Convergence of Spectrums: Neuronal gene network states in Autism spectrum disorder

Abstract

Autism spectrum disorder (ASD) is a prevalent neurodevelopmental disorder characterized by social deficits and associated restrictive and/or repetitive behaviors. The breadth of ASD symptoms is paralleled by the multiplicity of genes that have been implicated in its etiology. Initial findings revealed numerous ASD risk genes that contribute to synaptic function. More recently, genomic and gene expression studies point to altered chromatin function and impaired transcriptional control as additional risk factors for ASD. The consequences of impaired transcriptional alterations in ASD involve consistent changes in synaptic gene expression and cortical neuron specification during brain development. The multiplicity of genetic and environmental factors associated with ASD risk and their convergence onto common molecular pathways in neurons point to ASD as a disease of gene regulatory networks.

Autism spectrum disorder (ASD) is a heterogeneous neurodevelopmental disorder affecting more than 1% of individuals [1]. The main manifestations of ASD are impaired social communication and interaction, repetitive behaviors, and/or restricted interests. The scale of social impairment in individuals with ASD is highly variable and ranges from subtle to most severe conditions that can leave patients unable to lead an independent life. This wide range of symptom severity implies a different degree of impairment of the neuronal networks that regulate social interaction and behavior.

The initial hints about possible neurobiological causes for the etiology of ASD came from human genetic studies that revealed an association between ASD and mutations in genes encoding different components of the synaptic machinery [2]. Clinical data further showed that mutations in the same gene can result in phenotypic variability with a wide range of clinical presentations, as observed in individuals with SHANK3 haploinsufficiency or Phelan-McDermid syndrome [3]. These early findings led to the widely accepted view of ASD as a “synaptopathy” characterized by abnormal neuronal circuit formation during brain development followed by impaired behavior [4] (Figure 1).

Figure 1. Converging pathways in ASD.

Scheme shows the two major pathways implicated in ASD risk based on genomic studies and gene expression analysis of affected individuals. (Left) Dysregulation of gene expression at the level of chromatin modifications, chromatin remodeling, regulation of transcription, and RNA splicing, as well as (right) alterations in synapse development and function are strongly associated with ASD risk. Recent data suggest a convergence of the two pathways in ASD pathology, where changes in neuronal gene expression regulation during fetal brain development preferentially affect genes important for synapse function and neuronal differentiation, and conversely, changes in genes important for synaptic function and neuronal specification indirectly affect neuronal gene expression regulation.

Several recent large-scale sequencing studies, however, led to a shift in our understanding of the genetic mechanisms contributing to the synaptic impairments in ASD. Studies of thousands of families with children with ASD led to the identification of a large number of novel high-confidence genes conferring risk for the disease [5–8]. Many of the affected genes encoded functionally distinct regulators of gene expression that range from chromatin modifiers to bona fide RNA transcription factors [5–8] (Table 1, Figure 1). The role of impaired transcriptional regulation in ASD was furthermore underscored by studies that link ASD with genetic variation in non-coding gene regulatory sequences such as promoters [9]. Machine learning approaches applied to whole-genome sequencing data from 1,902 families identified the strongest association between ASD risk and de novo mutations in evolutionary conserved loci, including transcription factor binding sites, located within distal promoter regions [9].

Table 1. High-confidence ASD risk genes encoding chromatin and transcription regulators.

For each gene, the references for the whole-exome sequencing (WES) studies that implicated the gene in risk with a false discovery rate (FDR) < 0.1, the associated condition reported in the Online Mendelian Inheritance in Man® database (OMIM), the inheritance indicated in OMIM, the OMIM ID, and the references (PMIDs) to mouse models and gene expression studies in mouse models are indicated. For inheritance, AD indicates autosomal dominant, AR autosomal recessive, XLD X-linked dominant, and XLR X-linked recessive. Please, note that all WES studies indicated focused on autosomal genes supposedly acting as haploinsufficient.

	Gene	WES study for ASD association	Condition (OMIM)	Inherita nce	OMI M ID	Mouse models	Gene Expression data
Chromatin regulator	ADNP	De Rubeis et al., 2014; Iossifov et al., 2014; Sanders et al., 2015; Satterstrom et al., 2018	Helsmoortel-van der Aa syndrome	AD	615873	Hacohen-Kleiman et al., 2018; Malishkevich et al., 2015; Vulih-Shultzman et al., 2007	Hacohen-Kleiman et al., 2018
	ARID1B	De Rubeis et al., 2014; Iossifov et al., 2014; Sanders et al., 2015; Satterstrom et al., 2018	Coffin-Siris syndrome 1	AD	135900	Celen et al., 2017; Jung et al., 2017; Shibutani et al., 2017	Celen et al., 2017; Shibutani et al., 2017
	ASH1L	De Rubeis et al., 2014; Sanders et al., 2015; Satterstrom et al., 2018	Mental retardation, autosomal dominant 52	AD	617796	Zhu et al., 2016
	ASXL3	De Rubeis et al., 2014; Satterstrom et al., 2018	Bainbridge-Ropers syndrome	AD	615485
	CHD2	Iossifov et al., 2014; Sanders et al., 2015; Satterstrom et al., 2018	Epileptic encephalopathy, childhood-onset	AD	615369	Kim et al., 2018	Kim et al., 2018
	CHD8	De Rubeis et al., 2014; Iossifov et al., 2014; Sanders et al., 2015; Satterstrom et al., 2018				Gompers et al., 2017; Jung et al., 2018; Katayama et al., 2016; Platt et al., 2017; Suetterlin et al., 2018	Gompers et al., 2017; Jung et al., 2018; Katayama et al., 2016; Platt et al., 2017; Suetterlin et al., 2018
	CREBBP	Satterstrom et al., 2018	Rubinstein-Taybi syndrome 1	AD	180849	Merk et al., 2018; Zheng et al., 2016
	KDM6B	Sanders et al., 2015; Satterstrom et al., 2018				Park et al., 2014	Park et al., 2014
	KMT2C	De Rubeis et al., 2014; Sanders et al., 2015; Satterstrom et al., 2018	Kleefstra syndrome 2	AD	617768
	KMT2E	Iossifov et al., 2014; Sanders et al., 2015; Satterstrom et al., 2018
	MBD5	Sanders et al., 2015; Satterstrom et al., 2018	Mental retardation, autosomal dominant 1	AD	156200	Camarena et al., 2014
	NSD1	Satterstrom et al., 2018	Sotos syndrome 1	AD	117550
	PHF12	Satterstrom et al., 2018
	PHF2	Iossifov et al., 2014; Sanders et al., 2015; Satterstrom et al., 2018
	PHF21A	Satterstrom et al., 2018
	SETD5	De Rubeis et al., 2014; Sanders et al., 2015; Satterstrom et al., 2018	Mental retardation, autosomal dominant 23	AD	615761	Deliu et al., 2018 Moore et al., 2019	Deliu et al., 2018 Moore et al., 2019
	SKI	Satterstrom et al., 2018	Shprintzen-Goldberg syndrome	AD	182212
	SMARCC2	Satterstrom et al., 2018				Tuoc et al., 2017; Tuoc et al., 2013
	SUV420H1/K MT5B	De Rubeis et al., 2014; Sanders et al., 2015; Satterstrom et al., 2018	Mental retardation, autosomal dominant 51	AD	617788
	ZMYND8	Satterstrom et al., 2018
Transcription regulator	BCL11A	De Rubeis et al., 2014; Sanders et al., 2015; Satterstrom et al., 2018	Dias-Logan syndrome	AD	617101	Dias et al., 2016	Dias et al., 2016
	DEAF1	Satterstrom et al., 2018	Dyskinesia, seizures, and intellectual developmental disorder/Mental retardation, autosomal dominant 24	AR/AD	602635/615828	Luckhart et al., 2016; Vulto-van Silfhout et al., 2014
	DYRK1A	De Rubeis et al., 2014; Iossifov et al., 2014; Sanders et al., 2015; Satterstrom et al., 2018	Mental retardation, autosomal dominant 7	AD	614104	Arque et al., 2009; Arque et al., 2008; Benavides-Piccione et al., 2005; Fotaki et al., 2002; Fotaki et al., 2004; Raveau et al., 2018
	FOXP1	Iossifov et al., 2014; Sanders et al., 2015; Satterstrom et al., 2018	Mental retardation with language impairment and with or without autistic features (FOXP1 syndrome)	AD	613670	Araujo et al., 2015; Araujo et al., 2017; Bacon et al., 2015; Frohlich et al., 2017; Usui et al., 2017	Araujo et al., 2015; Araujo et al., 2017; Usui et al., 2017
	FOXP2	Satterstrom et al., 2018	Speech-language disorder-1	AD	602081	Chen et al., 2016; French et al., 2018; Medvedeva et al., 2018; Shu et al., 2005; Xu et al., 2018	Medvedeva et al., 2018; Vernes et al., 2011
	MECP2		Rett syndrome/Mental retardation, X-linked syndromic/Lubs type (MECP2 duplication syndrome)	XLD/XLR	312750/300260	Guy et al., 2007; Hao et al., 2015; Lu et al., 2016; Moretti et al., 2006; Shahbazian et al., 2002; Sztainberg et al., 2015
	MYT1L	De Rubeis et al., 2014; Sanders et al., 2015; Satterstrom et al., 2018	Mental retardation, autosomal dominant 39	AD	616521
	NCOA1	Satterstrom et al., 2018
	POGZ	De Rubeis et al., 2014; Iossifov et al., 2014; Sanders et al., 2015; Satterstrom et al., 2018	White-Sutton syndrome	AD	616364
	SATB1	Satterstrom et al., 2018				Balamotis et al., 2012
	SIN3A	Satterstrom et al., 2018	Witteveen-Kolk syndrome	AD	613406
	TBL1XR1	Satterstrom et al., 2018	Mental retardation, autosomal dominant 41/Pierpont syndrome	AD	616944/602342
	TBR1	De Rubeis et al., 2014; Iossifov et al., 2014; Sanders et al., 2015; Satterstrom et al., 2018	Intellectual developmental disorder with autism and speech delay	AD	606053	Fazel Darbandi et al., 2018; Huang et al., 2014	Fazel Darbandi et al., 2018; Huang et al., 2014
	TCF20	Satterstrom et al., 2018
	TCF4	Satterstrom et al., 2018	Pitt-Hopkins syndrome	AD	610954	Crux et al., 2018; Kennedy et al., 2016	Kennedy et al., 2016
	TCF7L2	Iossifov et al., 2014; Sanders et al., 2015; Satterstrom et al., 2018

A number of large-scale gene expression studies comparing postmortem brains from individuals with ASD and matched unaffected controls underscore the role of impaired epigenetic/transcriptional regulation along with synapse dysfunction in ASD [10–14]. The importance of transcriptional dysregulation in ASD receives additional support from abnormal patterns of RNA splicing and isoform usage in the brains of individuals with ASD. This may have a particular impact on brain development and physiology, since alternative splicing occurs more frequently in the brain than in any other tissue [15]. One specific group of genes carrying microexons (3 to 27 base pairs) is found to be not only preferentially expressed in the brain but also frequently dysregulated in ASD [12,15,16]. Overall, genes showing isoform dysregulation in ASD are likewise enriched in regulators of gene expression and synaptic function [13]. Besides alterations in gene expression and isoform usage, widespread changes in RNA editing in brains from individuals with ASD [17] further point to a broad spectrum of impairments that can contribute to transcriptional dysregulation in ASD.

The possible causal role of perturbed epigenetic and transcriptional processes in ASD has been gaining strong support from animal studies that employ genetically engineered mice lacking/bearing specific ASD risk genes. The haploinsufficiency or neuron-specific targeting of ASD risk genes encoding the chromatin remodelers Arid1b [18–20] and Chd8 [21–23], the histone methyltransferases Ehmt1 [24,25], and Setd5 [26,27], and the transcriptional regulators Foxp1 [28–31] and Foxp2 [32–35] result in ASD-like behavioral phenotypes in mice (Table 1). Transcriptomic analyses of the brains of mice with haploinsufficiency or cell-type specific ablation of chromatin modifiers or transcription regulators revealed consistent changes in the expression of synaptic genes [18,24,26,27,29,36–38] (Table 1), suggesting impaired transcriptional regulation as one of the key mechanisms for altered synaptic gene expression and function in ASD (Figure 1).

One of the peculiar aspects of the ASD-related transcriptional changes deals with the selective impairment of the expression of genes of extended length (>100kb). Transcriptome analysis in the human and mouse brain revealed a significant enrichment in the expression of long genes as compared to any other organ [39,40]. Additionally, genes of extended length are enriched in genes encoding synaptic proteins and ion channels, including those that have been linked to ASD [39,41,42]. The efficiency of gene transcription, as defined by the abundance of the full-length mRNA transcript as well as the pattern of RNA splicing, depends greatly on the fidelity of RNA elongation [43]. The latter is governed by numerous factors that collectively support the movement of the RNA polymerase at a defined speed of RNA elongation [44]. Therefore, it is very likely that the transcription of genes of extended length is more sensitive to changes in the transcriptional machinery as compared to shorter genes. Recent data revealed the particular dependency of ASD risk genes of extended length on the Top2b-dependent transcriptional elongation process [41]. Moreover, long genes linked to ASD appear to be distinct from other genes by harboring expanded enhancer domains [45], and by being particularly sensitive to transcriptional repression in response to pharmacological inhibitors of the bromodomain - containing proteins of the BET family [42].

The impaired transcription of genes of extended length as well as other structural and functional alterations associated with transcriptional regulation in ASD may also lead to erroneous timing of gene expression during the tightly regulated developmental trajectories of neurons in the developing brain. The differentiation of specific neuronal subtypes during brain development is governed by complex gene regulatory networks [46], with each cell type acquiring a unique expression profile dictating their morphological and phenotypic specialization [47–49]. Time course analysis of developing human brain tissue revealed that late cortical development in the fetus is characterized by widespread changes in gene expression patterns, including increased neuron subtype-specific signatures and the expression of genes associated with synapse development and neuronal functions [50]. These data suggest that the convergence of the varied ASD risk genes onto the same behavioral outcome may reflect the defective timing of neuronal subtype specification and associated circuit formation during pre- and post-natal brain development. In other words, genetic lesions associated with ASD act as a form of “chaotropic” agents that, by acting on multiple pathways during the extremely precise differentiation processes in the fetal brain, affect the stability of neuronal networks. As a consequence, the social challenges that occur imminently after birth confront a neuronal network with impaired robustness and hence increased susceptibility to activity-driven changes that may lead to the establishment of the disease phenotype. This notion is in line with evidence of post-natal or adult activity-driven transcriptional changes in the brain of different mouse models of the disease [24,42,51,52]. The described scenario views attenuated brain robustness, rather than defects in a specific gene or cell type, as an underlying mechanism for ASD pathophysiology.

The robustness of cell differentiation reflects the so-called “canalization” process towards a specific outcome from uncertain starting conditions [53] (Figure 2). In Waddington’s “epigenetic landscape”, environmental signals lead to the establishment of “valleys” that guide the direction of the differentiation processes to the finite cell type [53] (Figure 2A). Projected into this landscape (Figure 2B), the process of neuronal differentiation and sub-specification is reflected in the trajectories of a multipotent neuronal progenitor cell. Exposed to a combination of external and internal signals (growth factors, signaling molecules, transcriptional regulators), the progenitor cell is guided through the differentiation process acquiring new features that ultimately will define its distinct neuronal identity and function. This process of neuron sub-type specification is governed by high “ridges” that prevent divergence during the differentiation process and stabilize the newly acquired phenotypes.

Figure 2. Robustness of cell/neuron differentiation during development.

Modified scheme of Waddington’s “epigenetic landscape” [53] illustrates cell differentiation during development. (A) Multipotent stem cells and (B) neuronal progenitor cells (white circle) follow a robust developmental trajectory or “canalization” towards a specific outcome. In this “epigenetic landscape”, distinct extrinsic and intrinsic signals lead to the establishment of “valleys” and “ridges” that ensure robustness and guide the differentiation processes towards distinct differentiated cell types (A) or neuronal subtypes (B).

The Waddington landscape reflects the state of gene regulatory networks that operate within cells [54]. In turn, the notion of a stable dynamical state invites the comparison to the so-called “attractor state” which, in material physics, is defined as a place where the dynamical system is exerting a minimal amount of energy [54]. The attractor state, which has been widely discussed in the context of gene network regulation during development [54], represents a defined outcome of numerous interactions within any given network, from transcriptional networks to intercellular interactions. Each of the interactions within such a network represents a single dimension and, accordingly, multiple interactions yield highly complex multidimensional manifold. The topography of the manifold is molded into a conformation of “ridges” and “valleys”, where unstable, high energy states occupy the top of the “ridges” and low energy states form the energetically favorable attractor states or “valleys” [55] (Figure 3). Thus, the low energy level of the attractor state contributes to its stability and protection against environmental perturbation [56]. The stability of a given attractor state can be attenuated by introducing systemic alterations (genetic mutations and/or environmental insults) that “lower” the protective “ridges” of the attractor state or increase the instability of the network by altering numerous different network components or by targeting a key regulator of the network [57,58]. Both of these scenarios may yield abnormal cell or tissue function.

Figure 3. Epigenetic landscape of normal brain development and ASD.

(A) Modified scheme from Huang et al, 2009 [53] of an epigenetic landscape during brain development. The landscape is a schematic projection of a complex gene network into a two-dimensional state space. The y axis represents the relative stability of individual cell states where higher positions indicate less stability. The valleys represent stable attractor states that occupy the low-energy stable basin and are resistant to perturbations. Normal developmental trajectory (blue) progresses from back to front towards a stable attractor state, which represents a distinct neuronal state, and is prevented from entering unused “abnormal attractors” (red dashed circle) along the path due to epigenetic barriers (orange area). Mutations or environmental insults can lower this barrier, thus opening access to unused attractors that encode an abnormal phenotype = ASD attractor state (red dashed arrow). Alternatively, the ASD gene network state may reflect an unstable neuronal state that hinders the formation of stable neuronal networks. (B) Model of ASD network states and their potential reversibility. Scheme shows highly simplified version of proposed model for normal (blue) and ASD (red) gene network states. The y axis represents the relative stability of individual cell states where higher positions indicate less stability. The x axis represents the specific space coordinate of a given neuronal network state. Future potential therapies could be aimed at trying to reverse symptoms of ASD by targeting the ASD network states (red, novel ASD attractor state or unstable state) followed by their conversion into a normal stable neuronal network state (blue).

Notably, it has been shown that undifferentiated progenitor cells can reach the same finite differentiation state in response to a set of different stimuli [54]. Despite the differences in individual stimulus-induced pathways, the distinct gene expression processes that follow the signal eventually converge onto the same attractor state/gene expression profile [54]. The same principle could be applicable to neurodevelopmental disorders like ASD, where distinct alterations within the neuronal gene network may lead to a similar outcome such as i.e. the dysregulation of synaptic gene expression and associated functions in ASD. Since diseases are not a hardwired part of our evolution, the disease-associated attractor states could be highly individual and reflect subtle differences in the energy landscape of the interacting components. The impaired attractor state model has been discussed in the context of malignant transformation, where genetic mutations and environmentally induced expression changes of distinct genes trigger the formation of immortal and invasive states characteristic of cancer cells [56,58,59]. Following this model, it is possible that the different genetic and environmental factors contributing to ASD risk, while targeting different sets of genes during the critical phase of fetal and early post-natal brain development, may converge on a common neuronal “attractor state” and induce a switch to a new phenotypic state. This newly acquired state may either become stabilized in form of a “new attractor state” (Figure 3) or remain permanently unstable, hence generating multiple different, slightly heterogeneous phenotypes (Figure 3). In summary, it could be feasible to view ASD as a pathological and perhaps unstable attractor state, where slight variations in the input factors drive the severity of the clinical manifestations. The important aspect of this model is that it directs much of the attention from individual genes towards genetically and/or environmentally impaired gene regulatory networks and associated alterations in neuronal function during development. This very concept of the attractor state bears a certain futility as the multiplicity of interactions may preclude the identification of key driver genes or cell types.

The idea of ASD reflecting an unstable neuronal regulatory network state may be particularly interesting from a therapeutic point of view since it suggests a possible reversibility of the phenotype (Figure 3B). This scenario is supported by recent data showing that some of the transcriptional and/or behavioral changes in mice are reversed by restoring expression or function of ASD risk genes in the adult brain [42,60–65]. Collectively, these data suggest the exciting possibility that ASD-associated neuronal network states could be reverted to their predestine state (Figure 3B) and allow the amelioration of social deficits. This concept is supported by intriguing examples of reported temporary alleviation of ASD symptoms in humans in response to strong perturbations, such as increased body temperature during fever [66], and suggest potential novel therapeutic approaches could be aimed at rewiring neuronal gene networks/attractor states in ASD.