cases
Finally, we examined genes showing a main effect of diagnosis to identify differentially expressed between autistic and control cases independent of age (i.e., we did not confine attention to genes exhibiting diagnosis x age interaction effects or age-specific effects). More than 2000 such genes were detected based on a simple contrast between autism and control brains (p<0.05, empirical FDR
0.13; Figure S1
, Table S5
). Enrichment analyses of these genes using MetaCore suggested that seventeen Map Folders were altered in all autistic brains (p<0.05, FDR
0.1; , Diagnosis main effect map folders). The three most significant Map Folders included DNA-damage response, apoptosis and immune system response functions (, Figure S3
). In addition, the top 25 M-Pathways (, Diagnosis main effect map folders, Table S6
) included cell cycle [14-3-3 (YWHAZ), CDC25A, CDCD25C, ATRX], proliferation [CTNNB1 (beta-catenin), FSHB, PRKACB, PRKCZ], apoptosis (BAD, CASP8, CASP10, MDM2), cytoskeleton and extracellular matrix remodeling (ErbB4, MMP2, NID1, TIMP1, COL4A3) and growth and development [RELN, ROBO1, ADORA2A, p21 (CDKN1A), 14-3-3, HGF, FGFRL1, TSC1] functions. Specifically, the p53 signaling pathway and the PTEN pathway were among these dysregulated M-Pathways (, Diagnosis main effect map folders).
Dysregulated gene expression in top two MetaCore Map Folders in autism independent of age.
A number of genes in these M-Pathways have known functions in neuronal development: PTEN signaling regulates proliferation 
and is associated with macrocephaly in autism 
. TSC1 is mutated in tuberous sclerosis, acts downstream of PTEN in the mTOR pathway 
and affects cortical lamination, neuronal migration and axon pathfinding 
. CTNNB1, a key member of the WNT pathway, regulates cerebral cortical size. CTNNB1 transgenic mice have enlarged and folded cortices 
. Furthermore, RELN is critical for human neuronal migration 
and has been previously linked to autism 
. Genes differentiating autistic cases from controls independent of age have important developmental, immune and cytoskeletal remodeling functions.
Finally, given the novel cytoskeletal age-independent M-Pathways identified in this analysis, we revisited the 102 significant genes identified in the young autism vs. control analysis. We found 17 cytoskeletal and matrix remodeling genes (). These analyses support a role for cytoskeletal dysregulation in young autistic brains as well. Cytoskeletal elements have been linked to defects in neuronal migration in other neurodevelopmental disorders such as lissencephaly 
Comparison with a published functional genomic study shows overlap in repair and immune response genes
Due to commonalities between previously identified candidate loci and the genes we found to exhibit expression differences between young autism vs. control, we also investigated similarities between the genes we identified and those identified in a recent functional genomic study. We compared genes showing a main effect of diagnosis in the present study with differentially expressed genes implicated recently by Voineagu et al. 
Twenty-five probes detected as p<0.05 in comparing autism and control cases and 21,564 probes detected as p>0.05 in both studies were identified as overlapping (p<0.0001). Among the probes detected at p<0.05 were 6 genes important for Tissue Remodeling and Wound Repair (p
1.622E-4, FDR<0.05) and 7 genes important for Immune System Response (p
4.772E-4, FDR<0.05). Ultimately, we found some consistency between genes detected in our analyses with those of a previous functional genomic study, particularly within domains of repair and immune response.
Genetic variation may underlie cell cycle and cytoskeleton gene dysregulation
To further investigate the findings of dysregulated genetic functions in young autistic cases particularly, we determined whether CNVs in autistic cases were distinct from those in controls. We genotyped prefrontal cortex samples from 55 of the 57 total autistic and control cases in our study. After quality control, CNV enrichment was analyzed in 30 DLPFC cases from male and female autistic and control cases (Table S1
) using PennCNV with GC adjustment and CNVision programs 
. Nearly all (>99%) brain-derived CNVs in autistic and control cases were deletions (Table S7
, Table S8
). There were no differences between autistic and control cases in numbers of CNVs, numbers of genes per CNV or average CNV size (Figure S4
Restricting analysis to male cases, genes identified in the autistic group were enriched in cell cycle (mitosis and S phase, including BUBR1, Cyclin B, BRCA1, RAD51, CRM1, RFC2 and LIS1) and cytoskeleton (spindle and cytoplasmic microtubules, including CLIP170, Tubulin alpha, Dynein light and heavy chains, DNAL1 and ROD) GeneGO Process Networks. No significant enrichment was identified in the gene content of controls (). Similar network enrichment was noted in male autistic cases after filtering out common, non-pathogenic CNVs (Materials and Methods
), while no significant network enrichment was found in controls (). Notably, BRCA1 expression was found to be dysregulated in the young autistic brain () and predicted as deleted in the CNV analysis (Table S7
). Subsequent enrichment analysis of CNVs present in male and female autistic and control cases yielded similar results (Figure S4
). These results are consistent with the above described gene expression results showing abnormal expression of cell cycle and developmental pathways in the autistic cases.
MetaCore process networks and network maps in autism based on gene deletions located in CNVs from DLPFC, and genetic association results.
To extend these genotypic findings and specifically test whether common variants of cell cycle genes are associated with autism, we determined whether common variants in processes involved in cell cycle and other hypothesized regulatory processes were associated with autism using two previously analyzed SNP datasets: one from the Autism Genetic Resource Exchange (AGRE) and National Institutes of Mental Health genotyped at the Broad Institute and Johns Hopkins Medical Institute (Broad/JHMI) and another from AGRE genotyped at the Children's Hospital of Philadelphia (CHOP) 
. Set-based association analysis was applied to the Broad/JHMI dataset (Materials and Methods
). Four sets of genes postulated to be involved in brain overgrowth (cell cycle, WNT pathway, growth factors and apoptosis) were used as experimental sets for analysis, while other postulated pathological processes such as synaptogenesis and inflammation were used as controls.
We found nominal associations of only cell cycle-related genes with ASD (, p
0.037). Cell cycle genes were then subdivided into twenty-one subgroups. Genes regulating cell cycle senescence (p
0.011), the G2-M phase transition (p
0.028) and the G0-G1 phase transition (p
0.051) in particular were associated with autism in the Broad/JHMI dataset (). Because these analyses were not statistically corrected for the number of pathways considered, the findings were replicated for the three significant subcategories (p<0.05) of cell cycle genes using the independent CHOP dataset (). Two of the subcategories (senescence and G0-G1 phase transition) were replicated (p<0.0328 and p<0.0093 respectively), while the third (G2-M phase transition) was not. These results suggest that abnormal expression of cell cycle and developmental pathways may originate from genetic variation.