ASC is understood to be a whole brain disorder, whose neural correlates range from alterations in synapses and neural transmission to brain volume and structure. Alterations in functional connectivity have long been implicated in the pathophysiology of ASC since differences in functional connectivity have been found in an estimated 90% of studies (Lau et al., 2013; Chen et al., 2015; Mohammad-Rezazadeh et al., 2016) although the direction of this difference, implicated brain areas, brain state, and age are disputed (Müller et al., 2011; King et al., 2019). A dominant but challenged theory (Picci et al., 2016) is the “Cortical Underconnectivity Theory” which states that there is long range underconnectivity particularly between hemispheres and frontal-posterior brain areas in ASC individuals which may explain symptomatology (Just et al., 2012). In support of this theory is the finding that many ASC individuals have decreased Corpus Callosum (CC) volume (Frazier and Hardan, 2009), and that both adults and children with congenital agenesis of the CC (AgCC) show deficits in social and emotional intelligence and score highly on the Autism Quotient (Lombardo et al., 2012; Lau et al., 2013) without intellectual disability (Lábadi and Beke, 2017). Furthermore, fMRI studies of individuals with AgCC show decreased long range functional connectivity, suggesting that decreased long range connectivity between hemispheres is what contributes to ASC symptomology (Owen et al., 2013).

Apart from underconnectivity driven by the corpus collosum, underconnectivity could also be seen by activity in the nodes of the Default Mode Network (DMN), a set of nuclei that are preferentially activated, as visualized by rsfMRI (resting state magnetic resonance imaging) scanning, while individuals are not focused on the external environment (i.e., “mind wandering” or “resting state”). While DMN activation ceases once cognitive effortful tasks commence (Anderson et al., 2011; Li et al., 2014; Padmanabhan et al., 2017), mind wandering involves internally generated thoughts about others, one’s self and remembering past or envisioning future events, as elucidated by designed self-report questionnaires and interviews (Andrews-Hanna et al., 2013; Smallwood and Schooler, 2015). The DMN identified in this manner consists of the medial prefrontal cortex (mPFC), the posterior cingulate cortex, precuneus, medial temporal lobe (which includes the hippocampus), and inferior parietal cortices (Li et al., 2014; Alves et al., 2019) which together incorporate memories (hippocampus) and emotions (mPFC) in order to create “mind wandering” (Poerio et al., 2017). Hence, adults with bilateral ventromedial prefrontal cortex (vmPFC) lesions show a reduction in self-reported mind wandering (Bertossi and Ciaramelli, 2016).

Decreased connectivity is often detected by fMRI in ASC individuals within the DMN, with underconnectivity having been found positively correlated to ASC social deficits, increased repetitive stereotyped behaviors measured by Autism Diagnostic Observational Score (ADOS; Yerys et al., 2015) as well the Autism Diagnostic Interview-Revised (ADI-R; Chen et al., 2021b). Though this data is necessarily correlative in nature, long range underconnectivity in DMN areas could be a neuroanatomical signature for ASC individuals who could then be tested for social deficits (Nair et al., 2020) using the Theory of Mind (ToM) tasks in which ASC individuals show marked deficits (Frith and Happé, 1994). However, other studies have shown short range, long range and overall brain connectivity to be functionally overconnected in adults (Belmonte, 2004; Kleinhans et al., 2016), though long range connections to and from the frontal cortex appear to be consistently under-connected (Courchesne and Pierce, 2005), with strength of connectivity between the PFC and right parietal cortex negatively correlated with autism symptom severity (Redcay et al., 2013). However, long range hyperconnectivity between the thalamus and auditory and somatosensory cortices as well as parietal regions has also been found in fMRI studies of adults with ASC (Tomasi and Volkow, 2019).

Varying long range connectivity in studies could occur due to the polygenetic nature of ASC with different genetic rodent models showing different functional connectivity patterns. Therefore, it may be possible to use varying patterns of connectivity as an endophenotype for identifying different forms of ASC (Zerbi et al., 2021). For example, homozygous knockout male mouse models of Fragile X Syndrome (FXS) show long range cortico-striatal connections to be functionally underconnected (Zerbi et al., 2018), whereas heterozygous knockout male TSC models mouse models show the same connections to be functionally hyperconnected (Pagani et al., 2021). This long range hyperconnectivity found within the TSC mouse model was also found to be associated with increased repetitive behaviors in the mice. Increased repetitive behaviors and increased long range hyperconnectivity are both reversed by the administration of the mTOR inhibitor rapamycin, which functionally reverses the effects of the loss of one of the two Tsc2 genes (Pagani et al., 2021).

Conversely, individuals with depression present increased rumination and self-referential mind wandering (Nejad et al., 2013; Nayda and Takarangi, 2021) that is correlated with hyperconnectivity between the long distance DMN nodes mPFC and the posterior cingulate cortex, suggesting that the DMN nuclei must maintain optimal connectivity for healthy cognition and perception of self (Wise et al., 2017). In humans the use of ketamine, a compound that is primarily an NMDA receptor antagonist (Sleigh et al., 2014) has been shown to alleviate symptoms of depression, as well as reduce functional connectivity within the DMN of healthy individuals (Scheidegger et al., 2012) and those with Major Depressive Disorder (MDD; Evans et al., 2018), suggesting that many behaviors are strongly correlated with functional connectivity. These studies show that functional connectivity in the DMN may be a marker of behaviors that denote increased or decreased self-reference.

What underlies differences in DMN connectivity in ASC individuals? Within discrete brain areas, this may be attributed to an Excitation Inhibition (E/I) imbalance and reduction in signal-to-noise ratio in neural circuitry that is implicated in ASC (Rubenstein and Merzenich, 2003; Testa-Silva et al., 2012). Indeed, the high incidence of epilepsy within the ASC population (Viscidi et al., 2013) as well a postmortem study showing upregulation of excitatory AMPA receptors (Purcell et al., 2001), may indicate that there is an increased excitation within the brain of ASC individuals. How increased excitation within the brain of ASC individuals causes local or long distance hyperconnectivity or hypoconnectivity is not fully understood though changing E/I ratios in local circuits to understand this question is now underway in a number of rodent models of ASC (See Section “Molecular Mechanisms That Link Dysfunctions in Neuromorphology to Behavior”). Interestingly, hyperconnectivity between various modes of the DMN, including that detected in a Tsc2^+/− mouse model of ASC using rsFMRI is linked to an excess of mTOR signaling and increased spine density in the insular cortex. Using in silico modeling, these increases in synapse density were linked to hyperconnectivity and use of a mTORC1 inhibitor, rapamycin, reversed both the increases in spine density and the hyperconnectivity in this model (Pagani et al., 2021). This strongly suggests that connectivity in the DMN is positively correlated to spine density which is in turn driven by mTORC1.

Clearly, DMN connectivity and the underlying E/I ratio could be dependent on neuronal architecture including dendritic spine dynamics reflected by spine density and spine morphology (Gao and Penzes, 2015). Dendritic spines are typically around 0.5–2 microns long, dendritic protrusions that receive signals from axonal boutons of other neurons to form synapses allowing neurons to transmit signals to one another (Horner and Arbuthnott, 1991). Human postmortem studies reveal that synapse formation increases throughout childhood from birth in both ASC and Typically Developing (TD) children, but is slightly elevated in ASC (Penzes et al., 2011). Synapses are then pruned from adolescence into adulthood, with pruning being more efficient in TD over ASC subjects, leaving ASC individuals with a greater number of synapses. In adulthood, this is followed by an equilibrium in synapse formation and elimination to maintain a stable number of synapses in TD individuals (Rakic et al., 1986; Masliah et al., 1993; Peters et al., 1998; Penzes et al., 2011); there is increased synapse maintenance during adulthood in ASC, often leading to a greater number of spines and synapses compared to TD individuals (Penzes et al., 2011). The typical development of cortical dendritic spines in rodents follows a similar pattern to humans, with an initial increase in dendritic spine density during the first two postnatal weeks, followed by an increase in spine pruning over spine formation during adolescence, with spine density then stabilizing in adulthood (Bhatt et al., 2009). Apart from spine number, the shape of spines is also important (Table 1). New immature spines have a small surface area and postsynaptic density (PSD), with decreased number of receptors; this could reflect an increased “neuroplasticity” of neural circuits, with a greater potential for strengthening, thus making them “learning” spines (Bourne and Harris, 2007). It should be noted that the classification of filipodia spines as being a defined and individual spine type or just the precursor to dendritic spines is disputed (Takahashi et al., 2003; Kanjhan et al., 2016); this is due to the dynamic nature of filipodia-type spines which give rise to all other spine types. However, since many studies investigating dendritic spine dynamics in rodent models of ASC report filipodia-type spines as part of the immature dendritic spine classification, and report changes in the quantity of filopodia type spines in various rodent ASC models, we have also decided to report filipodia type spines as a distinct spine classification within the larger category of immature spines (Dictenberg et al., 2008; Williams et al., 2008; Dunaevsky et al., 2014; Pyronneau et al., 2017; Sceniak et al., 2019; Skelton et al., 2020). Wider more mature “memory” spines have greater surface area and PSD for receptors and form very strong synapses with very little range for the synaptic strengthening/adaptation that is associated with Long Term Potentiation (LTP), on increased stimulation (Cooke and Bliss, 2006). Spine shape and consequently maturity are characterized by their length and width measurement ratios (Harris and Kater, 1994), with immature spines, i.e., filopodia, thin, long thin spines tending to be longer and thinner in size and mature spines, i.e., mushroom and stubby shorter and wider (Table 1). An increase in mushroom type mature spines and increase in overall spine width in the primate prefrontal cortex is associated with cognitive decline during aging in rhesus macaques (Dumitriu et al., 2010), suggesting inflexibility and loss of neural plasticity (Bourne and Harris, 2007, 2008).

A recent study using living humans showed that synaptic density (which may be indicative of spine density), as measured by PET scanning of synaptic glycoprotein 2A, is negatively correlated with symptom severity in adults with MDD, within the dorsolateral prefrontal cortex (Holmes et al., 2019). Such local hypoconnectivity is inversely related while long range hyperconnectivity between the dorsolateral prefrontal cortex and the posterior cingulate cortex is positively correlated with depression symptoms of clinically depressed individuals, suggesting that the underlying measurements at the synapse, i.e., spine density/morphology may have predictive value. In support of this idea, multiple rat models of depression also show a decrease in dendritic arbor within the hippocampal CA3 (Watanabe et al., 1992) and mPFC pyramidal cells (Goldwater et al., 2009), both key nodes involved in the DMN. Furthermore, ketamine inhibited dendritic spine loss in chronically stressed rats that displayed symptoms of depression via protection of mushroom shaped spines, and through the induction of spine formation (Ng et al., 2018).

Human postmortem studies, using Golgi-Cox staining, have shown an increase in spine density in ASC individuals compared to controls, independent of sex, within the amygdala, cortex, and temporal lobes (Hutsler and Zhang, 2010; Tang et al., 2014; Weir et al., 2018; Table 2). This increased spine density in ASC individuals (Weir et al., 2018) is correlated with increased head circumference/macrocephaly, as well as enlargement of specific brain areas including the amygdala and hippocampus (Schumann et al., 2004; Tilot et al., 2015; Weir et al., 2018). Only one human postmortem study assessed spine morphology via Golgi-cox staining in ASC vs. TD controls and found that ASC individuals had more “compact” spines with decreased spine length within the cortex, which indicates a more mature spine morphology (Hutsler and Zhang, 2010; Table 2). This along with the increased spine density found in ASC individuals may reflect an increase in local connectivity linking spine density to functional connectivity. One thing to note is that increased head size (macrocephaly) is a common finding in children and adults with ASC; toddlers with ASC show accelerated brain growth compared to TD controls to possibly result in an enlarged amygdala and hippocampus during adolescence (Groen et al., 2010) with the degree of amygdala enlargement being a predictor for ASC symptom severity (Schumann et al., 2004). A mouse ASC genetic model of Phosphatase and tensin homolog (Pten) haploinsufficiency also reflects the macrocephaly found within ASC individuals (Huang et al., 2016). Furthermore this model shows increased brain enlargement and macrocephaly that is correlated with hyperconnectivity and dendritic arborization (Huang et al., 2016), suggesting that ASC individuals with macrocephaly may also have similar hyperconnectivity with increased spine densities. A conditional knockout of Pten in a subset of auditory cortical neurons shows enhanced local and global hyperconnectivity with the strength of synaptic inputs from the thalamus and auditory neurons increased; this was decreased by the mTOR inhibitor rapamycin (Xiong et al., 2012). Together, these studies suggest that functional connectivity is correlated with both spine density and mTOR signaling and both these neural correlates may result in macrocephaly.

If mTORC1 signaling is a critical nexus for upstream regulators that are deleted or mutated in autism and if spine dynamics are an underlying mechanism for autism, then it is plausible that loss or mutation of upstream regulators may affect spinogenesis and/or spine pruning via mTORC1 signaling, just as it does social behavior (Section “Rapamycin, an Inhibitor of mTORC1, Reverses Autistic-Like Behaviors”). In ASC individuals, postmortem studies show a decrease of LC-III, an autophagosome marker, suggesting decreased pruning. Decrease of this marker is correlated with increased spine density and persistent PSD-95 marker expression in the temporal lobe throughout childhood and adolescence (Tang et al., 2014). Supporting this, a neuronal deletion of atg7, a E1 ligase required for phagosome formation in mice results in social interaction deficits and spine density on basal dendrites that stays elevated on cortical pyramidal neurons into adulthood. A second murine model that shows the significance of spine pruning is the Tsc2^+/− haploinsufficient mouse where reduction of this negative regulator of mTORC1 leads to increases in spine density in cortical projection neurons (Tang et al., 2014). Though rapamycin rescues both behavioral deficits and aberrant spine density in the Tsc2^+/− mouse, it does not do so in the Atg7 conditional knockout (cKO) mouse or in the double mutant Tsc2^+/−, Atg7^cKO mouse, demonstrating an absolute need for spine pruning in order to ameliorate social behavior deficits (Tang et al., 2014). A second upstream negative regulator of mTORC1 is the Fragile X mental retardation protein (FMRP; Sharma et al., 2010; Hoeffer et al., 2012; Darnell and Klann, 2013; Casingal et al., 2020), the Fmr1 KO shows increased spine density in the hippocampus which in turn is correlated with higher long term depression (LTD) and a lack of discrimination in the novel object recognition task. These phenotypes are ameliorated with shRNA to Raptor, an unique component of the mTORC1 complex, suggesting that overactivation of mTORC1 specifically leads to the cognitive deficit (Yan et al., 2018). This rectifying effect of decreasing Raptor levels could be abrogated by decreasing ATG7, a protein required for phagosome formation, suggesting that Raptor rescued spine and cognitive deficits via the activation of autophagy in hippocampal neurons (Yan et al., 2018). Consistent with the idea that rectifying mTORC1 activity is central to rescuing spine density and social behavior, ketamine, a mTORC1 activator because it also secondarily increases AMPA signaling, rescued the lack of fear memory and decreased spine density in the prefrontal cortex seen in a heterozygote mouse deficient in the RELN (reelin) protein. Since RELN is a positive activator of mTORC1, lack of this protein leads to lower activation of this pathway which can be rescued by activation with ketamine. Conversely, rapamycin, an inhibitor of mTORC1 can reverse the ketamine-mediated rescue of spine density and fear memory in the Reln^+/− mouse, suggesting that both neuromorphology and cognitive and social behaviors are highly sensitive to mTORC1 levels (Iafrati et al., 2013; Section “Rapamycin, an Inhibitor of mTORC1, Reverses Autistic-Like Behaviors” below).

If mTORC1 is a critical nexus for ASC, we would also expect that alteration of this pathway results in behaviors that denote autism in rodent models. To give an example, haploinsufficiency of Tsc2 or Tsc1 in the cerebellum using a conditional knockout (KO) mouse model is sufficient for the manifestation of social preference in the 3-chamber preference test and repetitive behavior in the marble burying task and self-grooming with all behavioral phenotypes ameliorated with rapamycin treatment (Tsai et al., 2012; Reith et al., 2013). Similarly, a selective deficit in social interaction but not in locomotor or food exploration activity was noted in both Tsc1 and Tsc2 heterozygote mutant mice. This deficit is reversed in both genotypes by rapamycin (Sato et al., 2012). Tsc2^+/− mice also show decreased memory in the Morris Water Maze and no discrimination between novel and training contexts in the contextual fear conditioning task, behaviors that could possibly be due to unstable late-phase LTP at the Schaeffer collateral synapse in the hippocampus; these were also reversed by rapamycin. This is a specific effect since early phase LTP, basal synaptic transmission and paired pulse facilitation are unaffected (Ehninger et al., 2008). Rapamycin reversal of specific behavioral and synaptic phenotypes demonstrates that mTORC1 signaling is specifically critical in driving autism-like behaviors in rodents.

The consequence of overactivation of the mTORC1 pathway would be an increase in translation and hence we would expect to see similar behavioral deficits in mice that overexpress or show greater activity of the downstream targets of mTORC1. Mice that are deleted for the repressor e4B-BP1, mimic mTORC1 overactivation since they have increased eIF4E activity and show increased excitation-inhibition (E/I) ratio, suggesting altered synaptic properties in the CA1 of the hippocampus. They also show decreased vocalization and socialization and no preference for conspecifics in the 3-chamber test as well as increased anxiety and repetitive behaviors such as self-grooming (Gkogkas et al., 2012). Interestingly, pharmacological inhibition of eIF4E normalized both social and repetitive behaviors and the E/I ratio. Similarly, a transgenic mouse that overexpressed eIF4E mimicking ASC individuals with a SNP in the promoter of the eIF4E gene that increases the level of this protein, also showed increased repetitive behavior and anxiety, decreased contact with conspecifics and higher mEPSCs from pyramidal neurons in the prefrontal cortex. This could be due to an increased spine density through an increase of smaller spines (Santini et al., 2012). Again, intracerebroventricular infusion of the eIF4E inhibitor, 4EGI-1, improved social behaviors and decreased anxiety (Santini et al., 2012). In Fmr1 (Fragile-X mental retardation 1) KO mice, where mTORC1 is overactive because of loss of inhibition of PI3K, S6K is also overactivated; deletion of S6K increased dendritic spine maturation with concomitant improvement in social interaction behaviors and mGluR-mediated long term depression (LTD; Bhattacharya et al., 2012). Taken together, all these studies suggest that overactivation of mTORC1 and S6K signaling is sufficient for the manifestation of autistic-type behaviors.

Several reviews stress the role of mTOR signaling in autism (Magdalon et al., 2017; Winden et al., 2018) while others explore the role of spine dynamics in autism, focusing on a few genetic models (Copf, 2016; Lin et al., 2016; Martínez-Cerdeño, 2017). In these sections, we specifically focus on the effects of upstream positive and negative regulators of mTORC1 on spine density and/or spine morphology in rodent models and possible correlations with synaptic physiology [Section “Many Upstream Regulators of mTORC1 Are Expressed at the Dendrite or at the Post-Synaptic Density (PSD)”] and social behaviors (Section “Alterations of mTORC1 Activity by Alterations in the Expression of Upstream Regulators Result in Autistic Behaviors”).

CDKL5 heterozygotes show reduced learning in the Barnes Maze and repetitive jumping behavior (Trazzi et al., 2018) while heterozygote reeler mice show decreased freezing to tone in renewal trials during fear conditioning (Iafrati et al., 2013). Similarly, the Dip2A KO shows lower social exploration time of a conspecific, increased self-grooming in both novel and home cage environments, decreased social novelty and increased marble burying indicative of higher anxiety as well as social recognition and interaction deficits (Ma et al., 2019).

Notably, rodent ASC models utilizing SHANK3 mutations have good face validity, reflecting both repetitive and altered social behavior phenotypes, including reliably increased self-grooming behavior, lower levels of social interaction and ultrasonic vocalization and increased anxiety in a domain-deletion independent manner (Wang et al., 2011; Jaramillo et al., 2016; Mei et al., 2016; Zhou et al., 2016). Most models also show decreased social recognition memory (Peça et al., 2011; Schmeisser et al., 2012; Kouser et al., 2013; Mei et al., 2016) with males scoring worse in this task than females (Wang et al., 2011). Furthermore, ankyrin-repeat and proline rich domain mutants show hypoactivity in the open field (Wang et al., 2011; Zhou et al., 2016). For the global regulators, though MeCP2 overexpression in mice recapitulates Rett syndrome, it is harder to see autistic-type behaviors. However, a recent study using MeCP2 overexpression in cynomolgus monkeys showed that these monkeys display repetitive circling in their cages and lower time with familiar or novel conspecifics, indicative of ASC (Liu et al., 2016). Similarly, UBE3A overexpression results in repetitive behaviors and social interaction deficits (Smith et al., 2011). In general, for positive regulators of mTORC1, tests of social interaction/recognition/motivation tend to report a deficit compared to tests of spatial memory where reversal of the memory is typically measured (Monteiro and Feng, 2017).

Pten KO mice show lower interaction with conspecific mice or novel mice and hyperactivity in the open field (Kwon et al., 2006). Unusually, in GRIN2B⁺/C456Y mice that mimic a human ASC mutation, NMDAR-dependent long-term depression (LTD) is decreased and though there is normal social interaction, there is an anxiolytic effect as measured by the elevated plus maze; the molecular mechanisms underlying this phenotype remain obscure (Shin et al., 2020). A similar anxiolytic effect is seen in a comprehensive panel of behavioral tests of the haploinsufficient SynGAP^+/− mice which also show no deficit in social interaction but hyperactivity and cognitive deficits in working and reference memory (Ma et al., 2019) and in a dentate gyrus-specific context discrimination task (Clement et al., 2012). A similar deficit in spatial memory is seen in two Shank2 PDZ-deletion models, combined with increased self-grooming behavior and anxiety along with decreased social interaction or novelty (Schmeisser et al., 2012; Won et al., 2012). Impaired social interaction and novel object recognition (NOR) along with hyperactivity is also seen in Fmr1 KO mice (Qin et al., 2011; Bhattacharya et al., 2012). For negative regulators of mTORC1, it is notable that spatial, hippocampal-based tasks show deficit and hyperactivity is seen in several models.

One of the reasons for E/I imbalance could be altered spine density/morphology or synaptic proteins, such as glutamate receptors at the PSD that are often altered in rodent models of autism that have alterations of the upstream regulator proteins of mTORC1.

For positive upstream regulators of mTORC1: Hyperexcitability and a higher E/I ratio in the CA1 is due to higher localization of GluN2B and SAP102 in CdKl5^−/y mice, as revealed by subcellular fractionation of the hippocampal PSD (Okuda et al., 2017). In Shank3 KO mouse striatum, the disrupted interaction between Homer1b/c-mGluR5 also results in increased mGluR5 at the PSD. Antagonists of the mGluR5 receptor such as MPEP could reverse the increased self-grooming in these mice, suggesting that receptor density at the PSD is critical for behavior (Wang et al., 2016). In mice carrying SHANK3 deletion of different domains, several receptors such as NR1, NR2A, GluR2, and GluA2 are decreased (Monteiro and Feng, 2017 and references therein). Conversely, the decreased mEPSC amplitude in the Dip2a KO is due to a smaller number of NMDA subunits at the PSD (Ma et al., 2019). In MeCP2 overexpressing mice, despite an increase in both mGluR and NMDA receptors at the synapse in the somatosensory cortex (Chahrour et al., 2008; Jiang et al., 2013), there is a reduction in the EPSCs because NMDA/AMPA currents correlate to reduced GluN1 expression; this is not due to reduction in GABAergic interneurons (Sceniak et al., 2016) in the prefrontal cortex. Rather, this suggests, as borne out by brain activity c-Fos mapping, that reduced functional excitation rather than reduced inhibition drives the decrease in E/I ratio in these animals since this is reversed by ketamine, showing the importance of post-synaptic glutamate receptors (Kron et al., 2012). This could demonstrate that alteration in E/I ratio is driven by different mechanisms in different brain regions since reduction in the MeCP2 overexpressing somatosensory cortex is due to a reduction in GABAergic parvalbumin neurons in that region (Gogolla et al., 2009) but this is not true in the prefrontal cortex (Sceniak et al., 2016).

For negative upstream regulators of mTORC1: In Shank2 PDZ-deletion mice, the predominant pathway driving the atypical (for a negative regulator) decrease in spine density and social interaction deficit appears to be NMDA hypofunction since cycloserine, a NMDA agonist, can restore functionality (Won et al., 2012). However, AMPA receptors are also important; for example, SynGAP dispersal from the synapse, results in loss of Ras-ERK inhibition and leads to subsequent mTOR-mediated increased AMPA receptor insertion at the PSD, a prerequisite for LTP. Therefore, in SynGAP-deficient mice, though the proportion of mature spines and basal synaptic transmission are enhanced, no further enhancement is possible since increased AMPA receptor insertion in the basal state occludes LTP (Gamache et al., 2020). A third negative regulator is Tsc1/2; a pan-neuronal KO of Tsc1 shows reduced frequency of GABAergic mIPSC, resulting in increased excitability in the cortex and an increased E/I ratio; however, a PV-specific Tsc1 KO does not show altered E/I balance, suggesting that the local hyperexcitability is not due to loss of TSC signaling in PV interneurons (Zhao and Yoshii, 2019). Similarly in the hippocampus, though homeostatic plasticity via decreased synaptic AMPA receptors is seen, in experiments where homeostatic plasticity is not initiated, the increased E/I ratio is primarily due to impaired mGluR5-LTD at hippocampal synapses that may be because of reduced Arc and other proteins in the Tsc1 KO that are required to stabilize LTD (Bateup et al., 2013). In support of this idea, the Tsc1 mutant phenotype of reduced LTD and context discrimination can be rescued by an Fmr1^−/y mutant that shows the opposite, i.e., exaggerated mGluR5-LTD, suggesting that levels of PSD proteins maintained by these mTORC1 regulators underlie the deficit despite the fact that loss of Tsc1 and FRMP proteins both lead to overactivation of mTORC1 (Auerbach et al., 2011). In the amygdala of the Fmr1 KO mouse, principal cells show intrinsic hyperexcitability and increased synaptic plasticity (Svalina et al., 2021), suggesting that the enhancement of PSD-95 and Arc levels seen in the PSD density of spines of hippocampal neurons (Yan et al., 2018) may also be true of other limbic regions to result in greater excitability.

Though the relationship between spine density and synaptic physiology appears to be complex, looking at spine morphology paints a clearer picture (Tables s 3 and 4), with an increase in immature spines in many models. A number of knockout rodent models of negative regulators such as FMRP, GRIN2B, and PTEN show an increase in the proportion of thin or filopodial spines vs. mushroom spines. Similarly, in the Dip2a KO, there is an increase in spine density but a decrease in mushroom shaped spines and a flattening of the PSD with a decrease of NMDA receptors in a smaller spine volume, possibly leading to a lower E/I ratio (Ma et al., 2019).

Some of this could also be due to spine instability. This is seen in Fmr1 KO mice where there is a developmental delay in the transition from immature to mature spines in Layer2/3 pyramidal neurons of the cortex; thin spines are unusually unstable. Normally, these spines search for presynaptic partners in a lengthening process facilitated by mGluR5; however, mGluR5 agonists are unable to lengthen spines and achieve this in Fmr1 KO mice (Ginsberg et al., 2012) suggesting that stabilization via contact with pre-synaptic partners is deficient when FMRP is absent. Interestingly, spines are stabilized in an ERK-dependent manner solely in clusters whose numbers are increased in mice overexpressing MeCP2 resulting in a higher proportion of mushroom shaped spines; isolated spines are formed and stabilized at the same rate as the wildtype mouse. Such stabilization of spines at a young age is correlated with increased stability on the rotarod, a marker of perseverative behavior in these mice (Ash et al., 2021a, b) though it is unclear if this is cause of effect. For example, repetitive task learning that is cortex-dependent induced spines preferentially in clusters; these clustered spines showed greater persistence past learning than non-clustered spines (Fu et al., 2012) and hence the functional interpretation of spine clustering in ASC models is difficult to ascertain.

Spine morphology also has functional consequences; shRNA targeting Pten results in increased spine density in the basolateral amygdala (BLA) of the mouse with an increased mEPSC amplitude and frequency in the BLA. This is most likely due to a shift in spine morphology with an increased proportion of mushroom spines but decreased filopodia (Haws et al., 2014). Consistent with this finding, though granule cells of neonatal Pten KO mice show hypertrophy and therefore might be expected to be less excitable, increased density of mushroom shaped spines concomitant with increased dendrites and protrusions drives the hyperexcitability observed in this ASC model (Williams et al., 2015). This increase in spines is larger when Pten KO granule cells are sparsely surrounded by wildtype cells when compared to when they are densely surrounded by other Pten KO cells (Skelton et al., 2019) though the mechanism underlying this and its significance is unknown. It could be that secondary effects on network activity amongst surrounding cells and subsequent tertiary feedback effects by these neighboring cells on the Pten KO neuron drive the ASC phenotype (LaSarge and Danzer, 2014).

Corticostriatal circuitry which underlies motivated sensorimotor by interacting with the limbic system is the focus of several neurodevelopmental disorders, including autism (Shepherd, 2013), with several alterations to this circuit. In ASC individuals, intrinsic functional connectivity studies done using fMRI imaging shows that those with highly repetitive behaviors show reduced frontoparietal and motor connectivity with the striatum but increased connectivity with the limbic system compared to those with low repetitive behaviors, underscoring the importance of the corticostriatal circuit (Abbott et al., 2018). These differences in neuroanatomy particularly for the striatum are underscored by a longitudinal structural imaging study which revealed a faster growth rate of the caudate nucleus during late childhood to early puberty in ASC individuals that was strongly correlated with repetitive behaviors seen in these individuals that developed during the preschool period (Langen et al., 2014). These alterations in mice are linked to spines; conditional dorsal pallium-specific KOs of Met, a tyrosine kinase that regulates spinogenesis, demonstrates local hyperconnectivity in this circuit with stronger afferents from corticostriatal neurons in layer 2/3 to layer 5 projection neurons in the anterior frontal cortex in heterozygotes and KOs compared to WT controls (Qiu et al., 2011).

Developmental studies of Shank3, a positive regulator of mTORC1, show that the striatal excitability is higher due to increased cortical glutamatergic efferents to the striatum (Peixoto et al., 2016) at early postnatal stages but there is reduced corticostriatal connectivity and EPSCs as adults (Mei et al., 2016; Wang et al., 2016). In addition, disruption of the dopaminergic receptor D1R-SYNGAP complex in the prefrontal cortex during development results in decreased migration of GABAergic interneurons with concomitant increased cortical excitability. This cortical excitability that develops as a result of disruption of the function of SYNGAP and D1R is instrumental in the social deficits seen in this model during adulthood (Lai et al., 2021). In Tsc2^+/− mice, resting state fMRI reveals functional hyperconnectivity within the corticostriatal circuit due to increased synaptic coupling possibly as a result of increased spine density; this was strongly correlated to repetitive behaviors. All parameters, i.e., connectivity, spine density, synaptic coupling, resting state default mode network signatures, and social behaviors are rescued to normal levels with rapamycin, suggesting that mTORC1 overactivation in this circuit is crucial to spine dynamics that underlie social behaviors. This hyperconnectivity signature is valid cross-species—children with ASC showed similar fronto-insular-striatal hyperconnectivity and mTORC1 interacting genes are enriched in the ASC cortical transcriptome, suggesting that this Tsc-mTORC1 pathway is a critical driver of functional hyperconnectivity (Pagani et al., 2021). Cross-species connectivity is also shown with mutations of NF1, a negative regulator of mTOR similar to Tsc1/2, though Nf1 KO mice show decreased spine density. Children with NF1 mutations with ASC show reduced functional connectivity between the striatum and the frontoparietal network and increased striatal functional connectivity with the limbic system. Similar to the human scenario, Nf1^+/− mice also show disrupted corticostriatal connectivity, as revealed by resting state imaging (Shofty et al., 2019). This shows that both positive, e.g., SHANK and negative regulators. e.g., TSC of mTORC1 typically show spine density alterations in the corticostriatal circuit that are commensurate with local connectivity signatures and may underlie the repetitive behaviors seen in both humans and rodent models.