ASD is a neurological disorder with a strong genetic component that has been linked to a number of gene defects. These genes have a broad range of activities, ranging from membrane receptors and scaffold proteins to metabolic regulators and transcription factors [25, 35]. Despite this diversity, ASD patients manifest with similar behavioral and neuronal phenotypes, albeit with different severities. This commonality of neurological phenotype suggests that the genetic defects may act through a limited set of pathways. In this report, we employed shRNA knockdown of eight ASD relevant genes in neuronal culture to explore the downstream effects and identify common pathways or transcriptional signatures.
Following microarray analysis of all samples, we performed cluster analysis on the intensity values. As expected, samples clustered by treatment group, demonstrating an overall consistency and quality of the knockdown experiment and subsequent gene expression quantitation. It also illustrates the distinctiveness of the downstream expression effects of knockdown of individual genes. Knockdown of Pten and Mecp2 had the most dramatic effects on gene expression. Given Pten’s broad role in numerous cellular processes and Mecp2’s role as a transcription factor, these results were not unexpected. For example, mutations in Pten have been linked not only to ASD but also cancer and diabetes [53, 54]. Fragile X mental retardation protein (FMRP), the protein product of Fmr1, has been shown to interact with a larger number of target proteins in relation to dendritic control of translation. A list of FMRP target proteins showed significant enrichment in the transcriptional profiles of shRNA for not only Fmr1, but also Mecp2, Pten, Shank3, Nlgn1 and Nlgn3. We further compared the genes affected in one or more knockdown experiments to a list of ASD interactome genes . This comparison indicated that knockdown of the eight ASD genes resulted in changes to a significant number of ASD interactome genes and the genes affected by the luciferase shRNA condition had little overlap with the ASD genes (Table 3). This control comparison is important, as other groups have reported nonspecific adverse effects of other shRNA and siRNA constructs [55, 56]. The luciferase shRNA versus untransduced comparison yielded almost 1,000 differentially expressed transcripts, with an impact on BDNF measured at the protein level. Thus, by identifying the changes in the luciferase shRNA versus untransduced experiments and subtracting those, the subsequent pathway analyses could focus on pathways that were specifically targeted by knockdown of the ASD-relevant genes and not identify artifacts of the transduction.
We next analyzed the gene lists from the shRNA experiments by two pathway analysis approaches to obtain different perspectives on the data. The most prominent pathways revealed through analysis with NextBio were a number of pathways related to neurologic signaling and function (Table 4). Secondarily, NextBio indicated that several pathways involved general cellular metabolism and growth were also affected. One prominent pathway, the Peroxisome proliferator-activated receptor gamma coactivator 1-A (PGC1A) pathway, is based on the MSigDB’s version of BioCarta’s pathway and contains Mef2A and several calcium-dependent kinases, which show gene expression changes in all shRNA experiments. One aspect of the NextBio analysis is that directionality of change (that is, upregulated or downregulated) is reported. The majority of the pathways are downregulated with ASD shRNA knockdown, suggesting that the genes we chose for this work are needed for the expression of these pathways and thereby their activity. Pathway analysis with GenSensor also identified a number of pathways related to neuronal signaling and function (Table 5). As with the NextBio analysis, several growth and metabolism pathways were also affected.
During an examination of the individual pathways identified by the two pathway analysis methods, we noted a recurring involvement of the mitogen-activated protein kinase kinase (MEK)/ERK signaling pathway. These effects would occur either directly through a kinase signaling cascade (downstream of BDNF/TRK) or via cAMP (as in the case of the dopamine and serotonin G-protein coupled receptors). To further investigate this potential commonality, we employed CRE analysis to identify potential underlying mechanisms (CRE hypotheses) in the shRNA datasets. Unlike pathway analysis, which identifies pathways with altered gene expression, CRE analysis predicts potential mechanisms behind gene changes based on the concordance of the number of genes that change expression, and the directionality of that change . The results of CRE analysis are interlinked hypotheses of potential driving mechanisms or experimental treatments that exhibit similar gene changes. It is interesting that three of the eight most conserved hypotheses have a biological function suggestive of growth and/or immune function, suggesting similar driving mechanisms (Figure 3). Likewise, there are highly conserved hypotheses involved with neurogenesis, synaptic activity and differentiation, as expected, although not mutually exclusive. Choosing the Mef2d experiment as a representative of the six most conserved shRNA treatments, the top ranking clusters can be connected as a molecular interaction map with (Figure 4), cyclic AMP and ERK serving as dual hubs of the network directly connecting seven and eight related hypotheses respectively. Three of the experiments, Mecp2, Mef2d, and Nlgn1, shared 11 of the 12 hypotheses in the network. As more experiments are included, the shared number of hypotheses decreases, for example, the light blue grouping of six experiments including Shank3 is based on four hypotheses.
In addition to this work, other work directly or indirectly supports a role for ERK signaling in the development of ASD. For example, maternal use of one of several different classes of drugs relevant to ERK modulation has been reported to increase the risk of having children born with ASD . Cocaine use during pregnancy has been reported to increase the rate of autism by 11%. Cocaine use has also been shown to alter dopamine-induced phosphorylation of ERK via cAMP [58, 59]. Recently, Hoffmann et al. showed that chronic cocaine use in rats can lead to attenuated ERK signaling . Chronic maternal cocaine use might thereby attenuate ERK signaling in the fetus. Similarly, mothers taking valproic acid, an inhibitor of gamma-aminobutyric acid (GABA) function, have been demonstrated to have an increased risk of have children with autism . As with cocaine, valproic acid activates ERK signaling . Zou et al. demonstrated that RAS/RAF/ERK1/2 signaling was upregulated in the brains of the BTBR mouse model of autism . Recently, the upregulation of this pathway (and of ERK5) has been shown to occur in the brains of autistic subjects . Although misregulation of ERK does appear to be a common feature of ASD, the observed directionality of that misregulation has been contradictory. In the case of Rett syndrome, ERK signaling through the BDNF pathway in particular is reduced. BDNF levels are reduced in Mecp2-null mice, and exogenous BDNF has been shown to rescue deficits due to Mecp2 deficiency [7, 9]. In human Rett syndrome patients, a Val/Met polymorphism in BDNF has been associated with disease severity . In the present study, Mecp2 shRNA produced a significant reduction in Bdnf at both the mRNA and protein level, both of which were inversely affected by Fmr1 knockout. Given the diverse functions of BDNF in neurons, it would be interesting to determine in follow-up studies whether inverse functional outcomes may be observed with these treatments.
Based on the experimental data presented here and previously existing data, we have put together a pathway model to show that the transcriptional regulation exerted by a diverse set of ASD-associated genes converges on ERK signaling. A central role for ERK signaling would explain many of the features associated with ASD. Early work on the ERK proteins described these as microtubule-associated protein 2 kinases, and were shown to phosphorylate MAP2 kinases, proteins known to be involved in neuronal architecture [65–67]. Later work demonstrated that ERK plays a critical role in microtubule formation and thereby to axon/dendrite formation [68, 69]. A review article by Hoogenraad and Akhmanova has summarized the criticality of microtubules in synaptic plasticity . Mutations that lead to altered ERK activity would then be expected to have alterations in axon extension and/or retraction and thereby, synaptic plasticity. Mazzucchelli et al. found that ERK1-knockout mice exhibit enhanced synaptic plasticity, most likely through the compensatory activation of ERK2 . Voineagu et al. recently reported that the expression differences between the temporal and frontal lobes are significantly attenuated in individuals with autism . They further suggested that this lack of differentiation is the mechanism behind the lack of long-range axonal connections and the decreased myelin thickness in autistic prefrontal lobes as reported by Zikopoulos and Barbas [21, 22]. In some instances altered ERK activity could interfere with neuroglia wrapping of neuritis to form the myelin sheath. Newbern et al. recently reported that ablation of ERK1/2 in Schwann cell precursors resulted in hypomyelination of axons .