Autism spectrum disorder (ASD) comprises a heterogeneous class of neurodevelopmental disorders characterized by impaired social interactions, restrictive interests and repetitive behaviors (Landa, 2008). ASD typically presents with other mental and physical disabilities such as anxiety, attention-deficit/hyperactivity disorder (ADHD), intellectual disability (ID), epilepsy and impairments in motor coordination. In up to 25% of individuals diagnosed with ASD, an identifiable or genetic variant can be identified, providing valuable insights into the mechanisms involved in proper neurodevelopment (Huguet et al., 2013). A large number of ASD-linked genes are also associated with broad processes such as metabolism, chromatin remodeling, mRNA regulation, protein synthesis and synaptic function.

The human brain contains about 86 billion neurons making trillions of connections (Azevedo et al., 2009). Most neurons are produced in the ventricular zone (VZ) and migrate radially out into the developing neocortex, and it is estimated that about 75% of rodent, and up to 90% of human neurons use glial-guided migration (Letinic et al., 2002). The movement of neurons along migratory routes is guided by a number of guidance molecules that direct their movement into the cortex and the formation of an organized 6-layered structure (Marín and Rubenstein, 2001; Huang, 2009; Valiente and Marin, 2010).

After neuronal migration, neurons must undergo extensive morphological changes. Long axonal processes are extended and are required to connect to target neurons with precision, while complex dendritic arbors must grow and branch to occupy specified dendritic field volumes. These processes take place during prenatal and early postnatal periods and lay the foundations for neuronal connectivity within and across brain regions. Ultimately, with activity-dependent structural remodeling, neurons form synaptic connections and incorporate into functional neuronal networks for proper brain function. It’s of no surprise that disruptions in any of these intricate processes would cause abnormalities in brain development and function, leading to neurodevelopmental disorders including ASD.

While ASD shares characteristic features at the behavioral level, its underlying causes are highly heterogeneous. Developmental dysregulation in ASD may affect processes ranging from progenitor cell proliferation and neuronal differentiation to neuron migration, axon guidance, dendrite outgrowth, synaptogenesis, synaptic function and neural circuitry.

Studies of ASD-related brain pathologies indicate that abnormal acceleration of brain growth in early childhood (Wegiel et al., 2010) accompanied with impaired neuron morphological development and brain cytoarchitecture are common features in ASDs (Bauman and Kemper, 1985; Bailey et al., 1998; van Kooten et al., 2008). Additionally, impairments in synapse formation and synaptic plasticity (Bourgeron, 2007, 2009), which ultimately lead to functional and cognitive impairments, are fundamental causative factors underlying ASD pathology.

Analysis of the Simmons Foundation Autism Research Initiative (SFARI) gene database shows that ASD causative genes display vast diversity involving up to one thousand genes (Banerjee-Basu and Packer, 2010; Abrahams et al., 2013). Additionally, a large number of rare genetic variants in protein-coding genes are causative for ASD, none of which individually account for more than 1% of the total number of ASD diagnoses. Therefore, the complexity and heterogeneity of autism genetics is a major challenge when investigating the underlying neurobiological pathways that are shared within ASD (Happe et al., 2006).

The ability to classify ASD patients according to genetics has been enhanced by the substantial amount of work done to understand ASD-linked genes, and the role of their encoded genes, in the underlying neuropathologies. For example, well known ASD-linked genes, such as neurexin, neuroligin and Shank have been well characterized for their roles in synaptic formation and function (Toro et al., 2010). Additionally, co-expression network analyses of ASD-linked genes have identified that early developmental periods when neurogenesis and synaptogenesis occur are commonly disrupted processes in ASD (Parikshak et al., 2013). Genes involved with regulating developmental processes such as Chromodomain Helicase DNA-binding protein 8 (CHD8), T-Box Brain Protein 1 (TBR1), and Fragile X Mental Retardation 1 (FMR1), have also been linked to ASD (Parikshak et al., 2013; Willsey et al., 2013). These findings suggest that multiple processes during prenatal and early postnatal brain growth are linked to ASD, underscoring the vast heterogeneity of causative factors leading to ASD pathogenesis. ASD-risk genes, their associated disorders and the neurodevelopmental processes they affect are shown in Table 1.

Neurons are highly specialized cells with distinct morphologies comprised of three distinct sections: the soma which contains the nucleus and the majority of the cellular organelles; a long axonal process to transmit information; and a complex dendritic arbor to receive information from neighboring neurons. The dendrites are highly branched and elaborate and therefore occupy a large area within neural tissues (Tahirovic and Bradke, 2009).

Dendrite growth can be viewed as discreet steps. After birth from neural progenitors, neurons start as a simple round soma and must first undergo cellular polarization. Neurons first adopt a multipolar morphology with the extension of minor neurites. The movement of migrating neurons is achieved through the formation, maintenance and constant transformation of microtubules in response to extracellular cues and intracellular polarity signals. During migration, the cell first extends a leading process. Stabilization of a single neurite is required for newly generated neurons to exit the multipolar stage to enter the cortical plate. This stabilization ultimately results in formation of the leading process while the trailing process eventually develops into the future axon (Shim et al., 2008; Witte and Bradke, 2008). Next, in order to move forward, the nucleus undergoes a translocation into the stabilized leading neurite. During this process, termed nucleokinesis, the attachment of microtubules from the centrosome to the nuclear envelope exerts a traction force, pulling the nucleus into the leading neurite attached to a radial glia process (Bellion et al., 2005; Tahirovic and Bradke, 2009).

Once neurons reached their destined cortical layers, substantial dendritic outgrowth is undertaken to form dendritic arbors characteristic of a neuron subtype. At this stage, neurons are tasked with the processes of growing each branch to the correct size, initiating a new branch at the right site to complete a specific branching pattern, and directing each branch’s growth into an appropriate spatial location (Tahirovic and Bradke, 2009; Jan and Jan, 2010). The early postnatal stages of neurite growth are depicted in Figure 2.

Dendrites are vastly different from axons in their ultimate morphology, function and developmental processes. In mammalian neurons, a notable distinction is the manner in which the microtubules are organized. Initially all neurites contain microtubules oriented with their plus-end localized distally from the soma. The neurite specified to become the axon will maintain this distal plus-end orientation, while neurites that become dendrites will adopt a mixed orientation (Baas et al., 1988, 1989; Burton, 1988). Additionally, in comparison to axons, dendrites have an enrichment of cellular organelles that are transported with the aid of microtubule motors like dynein, such as Golgi outposts and mitochondria, which are utilized to supply the necessary for cellular materials required for growth (Horton and Ehlers, 2004; Horton et al., 2005; Kapitein et al., 2010).

The morphology of the dendritic arbor is largely developed early during the embryonic period; however, dendrites are highly dynamic and they maintain the overall morphology with various mechanisms into adulthood. Disruptions in dendritic growth, or breakdown of the mechanisms to maintain their morphology can be deleterious, resulting in aberrant network function. Abnormalities in neuronal connectivity between the higher-order association areas have been considered one of the major defects in ASD (Geschwind and Levitt, 2007). MRI studies have shown both large and small changes in connectivity of neuronal networks; however, the neurobiological basis for this disconnectivity remains to be fully elucidated (Casanova and Casanova, 2014; Maximo et al., 2014). A large number of autism studies have focused on dendritic development, although mostly relating to spine morphology and synaptic function, which will be discussed in detail below (Persico and Bourgeron, 2006; Kelleher and Bear, 2008; Bourgeron, 2009). However, the current body of research suggests that this is not the only morphological defect, as children with autism are frequently identified with large-scale anatomical abnormalities, suggesting dysregulation in dendritic growth and development. For example, both macrocephaly and microcephaly are identified in ASD adolescents, with 15% of ASD patients presenting with macrocephaly, while 20% display microcephaly (Lainhart et al., 1997; Fombonne et al., 1999; Cody et al., 2002; Pardo and Eberhart, 2007).

One of the proposed causes leading to macrocephaly has been suggested to be a result of increased dendrite number and size, which may be a product of excessive dendrite arborization, and/or decreased dendrite pruning (Jan and Jan, 2010). Other factors can play a role in macrocephaly, such as increased numbers of neurons and glia, however the number of dendrites may better explain certain forms of macrocephaly. Brain volumes in infants diagnosed with ASD are typically normal, however they display aberrant overgrowth as development progresses. This later onset of increased brain volume should not result from an increase in neuronal number, as the largest areas of the brain have mostly completed neurogenesis prior to birth, except for areas like the hippocampus where smaller amounts of neurogenesis occur later in life (Ming and Song, 2005; Zhao et al., 2008). The timing of increase brain size can be explained by aberrant dendrite growth and branching, which largely occurs postnatally. In humans, neurons grow and continue to elaborate their dendritic arbors, axons and form new synaptic connections until the age of five, with experience-based remodeling of synapses until 20 years of age (Stiles and Jernigan, 2010; Tau and Peterson, 2010; Pescosolido et al., 2012). Dendrites continue to extensively grow after birth, in an activity-dependent manner. Therefore, a large contribution to the increase in brain size postnatally likely results from increased dendritic growth (Redmond et al., 2002).

Previous genetics studies have described multiple high-risk genes for autism that play roles in diverse functions, including synaptic connectivity and synapse function, dendritic and axonal growth, trafficking, transcription and translation (Volders et al., 2011; de Anda et al., 2012; Miao et al., 2013; Bakos et al., 2015). Recently, estimates have suggested that 88% of the genes that are considered to be high-risk for ASD play a role in early neurodevelopmental functions such as neurogenesis and differentiation of neuroblasts. Importantly, ~80% of these genes are involved in later phases of neurodevelopment and regulate processes involved in neurite growth and synapse formation (Casanova and Casanova, 2014).

Dendrites from an individual neuron can have a thousand or more spines, with each spine forming an excitatory synaptic connection upon maturing. Spines and synapses are produced in excess numbers during development, but synaptic numbers are later fine-tuned through activity-dependent stabilization or elimination (Changeux and Danchin, 1976). Spines are highly dynamic and undergo constant turnover and morphological plasticity with a dependency on both developmental stage and activity.

Dendritic spines are typically classified as thin, stubby and mushroom, with the latter considered more mature. Mushroom and stubby morphologies are more permanent and form strong excitatory connections (Trachtenberg et al., 2002; Kasai et al., 2003). Alternatively, thin spines are highly dynamic, shorter lasting, and form weak synaptic connections or no connection at all. Within the spine head, actin becomes enriched helps play a key role in spine formation and structural dynamics (Chazeau and Giannone, 2016). Previous studies have shown that spine formation maintenance are a major cellular processes affected in ASD (Kelleher and Bear, 2008; Bourgeron, 2009; Phillips and Pozzo-Miller, 2015). Using Golgi staining, postmortem analysis of cortical neurons from ASD brain samples showed an increase in dendritic spine density compared to normal patients (Hutsler and Zhang, 2010). Some of the high-risk autism genes implicated in dendritic growth and branching spine formation and synapse maturation are described below.

Methyl CpG binding protein 2 (MECP2) is an X-linked gene that codes for a protein that functions as a transcriptional repressor and an activator (Chahrour et al., 2008). Mutations in MeCP2 were initially linked to Rett syndrome (RTT), a neurodevelopmental condition that presents with motor and speech impairments, cognitive deficits and autism (Amir et al., 1999). RTT is typically caused by loss-of-function mutations in MeCP2; however, there are rare cases that are also caused by MeCP2 duplications.

Using Golgi staining in brain slices from MeCP2 KO mice, reductions in dendritic growth and branching have previously been reported in both apical and basal arbors of motor cortical neurons (Kishi and Macklis, 2004; Stuss et al., 2012). Additionally, to investigate MeCP2 duplication, mice over-expressing the human MeCP2 gene have shown excessive dendritic branching, indicating that MeCP2 over-expression can induce dendritic overgrowth (Jiang et al., 2013). It is possible that in subtypes of neurons MeCP2 can act as a repressor while acting as an activator in others, or it can change roles at different periods in neuronal development. For instance, MeCP2 can activate genes involved in early dendrite growth and repress genes later during dendritic remodeling. An important aspect to these findings is that MeCP2 overexpression affects the dendrites, having no impact on axon growth, suggesting a defined role for MeCP2 in dendritic development. Recent work has provided evidence for the role MeCP2 plays in dendritic development, showing that MeCP2 also regulates the expression of genes post-transcriptionally. These findings show that MeCP2 regulates microRNA (miRNA) processing via a direct interaction with DiGeorge syndrome critical region 8 (DGCR8), which plays a role in mediating the genesis of miRNAs thought to regulate dendrite morphogenesis (Gregory et al., 2004; Cheng et al., 2014).

RTT animal models also show changes in excitatory hippocampal synapse numbers. The loss of the X-linked RTT protein MeCP2 has also been show to result in abnormal dendritic spine morphology and a decrease in spine density (Zhou et al., 2006; Chapleau et al., 2012; Stuss et al., 2012). Additionally, MeCP2 over-expression produces dendritic overgrowth in mice and these animals show a greater rate of spine turnover, with a bias toward spine removal (Jugloff et al., 2005; Jiang et al., 2013). Taken together, MeCP2, the established factor underlying RTT with autism, has a clear role in dendritic morphogenesis and synapse formation, both of which should play a major function in the cognitive impairments seen in RTT patients.

Fragile X mental retardation gene 1 (FMR1) is the gene underlying the disorder Fragile X syndrome (FXS), which results in ID with 15%–30% of patients also displaying ASD phenotypes (Krawczun et al., 1985; Persico and Bourgeron, 2006; Kelleher and Bear, 2008; Santoro et al., 2012). FXS is usually results from an expansion of a CGG triplet in the 5′-UTR region of the FMR1 gene, however a few missense mutations and deletions have been identified (Santoro et al., 2012). Fragile X mental retardation protein (FMRP), the FMR1 gene product, plays a key role in negatively regulating translation, especially local translation at the synapse (Santoro et al., 2012). In terms of FMRP’s role in dendritic growth, mouse studies have shown somewhat contradictory results. In visual cortex pyramidal neurons of FMR1 KO mice, one study has shown defects in dendritic spines, with no observable changes in dendritic morphology (Irwin et al., 2002). Conversely, multiple other studies have shown that FMRP is critical for dendritic growth and branching. In FMR1 KO mice, visual cortex pyramidal neurons show reduced basal dendrite length and branching (Restivo et al., 2005). Using neural stem cells isolated from FMR1 KO mice, or from postmortem tissues of FXS human fetuses, differentiated neurons showed fewer and less complex neurites with smaller somas (Castrén et al., 2005). An important caveat to these studies was that they involved loss-of-function or deletion of FMRP; however, FXS in humans is rarely caused by deletions or missense mutations in the FMR1 gene, but rather via an expansion of the CGG triplet repeat. This has been addressed in studies using transgenic mice with a FXS knock-in mutation consisting of 120–140 CGG repeats. Indeed, these animals display significantly impaired dendritic morphogenesis in addition to alterations in dendritic spine density and morphology (Berman et al., 2012). The studies from Drosophila have elucidated the mechanistic function of FMRP in dendritic growth and have shown that FMRP is involved with transportation of the mRNA for Ras-related C3 botulinum toxin substrate 1 (Rac1), a GTPase. In agreement with this, Rac1 was found to bind FMRP and affect dendrite arborization, indicating that FMRP’s role in dendrite morphogenesis could in part be through its interaction with Rac1 (Lee et al., 2003). These findings strongly support a role for FMRP in dendritic arbor morphogenesis.

FMRP is localized within dendrites and in addition to abnormal dendritic growth and branching, brains from FXS patients display an immature synaptic phenotype (Rudelli et al., 1985). FXS patients have an increased spine density on both apical and basal dendrites in neocortex and more spines characterized by an immature morphology. Additionally, Golgi studies from human FXS patients reveal a significant increase in long spines with less shorter spines compared to controls in multiple cortical areas (Hinton et al., 1991; Irwin et al., 2001). In FMR1 KO mouse models, spine phenotypes correlate with those observed in humans with FXS as these mice have an increase in longer spines and a corresponding decrease in shorter spines. Additionally, dendritic spines in FXS mice display a more immature morphology with fewer mushroom and stubby spines present (Irwin et al., 2001, 2002; McKinney et al., 2005). In addition, changes in synaptic proteins such as postsynaptic density-95 kDa (PSD-95) have been observed in FXS (Ifrim et al., 2015).

Although FMRP’s role in impaired spine formation and morphology in FXS has not been completely elucidated, FMRP has been identified to interact with multiple proteins that are linked to dendritic and spine regulation. Cytoplasmic FMRP-interacting protein 1 (CYFIP1), a binding partner of FMRP, is a protein that has recently generated interest as its genetic locus is chr15q11.2, a susceptibility area in ASD. When bound to FMRP, CYFIP1 can inhibit translation and regulates actin dynamics. This can therefore regulate the growth and removal, as well as the labiality and morphology of spines (De Rubeis et al., 2013). A down-regulation of CYFIP1 mRNA has been detected in subgroups of FXS patients with ASD (Nowicki et al., 2007). Conversely, over-expression of CYFIP1 results in higher dendritic complexity in vitro, while neurons that are haploinsufficient for CYFIP1 show decreased dendritic complexity and an increase in the relative levels of immature to mature dendritic spines (Pathania et al., 2014).

Pten is a phosphatase that de-phosphorylates PIP3, which serves to inhibit PI3K/AKT/mTOR signaling (Kwon et al., 2006; Ogawa et al., 2007; Garcia-Junco-Clemente and Golshani, 2014). Mice lacking Pten selectively in the central nervous system show an increase in activation of the AKT/mTOR/S6K pathway, an increase in neuron size, macrocephaly, as well as decreased and disorganized dendritic and axonal growth (Kwon et al., 2006). Pten KO during early development, as well as in adult mice disrupts neuron morphogenesis, suggesting that Pten plays an important role in dendritic growth and maintenance in adulthood in addition to its function during early neurodevelopment (Kwon et al., 2006; Chow et al., 2009). Therefore, Pten, a protein with a clear link to autism, has a definitive role in dendrite morphogenesis.

TSC1 and TSC2 are tumor-suppressing genes that have been linked to brain tumors in tuberous sclerosis complex (TSC; Huang and Manning, 2008). However, their involvement in neurodevelopmental disorders including epilepsy, autism and ID has become increasingly studied (Kwiatkowski and Manning, 2005; Persico and Bourgeron, 2006). In pheochromocytoma 12 (PC12) cells, an inducible neuron-like cell line, transfection of TSC1 antisense oligonucleotides was shown to increase neurite outgrowth through RhoA activation, while knockdown of Tsc2 decreased neurite growth (Floricel et al., 2007). In another study, overexpression of Tsc1/Tsc2 was found to impair axon formation. Knockdown of Tsc1/Tsc2 in vitro induced multiple axons, while genetic deletion in vivo in the mouse induced ectopic axons (Choi et al., 2008). There is strong evidence that TSC genes are linked to ASD risk and play a major role in neurite growth, therefore it will be important to further elucidate their mechanistic role in neuron morphogenesis and ASD in future research.

Mammalian target of rapamycin (mTOR is a serine/threonine kinase that regulates cellular growth and is induced by growth factors and environmental cues (Laplante and Sabatini, 2012). Via regulation of the cytoskeleton, mTOR plays an important function in regulating dendritic outgrowth, and proteins known to inhibit mTOR signaling have been associated with aberrant dendrite and spine development in ASD (Thomanetz et al., 2013; Skalecka et al., 2016), including Pten and Tsc 1/2 (Weston et al., 2014). Pten serves to inhibit the PI3K/AKT/mTOR signaling pathway, thereby affecting growth and protein translation, and Pten mutations have been discovered in multiple individuals diagnosed with ASD. Additionally, mice carrying PTEN mutations or genetic deletions show impaired social interactions and increased sensory responses. Loss of Pten produces an increase in dendritic growth, synaptic connectivity and disorganized dendritic and axonal processes (Kwon et al., 2006; Orloff et al., 2013). Conversely, knockdown of Pten in the amygdala decreases spine density, with an increase in mushroom spines and decrease in thin protrusions (Haws et al., 2014). Additionally, it has been previously shown that the TSC pathway plays an important role in regulating synaptic function, and in hippocampal neurons, loss of Tsc 2 expression produces an expansion of neuronal somas as well as spines (Tavazoie et al., 2005). These findings provide strong evidence that mTOR, and its interacting proteins, are involved in regulating both dendritic and spine development in ASD.

Neurofibromatosis is a condition in which tumors grow in the nervous system. Mutations in Neurofibromatosis-1 (NF1) produce neurofibromas and between 30% and 65% of children with NF1 mutations display learning disabilities and display significantly higher rates of autism, suggesting a causal relationship to autism (Rosser and Packer, 2003; Garg et al., 2013). Neurofibromin, the protein product of the NF1 gene, is a GTPase activating protein that negatively regulates the Ras signaling pathway (Costa and Silva, 2003). NF1 conditional KO mice display impaired dendritic morphogenesis and Golgi staining from NF1 KO mouse brains reveals shorter apical dendrites in pyramidal neurons. NF1’s function in dendrite morphogenesis has been shown to act through cAMP/PKA/Rho/ROCK signaling (Brown et al., 2012). RhoA plays a major role in actin dynamics and neurite growth and branching. With the inhibition of RhoA, an increase in the branching of neuronal processes is observed, and with the activation of RhoA, a decrease in the length and complexity of processes is observed (Li et al., 2000; Nakayama et al., 2000; Wong et al., 2000).

Loss of expression of KIAA2022/KIDLIA was previously identified by our group and others as the causative protein for severe ID and autistic behavior in multiple families (Cantagrel et al., 2004, 2009; Ishikawa et al., 2012; Van Maldergem et al., 2013; Charzewska et al., 2015; Kuroda et al., 2015). Clinical examinations of patients with KIDLIA mutations or a loss of expression, display enlarged ventricles, Virchow-Robin spaces, a thin corpus callosum and small cerebellar vermis. Additionally, strabismus has been observed in patients, with some dysmorphic features including a round face during early postnatal periods, febrile seizures, severely impaired or no language, stereotypical hand movements and delayed acquisition of motor milestones (Cantagrel et al., 2004, 2009). Interestingly, a patient with decreased expression of KIDLIA had only mild cognitive deficits with a significant delay in language acquisition as well as repetitive and stereotyped behaviors, indicating that the effects may be gene-dosage dependent (Van Maldergem et al., 2013).

With shRNA-mediated knockdown of KIDLIA in rat hippocampal neurons in culture, significant impairments in dendritic growth and branching are observed (Van Maldergem et al., 2013; Gilbert and Man, 2016). Mechanistically, we’ve recently reported that loss of KIDLIA significantly impairs actin dynamics and produces an aberrant increase in total and membrane-localized N-cadherin. N-cadherin was found to bind much greater levels of δ-catenin, thereby releasing the latter’s inhibition on downstream RhoA (Gilbert et al., 2016). Additionally, microcephaly has been reported in human patients with a loss of KIDLIA expression (Cantagrel et al., 2004; Van Maldergem et al., 2013; Kuroda et al., 2015). These findings suggest that the regulation of neuronal morphogenesis through dendrite growth and synapse formation are major underlying factors contributing to the cognitive impairments observed in patients with genetic deletions or functional mutations in KIDLIA.

Ubiquitin protein ligase E3A (UBE3A) is the causative gene for Angelman syndrome, a developmental disorder characterized by language impairments, ataxia, ID and hyperactivity (Williams et al., 2006). Individuals identified with Angelman syndrome are often co-diagnosed with autism (Steffenburg et al., 1996; Peters et al., 2004). In addition, Ube3A over-expression leads to autism in humans and in animal models (Bucan et al., 2009; Smith et al., 2011; Noor et al., 2015). UBE3A is an imprinted gene that encodes an E3 ubiquitin ligase, that typically expresses from both maternal and paternal alleles in most tissues, but is only expressed from the maternal allele in brain (Mabb et al., 2011). Previous studies have shown that Ube3A is required for normal dendritic morphogenesis in the mouse (Miao et al., 2013) and shRNA-mediated knockdown of Ube3A in vivo impairs dendritic growth cortical pyramidal neurons in the mouse (Miao et al., 2013). Additionally, studies in Drosophila strongly support a role for Ube3A in dendritic growth and branching. Both over-expression and loss of the Ube3A Drosophila homolog have been shown to result in decreased dendrite branching in larval sensory neurons (Lu et al., 2009). These studies suggest that Ube3A plays a significant role in dendrite growth and support that dysregulated neuron morphogenesis may underlie developmental disorders like Angelman syndrome and ASD.

Thousand-and-one amino acid kinase 2 (TAOK2) belongs to the family of MAP kinase kinase kinases (MAPKKK) and is located in chromosome 16, an area associated with increased ASD-risk (Weiss et al., 2008) and schizophrenia (McCarthy et al., 2009). Both Taok1 and Taok2 activate mitogen-activated protein kinase (MAPK) pathways via JNK and p38 which results in regulation of gene transcription (Chen and Cobb, 2001; Chen et al., 2003). Additionally, FMRP regulates Taok2 mRNA (Darnell et al., 2011), the protein underlying FXS, providing an additional link to its involvement in neurodevelopmental disorders. A recent study in vivo has shown that knockdown or overexpression of Taok2 showed opposing effects on basal dendrite development in the neocortex (de Anda et al., 2012). Specifically, primary dendrite numbers were reduced after Taok2 knockdown, while the number of primary dendrite numbers was increased with over-expression. Additionally, this effect was preferential for the basal dendrites on basal dendrites, as Taok2 knockdown produced no changes in apical dendrite morphology on the same neurons. The role of Taok2 on dendrite growth was dependent on interactions with Neuropilin 1, a membrane receptor that binds Semaphorin 3A, which leads to initiation of the JNK cascade (de Anda et al., 2012). Interestingly, this finding shows that Taok2 expression selectively affects specific dendritic areas, i.e., the basal dendrites, providing evidence that specialized molecular pathways are used for the formation of different dendritic areas.

Reelin, as discussed previously for its role in neuron migration, seems to also play a significant function in dendrite arborization in hippocampal and cortical neurons (Niu et al., 2004; Jossin and Goffinet, 2007; MacLaurin et al., 2007; Chameau et al., 2009; Hoe et al., 2009; Matsuki et al., 2010). Reelin, and its downstream signaling pathway through Vldrl/Apoer2-Dab1, serves to promote dendrite development. Reeler mice contain a loss-of-function mutation in RELN, and display decreased dendrite branching in the hippocampus (Niu et al., 2004). Additionally, Reelin has also been shown to affect cortical dendritic growth in vivo (Hoe et al., 2009) and Reelin application to Reelin KO mouse brain slices can promote dendritic growth (Nichols and Olson, 2010). The role of alterations in neurite growth and branching and the associated ASD-linked genes, are described in detail later.

Synapses are highly specialized structures required for signal transduction and plasticity within neuronal networks, making up the functional contact sites between neurons. A synapse is composed of the axon terminal, the presynapse, a synaptic cleft which contains adhesion proteins and the extracellular matrix, and the postsynaptic density with receptors on a target neuron’s dendrites. The presynapse contains mitochondria and is characterized by a pool of synaptic vesicles filled with neurotransmitters. Action potentials arriving at the axon terminal mediate calcium influx, which causes the vesicles to fuse with specialized regions of the plasma membrane called active zones to release their content into the synaptic cleft. A precise coupling between the electrical stimulus (action potential) and release of neurotransmitter is crucial for proper signal transmission to the postsynaptic neuron. The postsynapse is characterized by the presence of receptors that bind neurotransmitters released from the presynapse, which initiates signaling cascades that ultimately propagate the electrical signal in postsynaptic neuron. Receptors are clustered in a region called the postsynaptic density (PSD) in the excitatory synapse.

Synapses are classified as excitatory and inhibitory, depending on whether they use glutamate or γ-aminobutyric acid (GABA) as their main neurotransmitter, respectively. Additionally, the formation and structure of excitatory and inhibitory synapses is uniquely different. Whereas excitatory synapses made on dendritic spines, inhibitory synapses are formed directly on the dendritic shaft. Additionally, both excitatory and inhibitory synapses have distinct proteomic profiles to specialize each synapse with the appropriate receptors and signaling molecules. Synaptic dysfunctions, whether they arise from functional mutations in pre- or postsynaptic proteins, are a common underlying pathology in ASD and are discussed in detail below.

Based on the different ASD-linked pathways discussed previously, it can be inferred that changes in synaptic plasticity via altered synaptic strength and/or number, may be an underlying pathology in ASD patients and mouse models. Interestingly, many ASD-linked mutation results in altered gene transcription and protein synthesis of synaptic related transcripts, effects which can also be observed with changes in neuronal activity (Kelleher and Bear, 2008; Akins et al., 2009; Darnell et al., 2011; Qiu et al., 2012; Gilbert and Man, 2014).

Initial findings from FMR1 mutant mice, the mouse model of FXS, did not show impairments in LTP using a standard HFS paradigm (Godfraind et al., 1996; Li et al., 2002). However, using a low threshold stimulation protocol, LTP induction was reduced in multiple brain areas, including the hippocampus and somatosensory cortex (Larson et al., 2005; Zhao et al., 2005; Lauterborn et al., 2007). Additionally, mGluR-LTP was impaired in the basolateral amygdala and visual cortex (Wilson and Cox, 2007; Suvrathan et al., 2010). These findings provide evidence that different brain regions or synapse-specific deficits occur in mice with loss of FMRP. The most prominent change in synaptic plasticity has been observed with enhanced mGluR-LTD the cerebellum and in CA1 of the hippocampus (Huber et al., 2002; Koekkoek et al., 2005). An enhancement of LTD in the CA1 region has led to the theory that synaptic loss of FMRP produces increased signaling through mGluRs (Huber et al., 2002; Bear et al., 2004; Osterweil et al., 2010). Additionally, genetic reduction, or mGluR5 inhibition, rescues behavioral deficits in FMR1 mice (Dolen et al., 2007; Michalon et al., 2012). These rescue experiments therefore suggest the ability to reverse some of the impairments in FXS patients. The different impairments in synaptic function found across various brain regions in FMR1 mice provide evidence of a correlation between impaired synaptic plasticity and behavioral deficits in ASD. These studies also provide evidence that the FMR1 KO mouse may be a good model to explore the pathophysiology of ASD and investigate possible treatment strategies (Bhakar et al., 2012).

Synaptic plasticity has been studied in great detail in the UBE3A maternally deficient mice, the Angelman syndrome mouse model. In the CA1, HFS-LTP is reduced in transgenic mice and the reduction in CA1 LTP could be rescued with stronger stimulation protocols (Weeber et al., 2003). These findings indicate that Ube3a’s role in synaptic plasticity may be as a modulator of LTP, not necessarily required for LTP induction. In the same study, a reduction in calmodulin-dependent protein kinase II (CaMKII) activity was observed in transgenic mice, and genetic reduction of CaMKII’s inhibitory autophosphorylation rescued deficits in LTP and learning and memory tasks in transgenic animals (van Woerden et al., 2007). These findings provide evidence that altered CaMKII activity mediates the impairment in synaptic plasticity in UBE3A deficient mice; however, it remains to be elucidated how loss of UBE3A alters the activity of CaMKII.

Both of the TSC1 or TSC2 KO mice are embryonic lethal; however, mice harboring TSC1 or TSC2 heterozygous mutations display synaptic dysfunction and cognitive impairments (Kobayashi et al., 2001; von der Brelie et al., 2006; Ehninger et al., 2008). In TSC2 heterozygous rats, LFS-LTD was decreased and L-LTP was enhanced in the CA1 of the hippocampus; however, E-LTP was unaffected (von der Brelie et al., 2006; Ehninger et al., 2008), suggesting changes in protein synthesis pathways necessary for L-LTP expression. Additionally, in TSC2 heterozygous mice, mGluR-LTD was decreased but LFS-LTD was unchanged (Auerbach et al., 2011). A reduction in mGluR-LTD in TSC2 heterozygous animals opposes what is observed in the FMR1 KO mice; however, mGluR-LTD in both animals don’t show sensitivity to protein synthesis inhibitors (Auerbach et al., 2011). An additional study using mice with a specific deletion of TSC1 in cerebellar Purkinje cells displayed deficits in social interactions, an increase repetitive behavior as wells as defects in ultrasonic vocalizations. These animals however were not investigated for changes in synaptic plasticity in the cerebellum (Tsai P. T. et al., 2012). These findings suggest that impaired synaptic plasticity is a major pathology in TSC mouse models of autism, underlying the observed deficits in social interaction in both in TSC1 and TSC2 heterozygous animals.

In MECP2 KO mice, reductions in LTP and LFS-LTD in CA1 of the hippocampus have been reported. It’s interesting to note that younger mice (~3–5 weeks of age) show no impairments in synaptic plasticity, indicating that impaired synaptic plasticity correlates with the delayed deficits observed in human RTT patients (Chen et al., 2001; Asaka et al., 2006). A mouse model containing a truncation of MeCP2 (MeCP2308/Y) showed a reduction in LTP as well as paired-pulse stimulation, but no changes in mGLuR-LTD (Shahbazian et al., 2002; Weng et al., 2011). Interestingly, another study has shown that impaired LTP in CA1 of the hippocampus could be rescued by genetically reintroducing MeCP2 (Guy et al., 2007; Weng et al., 2011).

Shank proteins regulate levels and modulate the signaling of both metabotropic and ionotropic glutamate receptors at the synapse, and synaptic plasticity has been studied in multiple Shank3 mouse models (Tu et al., 1999; Uchino et al., 2006). Investigation of mEPSCs, paired pulse ratio (PPF and PPD), input/output curves and population spikes have indicated different abnormalities in synaptic transmission in hippocampal CA1 synapses of mice with different Shank3 mutations (Wang et al., 2011). Additionally, hippocampal LTP is reduced in the CA1 of the hippocampus in subsets of mice with Shank3 truncations, as well as LFS- LTD and mGLuR-LTD (Bozdagi et al., 2010; Bangash et al., 2011; Wang et al., 2011).

It has been hypothesized that disruptions in neuronal circuits involved with language and social behavior in subtypes of autism may be caused by unbalanced high levels of excitation, or disproportionately weak inhibition (Rubenstein and Merzenich, 2003; Gao and Penzes, 2015). With a more excitable cortex, the brain would have wide-ranging abnormalities in perception, memory, cognition and motor control, and would be highly susceptible to epilepsy (Rubenstein and Merzenich, 2003). In support of this theory, a decrease in glutamate decarboxylase 67 (GAD67) mRNA in autistic cerebellar Purkinje cells has been identified in human autism samples (Yip et al., 2007). Additionally, another study has shown a decrease in GABAergic inhibition in autism patient samples (Hussman, 2001), indicating that decreased GABAergic inhibition may disrupt the excitation/inhibition balance within neuronal networks in autism.

As a novel form of synaptic regulation opposing the Hebbian type plasticity, homeostatic synaptic plasticity (HSP) is a negative feedback response that serves to compensate for changes in network activity (Turrigiano et al., 1998; Hou et al., 2008, 2011, 2015; Yu and Goda, 2009; Pozo and Goda, 2010; Wang et al., 2012; Gilbert et al., 2016). In response to global decreases or increases in network activity from a homeostatic set-point, synaptic strengths are scaled up or down, respectively (Turrigiano et al., 1998). HSP may be important for maintaining network activity homeostasis and avoiding potential epileptogenic states and sleep may be necessary for synaptic homeostasis (Tononi and Cirelli, 2003; Kuhn et al., 2016). The studies in Drosophila for example, have shown that the size and number of synapses decrease after sleep and increase within a few hours of waking (Gilestro et al., 2009). Interestingly, Nlgns regulate levels of glutamatergic and GABAergic currents after sleep deprivation, indicating they play an important role in this sleep-dependent HSP (Huber et al., 2004; Gilestro et al., 2009; El Helou et al., 2013) and Nlgn defects could result in disturbances in sleep and circadian rhythms, a common disorder found in ASD patients (Bourgeron, 2007). FMRP, the RNA-binding protein involved with regulating dendritic protein translation, has also been shown to be required for increases in synaptic strength after neuronal activity blockade or application of retinoic acid in the mouse hippocampus, indicating that some symptoms of FXS may be due to impaired HSP (Soden and Chen, 2010). Release of brain-derived neurotrophic factor (BDNF) from postsynaptic neurons has previously been indentified to be required for a retrograde homeostatic up-regulation of presynaptic function. Increased BDNF levels in ASD patient blood samples has been observed as well as higher plasma levels of serotonin and N-acetylserotonin (NAS), a potent agonist of the BDNF receptor tyrosine receptor kinase B (TrkB; Jang et al., 2010; Halepoto et al., 2014; Kasarpalkar et al., 2014; Pagan et al., 2014). Excess levels of NAS could therefore increase TrkB-induced PI3K signaling, resulting in increased protein translation, similar to findings observed with mutations in components of the mTOR pathway.

ASD is diagnosed at the behavioral level with the presentation of its core phenotypes of impaired social interactions, restrictive interests and repetitive behaviors. Although the etiological factors of ASD are highly heterogeneous, recent research has strongly pointed to common cellular events that are impaired in ASD, including neurogenesis, morphogenesis, synapse maturation and synaptic plasticity. In this regard, loss-, or gain-, of-function mutations in single genes that are causative for ASD have given researchers unique opportunities to make important mechanistic insights. A common theme emerging within the field is that in the developing brain, alterations in dendritic growth, synapse formation and synaptic function result in neuronal network dysfunction, ultimately leading to complex social and cognitive dysfunction. Further elucidation into these pathways, as well as advancements in gene therapies and targeted drugs to modulate these processes, could provide exciting and promising new therapies for the treatment of ASDs.

JG and H-YM wrote and edited the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.