An estimated 1 in 100 children globally have autism spectrum disorder (ASD), and CDC estimates indicate even greater prevalence in America, where roughly 1 in 36 children is diagnosed with ASD (Zeidan et al., 2022; Maenner et al., 2023; Talantseva et al., 2023). As ASD is highly heritable (Sandin et al., 2017), much work has been done in recent years to identify genes that confer risk of developing ASD. Increased accessibility of DNA sequencing has allowed for the identification of hundreds of ASD risk genes, which range widely in the types of proteins for which they code (Satterstrom et al., 2020). Despite this molecular heterogeneity, ASD is still diagnosed through identification of behaviors that fall into two primary domains: deficits in social communication and interaction, and the presence of restricted, repetitive patterns of behavior (RRBs) (APA, 2022).

In individuals with ASD, RRBs can span a range of “lower order” and “higher level” behaviors. “Lower order” motor presentations may include self-stimulation or self-injury like head banging, hand flapping, twirling, lining up or manipulating objects, or repeatedly pressing buttons. “Higher level” repetitive behaviors include rituals, perseverative interests and insistence on sameness in a variety of situations (Caldwell-Harris, 2021). In addition to the repetitive behaviors recognized as core ASD symptoms, other motor presentations can include changes to gross motor skills such as balance, gait and posture, as well as alterations in fine motor skills and motor skill learning (Chukoskie et al., 2013). In studies of balance, individuals with ASD exhibit reduced postural control, in particular when somatosensory or visual challenges are introduced. This could occur when a subject is instructed to close their eyes, stand on one leg, or balance on a swaying platform, for example (Minshew et al., 2004; Travers et al., 2013). Atypical gait, which several studies have reported in individuals with ASD, may occur as a result of difficulties with balance and posture (Chukoskie et al., 2013). While specific changes in gait parameters are heterogenous across studies, a lack of smoothness, irregular trunk movements, and shorter stride length are commonly identified in individuals with ASD (Vernazza-Martin et al., 2005; Weiss et al., 2013). Foundational motor movements such as reaching and grasping have also been shown to be altered in children with ASD (Haswell et al., 2009; David et al., 2012), which may underlie some of the deficits seen in executing gross motor skills like throwing and catching, as well as fine motor skills like buttoning, manipulating small objects, and handwriting (Green et al., 2009; Chukoskie et al., 2013; Battah et al., 2023). Notably, handwriting has been reported to be significantly altered in those with ASD since the earliest descriptions of the disorder (Asperger, 1991). Although the early presence of motor symptoms is highly predictive of later overall ASD symptom severity, this remains an understudied and undertreated symptom domain (Troyb et al., 2016; Zampella et al., 2021). The use of common behavioral assays in tractable animal models of ASD can greatly assist in the identification of circuits that may underlie motor changes in autism.

Increasingly, the basal ganglia, and in particular the striatum, has been implicated in the manifestation of repetitive behaviors in ASD, because of the role of these circuits in motor learning, action selection, and habit formation (Fuccillo, 2016). Indeed, both structural and functional imaging studies identify aberrant striatal morphology and connectivity in individuals with ASD, in some cases strongly correlating with the presentation of repetitive behaviors (Hollander et al., 2005; Estes et al., 2011; Dichter, 2012). Magnetic resonance imaging (MRI) studies in mice support these findings, where a diverse range of genetic ASD mouse models exhibit altered striatal morphology and connectivity (Portmann et al., 2014; Ellegood et al., 2015; Lai et al., 2016; Wang et al., 2016). In this review we will discuss the relationship between striatal function and motor performance in mouse models of ASD, which has been illuminated through the use of a common behavioral assay of motor coordination and learning, the accelerating rotarod.

An increase in the identification of genes implicated in ASD risk paired with the genetic accessibility of animal models has allowed for the development of many genetic mouse models of ASD (Bey and Jiang, 2014). Targeting mutations in these mouse models to risk genes that have been identified in individuals with ASD provides construct validity (where the perturbation used to generate the disease model recapitulates the known etiology of the disease in people) (Nestler and Hyman, 2010). Face validity of these models (where the model displays key clinical manifestations of the disease) is more challenging to achieve given the heterogeneity and variability of ASD presentations in people. That said, a range of assays have been developed with the goal of measuring mouse behaviors analogous to those comprising the symptom domains of ASD (Bey and Jiang, 2014).

For the RRB domain of ASD, mouse behavioral assays primarily fit into the “lower order” and “higher level” domain distinctions detailed above. The former is typically measured with the open-field assay, allowing for detection of changes in general locomotor features such as speed and distance traveled, as well as the presence of motor stereotypies such as repetitive grooming, rearing, circling or jumping (Gandhi and Lee, 2020). Other assays like the marble burying test and the hole board take advantage of natural exploratory mouse behaviors like digging and head poking to detect increased repetition of these spontaneous behaviors (Bey and Jiang, 2014). More complex, “higher level” aspects of RRBs can also be assessed in mice, measuring resistance to change, cognitive inflexibility and perseveration in a range of reversal learning, set-shifting and response extinction tasks (Gandhi and Lee, 2020). The changes to gross motor function and coordination that appear to coincide with the repetitive behavior domain in individuals with ASD can also be assessed in mice using balance beams and commercially available systems for measuring and analyzing gait parameters (e.g., DigiGait, Neurocube) (Simmons et al., 2021). The recent development of deep-learning-based platforms such as DeepLabCut and MoSeq allows for unsupervised, data-driven detection and analysis of mouse behavioral parameters (Mathis et al., 2018; Wiltschko et al., 2020).

One behavioral assay commonly utilized in mouse models, the accelerating rotarod task, can be used both as a measure of gross motor coordination, as well as motor skill learning. Below we will outline the structure and parameters of the rotarod task, the way that learning occurs over the course of trials, and the brain regions and circuits implicated in rotarod performance.

First described in the 1950’s (Dunham and Miya, 1957), the accelerating rotarod task has historically been used as a measure of motor coordination and function in animal models of disease (Hamm et al., 1994; Heng et al., 2008; Lubrich et al., 2022) (Figure 1). However, performance on this test can also be used as a measure of motor skill learning. In the task, mice are trained to walk on a rotating rod as it increases in speed at a constant rate. Protocols utilized in the task vary, but typically the rod increases from 5 to 40 revolutions per minute over the course of 5 min. The latency to fall, or rotate backwards off the rod, is used to determine the terminal velocity in each trial, with increases in this measure indicating better performance. Over several trials, animals exhibit improvement both within a given training day, and over the course of training sessions (Luft and Buitrago, 2005). In this way, initial performance in the task can be isolated as a measure of basic motor coordination, with differences between mouse models at this early stage indicating gross motor deficits or altered baseline motor function. If initial performance is similar, but there are differences in improvement within a given training day and/or across training days, this indicates a difference in motor learning. Many different versions of this extended protocol have been used, ranging from 3 to 5 trials for 1 day up to ten trials a day for 8 days in longer versions of the task (Yin et al., 2009). Most common is to utilize 3–4 trials per day across 3–4 days of testing (Rothwell et al., 2014; Lynch 3rd et al., 2020; Benthall et al., 2021; Le Merrer et al., 2023) (Figure 1).

When given home cage access to a running wheel, animals perform better on the rotarod overall, but the rate of both intra-and intersession improvement remains the same, indicating that increasing performance in the task goes beyond gains in locomotor fitness (Buitrago et al., 2004). Instead, animals develop and optimize a sequence of movements that allows them to stay on the rod at faster speeds, which is exemplified by shifts in gait patterns across training from stepping to running (Buitrago et al., 2004). In some cases, differences in performance between models is only revealed in versions of the task that utilize faster speeds, up to 80 revolutions per minute, which necessitates even greater motor program optimization (Rothwell et al., 2014; DiCarlo et al., 2019; Lynch 3rd et al., 2020; Benthall et al., 2021).

Given the multiphasic nature of the accelerating rotarod task, several brain regions are implicated in task performance, including the cortex (Yang et al., 2009; Fu et al., 2012; Ash et al., 2021b), basal ganglia (Costa et al., 2004; Yin et al., 2009; Durieux et al., 2012), and cerebellum (Sathyamurthy et al., 2020; Simmons et al., 2021). In this review, we highlight the role of the basal ganglia, in particular the striatum, in the motor learning that occurs during rotarod training. Given the involvement of the striatum in a number of other motor learning functions, such as instrumental learning and extinction (Yin et al., 2005, 2006; Santos et al., 2015), active avoidance, response-based procedural learning (Pittenger et al., 2006), and shifting from action-outcome to stimulus–response performance (Hawes et al., 2015), altered rotarod performance, which is easily assessed in mice, likely translates into changes in these more difficult to measure corticostriatal-dependent behaviors. In this way, performance in the accelerating rotarod task is an informative indicator of the function of a frequently altered circuit in mouse models of ASD (Li and Pozzo-Miller, 2020).

The striatum, the main input center of the basal ganglia, is composed of GABAergic striatal projection neurons (SPNs) and local interneurons. SPNs, which make up over 95% of striatal neurons, send their outputs to downstream nuclei via two largely parallel pathways. Dopamine D1-receptor expressing SPNs of the direct pathway (dSPNs) send their primary projections to the substantia nigra pars reticulata and globus pallidus internal segment (SNr/GPi) and broadly facilitate movement when activated in bulk (Kravitz et al., 2010; Gerfen and Surmeier, 2011; Tai et al., 2012). D2-receptor expressing SPNs of the indirect pathway (iSPNs) send their primary projections to the globus pallidus external segment (GPe) and generally inhibit movement or suppress competing actions when activated as a population (Kravitz et al., 2010; Gerfen and Surmeier, 2011; Tai et al., 2012; Calabresi et al., 2014). During behavior, both populations of SPNs are activated in a coordinated way to orchestrate movement and decision-making. SPNs are innervated by a variety of inputs, most notably glutamatergic input from the cortex and thalamus, and dopamine input from the midbrain (Ding et al., 2008; Doig et al., 2010; Gerfen and Surmeier, 2011). Despite overall similar cytoarchitecture, the dorsal and ventral regions of the striatum are thought to be implicated in different functions, with the former controlling motor and cognitive functions, and the latter mediating limbic functions such as appetitive behavior and reward (Voorn et al., 2004) (Figure 1).

Further parsing of striatal regions, based primarily on differences in cortical inputs, implicates the dorsomedial striatum (DMS) as an associative region involved in the initial stages of learning action-outcome pairings and the dorsolateral striatum (DLS) as a sensorimotor region involved in the acquisition of habitual or procedural behaviors (Voorn et al., 2004). In both subregions, SPN ensemble activity and plasticity at striatal synapses is important for a variety of learning tasks, including motor skill learning (Costa et al., 2004; Barnes et al., 2005; Dang et al., 2006; Yin et al., 2009; Kupferschmidt et al., 2019). In the accelerating rotarod task, in vivo electrophysiological recordings showed that neurons in the striatum exhibit task-related activity that is highly correlated with performance (Costa et al., 2004; Barnes et al., 2005). Within the striatum, different subregions exhibit dynamic activity patterns throughout different phases of motor learning. In the DMS, positive modulation of firing rate in task-related SPNs predominantly occurs early in rotarod training, while in DLS, this firing rate modulation occurs after extensive training. Consistent with this, lesions of the DMS impair early learning while lesions of the DLS impair both early and late learning (Yin et al., 2009). Together this work establishes a key role for dorsal striatal circuits in rotarod learning.

While initial work highlighted the importance of the dorsal striatum in motor skill learning, several studies suggest that the ventral striatum may also play a role. In particular, a recent study showed that ablation of iSPNs in the nucleus accumbens (NAc) is sufficient to impair rotarod learning (Le Merrer et al., 2023). In addition, as discussed below, ventral striatal-specific manipulation of some ASD risk genes is sufficient to impact rotarod performance (Rothwell et al., 2014; Platt et al., 2017). This fits within the theory first introduced by Haber and colleagues that the ventral and dorsal striatum interact dynamically over the course of learning (Haber et al., 2000). Just as varying cortical inputs form a gradient across dorsolateral and ventromedial striatum, so too do the inputs to and outputs from dopaminergic substantia nigra. Ventral striatal subregions are proposed to influence behavioral gating in dorsal striatal regions through an ascending “spiral” of information through these striatonigrostriatal connections (Haber et al., 2000; Belin et al., 2007). Dynamic changes in the activity and functional roles of different SPN subtypes across this spiral likely occur during rotarod training.

In terms of the striatal cell types involved in motor learning, studies using ex vivo electrophysiology showed that D2-receptor expressing iSPNs of the DLS undergo significant synaptic potentiation during late training and that administration of a D2R antagonist late in training impairs rotarod performance (Yin et al., 2009). This suggests that plasticity of dorsal striatal indirect pathway activity may be important for rotarod learning. A study using adult neurotoxin-induced ablation of iSPNs throughout the striatum confirmed the importance of iSPNs for rotarod performance, particularly for early learning (Durieux et al., 2012). However, it was also shown that ablation of dSPNs throughout the striatum (Durieux et al., 2012), or selectively in the dorsal striatum (Durieux et al., 2012; Le Merrer et al., 2023), impairs rotarod performance, resulting in severe motor learning deficits. This is consistent with other studies showing that manipulations of dorsal striatal dSPNs can impact rotarod performance (Benthall et al., 2021; Ma et al., 2022). In terms of the ventral striatum, Le Merrer and colleagues showed that ablation of iSPNs (but not dSPNs) in the NAc disrupts rotarod performance (Le Merrer et al., 2023). Furthermore, reducing the excitability of dSPNs in the NAc has also been shown to impair motor learning (Rothwell et al., 2014). Together these studies provide evidence that multiple striatal circuits and subregions are required for motor learning and likely play a coordinated role in motor skill acquisition and maintenance.

The differential roles of striatal sub-regions as well as SPN subtypes during different stages of rotarod learning is likely driven by changes in cortical drive (Yin et al., 2009). Indeed, intact glutamatergic corticostriatal transmission is necessary for rotarod learning. Loss of the presynaptic scaffolding protein RIM1 from corticostriatal neurons, which disrupts excitatory transmission in the dorsal striatum, impairs rotarod learning (Kupferschmidt et al., 2019). In addition, striatal-specific deletion of glutamatergic NMDARs results in a significant deficit in learning in the task (Dang et al., 2006). Taken together, these studies show that changes in the synaptic properties of direct and indirect pathway neurons, throughout dorsal and ventral striatum, shape rotarod performance throughout different stages of the task. Our emerging understanding of the synaptic and circuit mechanisms that underlie rotarod learning make it a useful assay to apply to mouse models of disease.

Rotarod performance has been assessed across numerous mouse models with mutations in ASD risk genes, making it a useful assay for identifying potential convergent phenotypes. In surveying the literature, we find that many (but not all) ASD mouse models exhibit altered performance in this task, which can include altered initial performance, a global change in performance, or a difference in learning rate across trials (Table 1). One challenge with making general conclusions from this assessment is that multiple different rotarod protocols have been used. While utilizing a rod that increases in speed from 5 to 40 RPM over the course of 5 min per trial is most common, the number of trials implemented per day, and the total number of days of the task vary greatly across studies. In some cases where multiple protocols have been used, mice can show changes in one version of the rotarod task but not another (Rothwell et al., 2014; DiCarlo et al., 2019; Lynch 3rd et al., 2020; Benthall et al., 2021). Therefore, if no phenotype is reported with one rotarod protocol, it’s possible that performance would be altered if the acceleration speed, number of trials, and/or number of testing days were different.

With this caveat noted, we do find a group of models, including mice with loss-of-function mutations in Mecp2, Shank3 and Ube3a, which show consistent deficits in rotarod performance (Table 1). Some of these models exhibit poor performance from the first trial of the task, exemplified by decreased latency to fall from the rod in trial 1 compared to wild-type (WT) controls, owing to baseline deficits in motor coordination (see Table 1 - models with a deficit in coordination). In other models, trial 1 performance resembles that of WT controls, suggesting intact coordination, however, the latency to fall across trials either does not increase, or increases less than WT controls, indicating a deficit in motor learning (see Table 1 - models with a deficit in learning).

In the case of many loss-of-function Mecp2, Shank3 and Ube3a mutations, mice exhibit deficits in both initial coordination and motor learning. The phenotypes observed in these mouse models may reflect the motor deficits that occur in individuals with mutations in these genes (Chukoskie et al., 2013; Troyb et al., 2016; Caldwell-Harris, 2021). Specifically, while motor function can be quite variable across individuals with ASD as a whole, one of the core diagnostic criteria of Rett syndrome, which is caused by loss-of-function mutations in the MECP2 gene, is the deterioration of motor function, often resulting in complete loss of mobility in patients (Chahrour and Zoghbi, 2007). Similarly, patients with Angelman syndrome, a neurodevelopmental disorder caused by mutations in the UBE3A gene, generally exhibit severe motor dysfunction including orthopedic and movement difficulties, walking that is stiff or jerky, and a lack of coordination or development of complex motor skills (Rotaru et al., 2020). A comprehensive clinical assessment of 17 individuals with point mutations in the SHANK3 gene, a gene located within the 22q13.3 chromosomal region implicated in the neurodevelopmental disorder Phelan-McDermid syndrome, identified less severe motor dysfunction than typically seen in the above syndromes; however, nearly all individuals assessed exhibited hypotonia and gait abnormalities (De Rubeis et al., 2018). The identification of motor dysfunction as a common clinical presentation caused by mutations in these genes, alongside the consistently decreased rotarod performance seen in models of these syndromes lends face validity to the rotarod assay.

Notably, while phenotypic analysis of animal models of neuropsychiatric disorders often focuses on identifying deficits, there is a cohort of ASD mouse models that show increased performance on the rotarod task (Table 1). The enhanced performance in these models can either be apparent from initial testing onward or revealed over the course of training. In the remainder of this review, we will focus specifically on these “gain-of-function” cases and discuss how enhanced motor learning may reflect changes in striatal circuit function that could facilitate the development of RRBs.

The majority of studies assessing accelerating rotarod performance have used mouse models with global mutations in ASD risk genes; therefore, the brain region or circuit responsible for the phenotype is difficult to ascertain. As discussed above, striatal circuits have been identified as a key node in rotarod motor learning. To directly test the contribution of striatal neurons to rotarod phenotypes, conditional KO mice have been generated in which the ASD risk gene is manipulated selectively in striatal neurons. These studies have revealed a direct link between striatal function and rotarod performance.

In the case of Nlgn3, conditional deletion in cells of the direct, but not the indirect, pathway of the striatum results in increased rotarod performance (Rothwell et al., 2014). A deficit in inhibition specifically onto dSPNs in these mice, which is expected to enhance excitability of the direct pathway, likely contributes to their increased rotarod performance. Indeed, in WT mice, a manipulation that reduces the activity of indirect pathway cells, which would have the net effect of facilitating direct pathway activation of downstream basal ganglia nuclei, results in increased performance in the task. In addition, rotarod performance is restored to WT levels in Nlgn3 dSPN conditional KO mice via expression of the potassium channel Kir2.1, which decreases dSPN excitability (Rothwell et al., 2014). This study provides compelling evidence that altered balance between the striatal direct and indirect pathways can contribute to altered motor learning in ASD mouse models. Interestingly, conditional deletion of Nlgn3 in the NAc, and not broadly in the dorsal striatum, is sufficient to recapitulate the enhanced rotarod performance (Rothwell et al., 2014). This finding supports a potentially underappreciated role for the ventral striatum in motor learning. A possible explanation for the cell type and anatomical specificity of Nlgn3 deletion is that Nlgn3 is preferentially expressed in dSPNs of the NAc and therefore expected to have a greater effect when disrupted in these cells (Rothwell et al., 2014).

As discussed above, multiple studies of Chd8 mouse models have identified enhanced rotarod performance (Platt et al., 2017; Hulbert et al., 2020). One such study performed gene expression analysis across brain regions in Chd8^+/− mice, identifying the NAc as a region with significant gene dysregulation. Following this, Platt et al. injected Chd8-targeting sgRNA into the NAc in a Cas9 knock-in mouse to determine the impact of Chd8 reduction specifically in this region. Similar to the findings in the Nlgn3 study (Rothwell et al., 2014), reduction of Chd8 specifically in the NAc, and not the dorsal striatum, recapitulated the increased rotarod performance seen in constitutive heterozygous mice (Rothwell et al., 2014; Platt et al., 2017). Electrophysiological assessment of these mice found that SPNs of the NAc core have increased frequency and amplitude of spontaneous excitatory synaptic currents, as well as decreased amplitude of miniature inhibitory synaptic currents, suggesting overall increased excitatory drive of SPNs in the region (Platt et al., 2017). A distinction between dSPNs and iSPNs was not made in this study.

While the above studies implicate altered NAc function as a driver of enhanced rotarod performance, a recent study identified increased motor learning in a mouse model with dorsal striatum-selective Tsc1 loss (Benthall et al., 2021). In this study, dSPN-specific deletion of the ASD-risk gene Tsc1 resulted in increased performance on the accelerating rotarod, in particular at higher speeds (10–80 RPM). Mice with loss of Tsc1 in iSPNs did not exhibit changes in rotarod performance, consistent with the findings in Nlgn3 mice (Rothwell et al., 2014; Benthall et al., 2021). The D1-Cre line utilized in this study to target dSPNs is relatively restricted to dSPNs of the dorsal striatum, sparing the majority of NAc cells (Benthall et al., 2021). This, together with the results described above, indicate that altered direct pathway activity in either the dorsal or ventral striatum is sufficient to alter motor learning.

In Tsc1 dSPN KO mice, electrophysiology experiments revealed that Tsc1-KO dSPNs have increased glutamate release probability at cortical inputs, resulting in enhanced corticostriatal drive. This study also found a deficit in eCB-LTD onto Tsc1 KO dSPNs, which may explain the change in presynaptic release probability (Benthall et al., 2021). This prominent form of striatal synaptic depression works through the release of postsynaptic endocannabinoids that act on cortical presynaptic CB1 receptors, ultimately reducing the probability of neurotransmitter release (Kreitzer and Malenka, 2008; Lovinger, 2010). Loss of eCB-LTD onto Tsc1 KO dSPNs likely renders these cells unable to depress excitatory inputs, leading to increased corticostriatal drive over time. Interestingly, a similar deficit in eCB-LTD was identified in the dorsal striatum of a Nlgn3 mutant mouse model that exhibits enhanced performance in the rotarod task (Martella et al., 2018).

Here we have highlighted several examples of mouse models that exhibit enhanced rotarod performance. In a few of these models, striatal-specific manipulation of an ASD-risk gene was sufficient to induce changes in rotarod motor learning. In several studies, synaptic changes were reported that are expected to enhance striatal activation, particularly increase corticostriatal drive and/or excitability of the direct pathway (Rothwell et al., 2014; Platt et al., 2017; Benthall et al., 2021). Given the importance of striatal circuits for not only motor skill leaning but also habit learning etc., it seems plausible that gain-of-function at the neural circuit level facilitates the formation of fixed motor routines or perseverative behaviors. For ASD models where motor deficits predominate, the neural circuitry underlying the presence of repetitive behaviors remains to be established. In these cases, basic motor circuits may be disrupted such that rotarod deficits arise, but gain-of-function in other motor control circuitry likely drives the emergence of RRBs, for example in the case of Shank3 models (Drapeau et al., 2018). Further investigation into the motor phenotypes of ASD mouse models and the synaptic and circuit basis of repetitive behaviors will provide additional insight into this core aspect of ASD.

As the number and heterogeneity of identified ASD risk genes continues to expand, the utilization of common behavioral assays to identify convergent phenotypes in mouse models of ASD is of great benefit. When it comes to assessing motor symptom domains of ASD in mouse models, the accelerating rotarod task has proved very useful for identifying phenotypes. The task is relatively fast and straightforward to carry out, reveals information about gross motor coordination and motor learning, and has established underlying neural circuitry. Corticostriatal circuits are key regulators of rotarod performance and are increasingly implicated as a point of convergent alteration across a range of mouse models of ASD (Li and Pozzo-Miller, 2020). If these circuits are found to consistently contribute to altered behavior, they represent a potential site for targeted therapeutics, which may be applicable across ASDs of different genetic origin.

KC: Conceptualization, Funding acquisition, Writing – original draft, Writing – review & editing. HB: Conceptualization, Funding acquisition, Supervision, Writing – review & editing.