To our knowledge, this is the first study to explore motor learning in children with autism compared to both a TD control group and a clinical control group (ADHD group) known to be also associated with motor impairments [Denckla & Rudel, 1978
; Klein et al., 2006
; Klimkeit et al., 2005
; Mostofsky et al., 2001
]. Inconsistent with our first hypothesis, we found that children with HFA showed similar degrees of learning (i.e., similar rates of improvement in time on target) across the circular blocks compared to TD children. This lack of difference suggests that children with autism were able to learn the motor sequence. However, despite this similarity in performance, our findings of significant between-group differences in the effect of the interference (square) block suggest that, compared to TD children, children with HFA demonstrate differences in the pattern
(i.e., approach) of learning a novel motor sequence necessary to optimize accurate performance of a task. This was seen when all subjects performed the RP task under the same conditions (Experiment 1) as well as when the task was adjusted to minimize differences in motor execution (Experiment 2). In contrast, children with ADHD did not show impaired learning (demonstrated by a change in performance) on either task compared to TD children.
Our findings using the traditional RP design, with speed held constant across subjects (Experiment 1), revealed that children with autism stayed on-target for significantly less time and showed less change across trials than both
the ADHD and the TD groups independently. Therefore, while the findings from Experiment 1 reveal that children with HFA demonstrate less change across the blocks of trials, they left open the question of whether differences in visuomotor learning could be accounted for by problems with motor execution, manifested as lower time-on-target. The fact that children with ADHD show a similar pattern of performance as do the TD children despite having significantly poorer motor execution suggested otherwise. More critically, we were able to directly address the question of the impact of motor execution in Experiment 2. Using an approach adapted from van Gorp et al. 
, we were successful in minimizing group differences in motor execution, while keeping all other components of the testing the same. In contrast to Experiment 1, analyses of Experiment 2 revealed no significant differences in motor execution (measured as time-on-target) across the three groups. Consistent with Experiment 1, we found that even after minimizing group differences in motor execution, children with HFA showed significantly less change in performance across trials than did the TD group. Once again, change across the blocks of trials in the ADHD group did not significantly differ from the TD group, suggesting that differences in motor learning may be specific to autism.
More detailed analysis of the pattern of findings revealed that the HFA-associated differences in change across trials were primarily driven by an effect of the intervening square block (block 3). For both experiments, the children with ADHD and the TD children demonstrated a marked interference effect with the square block, while the children with HFA showed very little interference. This is despite the fact that children with HFA often show cognitive inflexibility and have associated difficulty with shifting (“transitioning”) from one task to another. The children with HFA did, however, show improvement in performance across the circular blocks of trials comparable to that seen in TD children and children with ADHD. The pattern of findings suggests that while children with HFA were able to improve their RP performance, the manner in which they did so differed from TD children (and children with ADHD). Likewise, other studies have revealed that despite seemingly intact performance by children with autism, their approach to the task either differed or they required greater resources (i.e., worked harder) than their comparison groups [Bowler, 1992
; Eisenmajer & Prior, 1991
; Happe, 1995
Evidence from prior studies suggests that children with autism show an over-reliance on declarative, rather than procedural, processes typically used to acquire motor skills [Klinger & Dawson, 2001
; Walenski et al., 2008
]. Increased reliance on declarative learning during RP would result in less “proceduralization” of the visuomotor (circular) sequence such that there would be less interference when a competing (square) pattern was introduced.
Declarative and procedural memory systems are not entirely independent systems. Instead, they appear to dynamically interact in both a cooperative and competitive manner during motor sequence learning. Declarative mechanisms are favored during the initial learning stages; however, once the sequence is acquired, improvement in terms of speed of performance appears to be related to procedural mechanisms [Brown & Robertson, 2007a
]. Given our findings, it may be that the pattern of circular motion was less well ingrained in the children with HFA than in TD children, and that the children with HFA had not yet recruited procedural mechanisms.
Alternatively, the findings of decreased interference effect might be explained by alterations in the mechanisms underlying procedural learning. Findings from previous studies suggest that autism may be associated with a preference for reliance on proprioceptive, rather than visual, information to guide acquisition of novel movement patterns [Masterton & Biederman, 1983
]. This might explain why the switch to a different visual presentation (from circle to square) disrupted performance to a significantly lesser degree in the group of autism subjects.
The altered pattern of motor sequence learning might help to explain impaired motor skill development in children with autism. In particular, parents often report that their children with ASD are delayed in the acquisition of motor skills that involve learning novel patterns of sequenced movements, such as peddling a tricycle, pumping legs on a swing and a range of fine motor skills (e.g. buttoning, zippering, tying shoelaces). These reports are consistent with a description by Wing [1969
, in Smith and Bryson] that, “clumsy children with autism reportedly have particular difficulty with learning
organized patterns of movements (e.g. skipping and dancing)” (p 267). Future studies should begin to focus on teasing apart whether these impairments are due to decreased reliance on, or abnormality in, procedural learning. A better understanding could help to shape intervention and treatment for children with ASD.
The findings that children with autism use a different approach to motor sequence learning may also provide insight into neurologic abnormalities associated with autism. Motor sequence learning relies on a broad neural network principally involving connections between frontal and parietal cortices and subcortical regions [Doyon, Penhune, & Ungerleider, 2003
; Eslinger & Damasio, 1986
]. The frontal regions, including motor and premotor areas, are particularly important for the storage of motor representations of movement sequences [Doyon et al., 2003
] and the retrieval of sequenced motor responses necessary to guide proper execution [Doyon et al., 1997
], while parietal regions are critical for the storage and retrieval of spatial and temporal representations of this movement [Heilman & Gonzalez Rothi, 2003
]. Findings from imaging studies point to additional distinctions in contributions from subcortical regions, including the cerebellum and basal ganglia. Using positron emission tomography, Grafton, Woods, and Mike 
found robust learning-dependent cerebellar activation in early RP learning. However, once participants learned the motor sequence and demonstrated optimal task performance, activity within the cerebellum became undetectable, suggesting that the cerebellum is particularly important for the early learning stages and acquisition of a motor sequence program [Doyon, Owen, Petrides, Sziklas, & Evans, 1996
; Jueptner, Frith, Brooks, Frackowiak, & Passingham, 1997
; Jueptner, Stephan et al., 1997
]. This appears to involve neuronal mechanisms of long-term depression (LTD) which contribute to refinement of the motor sequence based on internal models involving error detection through comparison of intended and actual response [Ito, 2005
]. In contrast, the basal ganglia, which is critical for movement selection, has been implicated in the acquisition of motor sequences as well as in the encoding and long-term storage of well-learned sequenced movements [Doyon et al., 2003
There is a good deal of evidence for autism-related abnormalities in regions within these cortical–subcortical circuits, providing a neuroanatomic basis for deficits in procedural/motor skill learning. Neurologic, neuropsychologic, postmortem, neurophysiologic and neuroimaging studies have long since suggested that motor impairments observed in autism may stem from dysfunction in frontal and subcortical structures and circuits [Bailey et al., 1998
; Carper & Courchesne, 2005
; Casanova, Buxhoeveden, & Brown, 2002
; Chugani et al., 1999
; Hardan, Kilpatrick, Keshavan, & Minshew, 2003
; Hughes, 1996
; Mostofsky et al., 2000
; Muller, Kleinhans, Kemmotsu, Pierce, & Courchesne, 2003
; Rinehart et al., 2006
]. Abnormalities in many of these structures are common findings in studies of autism. Abnormalities in frontal and parietal cortices as well as the basal ganglia and cerebellum have been reported in several imaging studies [Abell et al., 1999
; Carper & Courchesne, 2000
; Carper, Moses, Tigue, & Courchesne, 2002
; Hardan et al., 2003
; Kates et al., 1998
; McAlonan et al., 2002
; Muller et al., 2003
; Piven, Arndt, Bailey, & Andreasen, 1996
; Sears et al., 1999
] and histological studies reveal abnormalities in the minicolumn structure within the frontal cortex [Buxhoeveden et al., 2006
; Casanova et al., 2002
]. Particularly compelling is the fact that the decreased Purkinje cell count in the cerebellum is the most consistent finding in postmortem studies of autism [Bailey et al., 1998
; Bauman & Kemper, 1994
; Fatemi et al., 2002
; Ritvo et al., 1986
; Williams, Hauser, Purpura, DeLong, & Swisher, 1980
], which would prompt speculation that cerebellar dysfunction, perhaps related to impairments in LTD within the Purkinje cell, contributes to impaired motor sequence learning. However, findings revealing that children with HFA have intact motor adaptation [Mostofsky et al., 2004
] might suggest otherwise.
It is also worth considering whether the observed difficulties with motor sequence learning might not be related to dysfunction within a particular brain region, but rather whether it is a consequence of abnormalities in connections between these regions. There is increasing evidence in the current literature that suggests reduced connectivity between these distant brain regions in individuals with autism [Herbert et al., 2004
]. Anatomic imaging studies reveal an overgrowth of radiate white matter regions immediately underlying the cortex [Herbert et al., 2004
] with increased radiate white matter volume in primary motor cortex being a robust predictor of motor impairment in children with autism [Mostofsky, Burgess, & Gidley Larson, 2007
]. Investigators have further suggested that a relative undergrowth of more distant connections between cerebral cortical regions and subcortical structures [Happe & Frith, 2006
; Herbert et al., 2004
] results in impaired complex information processing [Minshew et al., 1997
] and “weak central coherence” [Shah & Frith, 1993
]. Given that motor learning relies on a distributed neural network, including cortico-cortical, cortico-striatal and cortico-cerebellar connections, it follows that individuals with autism may demonstrate a deficit in motor learning due to reduced connectivity between distant brain regions. Future research employing imaging techniques, such as diffuse tensor imaging and functional magnetic resonance imaging examinations of functional connectivity, may provide insight into abnormalities in the interconnections between brain regions critical for motor sequence learning.
The current findings that children with HFA utilize a different approach to motor skill learning may also have implications for development of the broader phenotype of social and communicative deficits in autism. After an extensive review of the literature describing motor deficits in autism, Leary and Hill 
concluded that the extent of the motor disturbances present in autism “can clearly have a profound effect on a person’s ability to regulate movement in order to effectively communicate, relate, and participate with others” (p 44).
Alternatively (or perhaps in parallel), abnormalities in procedural learning have been shown to be important not only for the acquisition of motor skills, but also for language and other aspects of social communication. There is robust evidence from behavioral, electrophysiology and imaging studies that support a model in which procedural learning is critical in the development of syntax and grammar, in contrast to lexical/semantic domains that appear to be dependent on declarative learning [for reviews, see Ullman, 2001
]. Differences in procedural learning may thereby contribute to problems with syntactic formulation in autism [Walenski et al., 2006
] and help to explain why children with autism frequently display overly “scripted” speech in which memorized, rote phrases are sometimes used in conversation instead of self-generated syntax.
Abnormalities in procedural learning might also impact the development of non-verbal communication and socialization. Recent findings from multiple groups of investigators reveal that for children with autism there exist robust correlations between performance of skilled gestures and measures of social/communicative impairment [Dziuk et al., 2007
; Freitag et al., 2007
]. It may be that procedural learning mechanisms critical for the development of motor skills may also contribute to impaired development of social and communicative gestures in autism. For TD populations, it would appear to be unlikely that social and communicative gestures (i.e. waving, blowing a kiss) are acquired through declarative learning mechanisms involving explicit memorization; rather it is more plausible that the automaticity of these social and communicative gestures is achieved through procedural means involving motor sequence learning.
While the experimental design and use of a clinical control group are strengths of the current study, there are limitations. Given the nature of learning tasks, it was necessary to use two independent groups of children for the two experiments. While this does introduce inter-individual variability as a confound, the groups were phenotypically similar; both groups met the same diagnostic and eligibility criteria and there were no statistically significant differences in gender distribution, age, IQ (as measured by PRI, with the exception of the HFA groups) or ADOS scores across the two groups. Furthermore, this study comprised a single session of learning. Research has suggested that practice and delay are integral to learning a novel motor sequence [Savion-Lemieux & Penhune, 2005
]. Thus, the examination of motor learning in children with ASD over days and months is an important area of future study.
It is also important to note that motor learning during RP and other tasks used in studies of autism (e.g. SRTT) are visually guided. Previous research has suggested that children with autism tend to rely on proprioceptive rather than visual feedback to adapt arm movements [Masterton & Biederman, 1983
]; and more recently abnormalities in motion perception, processing and coherence [Bertone, Mottron, Jelenic, & Faubert, 2005
; Dakin & Frith, 2005
; Milne et al., 2002
] have been reported in children with ASD. Examining motor learning across other modalities where visual guidance/feedback is not, or is less, necessary (i.e. tasks in which learning depends on somatosensory/proprioceptive feedback) is an important area of future study.
In summary, the present findings suggest that children with HFA show differences in the pattern of visuomotor sequence learning and that these differences persist even after minimizing individual differences in motor execution. Evidence for specificity of this impairment is demonstrated in that children with ADHD did not show differences in the pattern of motor learning compared with the same group of TD children. Detailed analysis revealed that while children with HFA showed similar gains across blocks of trials, they failed to show an expected decline in performance when an interfering pattern was introduced. The pattern suggests that autism may be associated with increased reliance on declarative learning, which could interfere with procedural learning, or, alternatively, might represent a mechanism by which children with autism compensate for underlying deficits in procedural learning. Future studies examining correlations between RP learning and measures of declarative memory would help to resolve this question. Furthermore, given that declarative memories are prone to decay, are less fixed and are highly flexible [Cohen et al., 1997
], studies of motor learning and retention over days to weeks would also help to understand the pattern of learning deficits in autism. Understanding autism-associated differences in learning could help to guide treatment intervention and might also provide insight into the biological basis of the motor deficits associated with autism, as well as impairments in socialization and communication that are the hallmarks of autism.