Visual and proprioceptive feedback mechanisms of precision manual motor control in autism spectrum disorder

Individuals with Autism Spectrum Disorder (ASD) show decits processing sensory feedback to reactively adjust ongoing motor behaviors. Atypical reliance on visual and proprioceptive feedback each have been reported during motor behaviors in ASD suggesting that impairments are not specic to one sensory domain but may instead reect a decit in multisensory processing, resulting in reliance on unimodal feedback. The present study tested this hypothesis by examining motor behavior across different visual and proprioceptive feedback conditions during a visually guided precision grip force test. development of sensorimotor abilities in ASD characterized by delayed maturation of precision sensorimotor control. These results help clarify the sensory feedback processes contributing to decits in online motor control in individuals with ASD and provide new insights into important neurodevelopmental processes that contribute to the disorder.


Abstract
Background Individuals with Autism Spectrum Disorder (ASD) show de cits processing sensory feedback to reactively adjust ongoing motor behaviors. Atypical reliance on visual and proprioceptive feedback each have been reported during motor behaviors in ASD suggesting that impairments are not speci c to one sensory domain but may instead re ect a de cit in multisensory processing, resulting in reliance on unimodal feedback. The present study tested this hypothesis by examining motor behavior across different visual and proprioceptive feedback conditions during a visually guided precision grip force test.

Methods
Participants with ASD (N = 43) and age-matched typically developing (TD) controls (N = 23), range [10][11][12][13][14][15][16][17][18][19][20] years, completed a test of precision gripping. They pressed on force sensors with their index nger and thumb while receiving visual feedback on a computer screen in the form of a horizontal bar that moved upwards with increased force. They were instructed to press so that the bar reached the level of a static target bar and then to hold their grip force as steadily as possible. Visual feedback was manipulated by changing the gain of the force bar. Proprioceptive feedback was manipulated by applying 80 Hz tendon vibration at the wrist to induce an illusion of muscle elongation. Force variability (standard deviation) and irregularity (sample entropy) were examined using multilevel linear models.

Results
While TD controls showed increased force variability with the tendon vibration on compared to off, individuals with ASD showed similar levels of force variability across tendon vibration conditions. Individuals with ASD showed stronger age-associated reductions in force variability relative to controls across conditions. The ASD group also showed greater age-associated increases in force irregularity relative to controls, especially at higher gain levels and when the tendon vibrator was turned on.

Conclusions
Our ndings that individuals with ASD show similar levels of force variability and regularity during induced proprioceptive illusions suggest a reduced ability to integrate proprioceptive feedback information to guide ongoing precision manual motor behavior. We also document stronger ageassociated gains in force control in ASD relative to TD suggesting delayed development of multisensory feedback control of motor behavior.
Multiple studies have indicated that individuals with ASD have de cits in processing sensory feedback to reactively adjust ongoing motor behaviors. Across multiple effector systems, including those involved in precision gripping [14] and postural control [11,12], individuals with ASD show increased variability and regularity of continuous motor behaviors. Variability represents spatial inconsistency in the movement and regularity represents temporally in exible motor behavior. Elevated variability and regularity of movement in persons with ASD indicate that they are not able to make spatially accurate and temporally precise adjustments to ongoing motor output in response to sensory feedback.
Additionally, de cits in sensory feedback processing for motor control in persons with ASD implicate multiple sensory modalities. In studies of motor learning, individuals with ASD learned to adapt to proprioceptive errors more e ciently than typically developing (TD) controls indicating that persons with ASD were over-reliant on proprioceptive feedback for motor learning [6,16,17]. In our studies of visuallyguided ne motor control, participants with ASD showed elevated motor variability and regularity compared to TD controls during precision gripping, especially when visual feedback was enhanced (high visual gain) or degraded (low visual gain) [14], indicating that persons with ASD were over-reliant on visual feedback even when it was degraded or ampli ed.
Collectively, behavior-speci c ndings of visual or proprioceptive bias in ASD suggest that sensorimotor de cits are not speci c to a sensory domain but may instead be task-dependent and re ect di culties integrating information across sensory domains to dynamically adjust motor output. Consistent with this hypothesis, several studies have found that individuals with ASD have de cits in multisensory integration, even though processing of simple, unimodal stimuli is largely intact [25][26][27][28]. During postural control -for which proprioceptive feedback is primary -individuals with ASD show elevated variability of their center of pressure (COP) when proprioceptive feedback is perturbed (tendon vibration), whereas TD controls are able to compensate for disrupted proprioceptive feedback by relying more heavily on a secondary source of feedback (in this case, visual) to minimize COP variability [13]. These results indicate that individuals with ASD are unable to reweight different sources of sensory feedback (i.e., upweight secondary sources) in response to perturbations of the primary sensory input.
To test this hypothesis, the present study manipulated visual and proprioceptive feedback within a visually guided precision gripping task to assess how each feedback source in uenced motor control in individuals with ASD. The precision gripping test used here involves continuous visual feedback, which has been shown to be the primary sensory feedback source for online control of visually-guided upper limb movements [29][30][31]. We expected individuals with ASD would show increased variability and regularity during precision gripping relative to controls, especially when visual (primary) feedback was enhanced or degraded. This nding would support the hypothesis that individuals with ASD have di culty down-weighting feedback from the primary sensory domain for visually guided movement. We also expected that force variability and regularity in individuals with ASD would be minimally impacted when proprioceptive feedback was manipulated with tendon vibration, consistent with an inability to utilize secondary sources of sensory feedback to optimize motor output.

Participants
Forty-three participants with ASD (11 females) and 23 TD controls (12 females) matched on age (range 10-20 years) and handedness completed tests of precision gripping with their dominant hand (Table 1).
Participants with ASD were recruited through our research registries comprised of individuals evaluated through the University of Kansas Health System who have consented to be contacted for research purposes, and though community advertisements. TD controls were recruited through community advertisements. ASD diagnoses were con rmed based on Diagnostic and Statistical Manual of Mental Disorders, Edition 5 (DSM-V) [1] criteria using the Autism Diagnostic Observation Schedule, Second Edition (ADOS-2) [32], Autism Diagnostic Interview -Revised (ADI-R) [33], and expert clinical opinion. Participants with ASD were excluded if they had a known genetic or metabolic disorder associated with ASD (e.g., Fragile X syndrome) or a full scale IQ below 60 as measured using the Wechsler Abbreviated Scales of Intelligence, Second Edition (WASI-II) [34]. TD participants were excluded if they scored ≥ 8 on the Social Communication Questionnaire [35]; reported a history of psychiatric or neurologic disorders; had a family history of ASD in rst-, second-, or third-degree relatives; had a family history of a developmental or learning disorder, psychosis, or obsessive compulsive disorder in rst-degree relatives, or had a full scale IQ below 85 as measured using the WASI-II. Participants also were excluded if they had a history of head injury, birth injury, or seizure disorder. No participants were taking medications known to affect sensorimotor behavior, including antipsychotics, stimulants, or anticonvulsants at the time of testing [36]. All participants had corrected or uncorrected visual acuity of at least 20/40. Adult participants provided written informed consent after a complete description of the study, in accordance with the Declaration of Helsinki and the approved University of Kansas Medical Center Institutional Review Board study protocol (IRB#: STUDY00140269). For participants under the age of 18 and adults who were under legal guardianship, a parent or legal guardian provided written informed consent on behalf of the participant, and the participant provided written assent. All study procedures were approved by the local Institutional Review Board.

Precision grip testing
Participants completed tests of precision gripping in a darkened room while seated 52cm from a 67cm (27in) Samsung LCD display monitor with a resolution of 1920x1080 and a 120 Hz refresh rate (Fig. 1). Participants sat with the elbow of their dominant hand comfortably positioned at 90 deg and their forearm resting in a custom arm brace xed to the table to provide stability during testing. To assess precision grip behavior when proprioceptive feedback was disrupted, participants completed grip testing with a tendon vibrator (VB 115, Techno Concept, Cereste, France) securely fastened on their wrist. A velcro strap held the tendon vibrator in place against the exor tendons, including carpi radialis and carpi ulnaris. Towels were placed underneath the participants' wrist to cushion the tendon vibrator from the surface of the table. The participants used their thumb and index nger of their dominant hand to press against two opposing precision load cells (ELFF-B4-100N; Entran) 1.27cm in diameter that were secured to a custom grip device attached to the arm brace. A Coulbourn (V72-25) resistive bridge strain ampli er received analog signals from the load cells. Data were sampled at 100 Hz with a 16-bit analog-to-digital converter (DI-720; DATAQ Instruments) and converted to Newtons of force using a calibration factor derived from known weights before the study [14].
To assess individuals' grip force control at a xed percentage of their maximum grip force, each participant's maximum voluntary contraction (MVC) was calculated for their dominant hand prior to testing. Participants completed three trials in which they were asked to press as hard as they could for three seconds. The average of the participant's maximum force output across these trials comprised their MVC.
During the precision gripping task, participants viewed two horizontal bars on the screen (Fig. 1B). A horizontal white force bar moved upward with increased force and downward with decreased force, and a static bar representing the target force was red during periods of rest and turned green to cue the participant to begin pressing at the beginning of each trial. Participants were instructed to press the load cells as quickly as possible when the red target bar turned green and to keep pressing so that the white force bar stayed as steady as possible at the level of the green target bar.
To test the impact of different sensory feedback processes on grip force behavior, participants completed testing across multiple levels of visual and proprioceptive feedback. As in our previous study [14], visual feedback was manipulated by changing the visual gain of the white force bar (i.e., the vertical distance measured in visual angle that the force bar moved in response to a unit of change in force output). For example, for the three visual gain conditions used in the present study, the force bar moved upward 0.059° per 1N increase in force output at the lowest visual gain, 0.623° per 1N increase in force at medium visual gain, and 6.575° per 1N increase in force at the highest visual gain. These gain levels were selected based on ndings from Vaillancourt et al. [40] that showed increases in force variability and regularity as visual angle increased up to 1°, beyond which force variability and regularity were relatively constant.
Proprioceptive feedback was manipulated by applying tendon vibration to the underside of the wrist during gripping. The tendon vibrator stimulates the proprioceptive system by exciting muscle spindle Ia afferents, eliciting a proprioceptive illusion of muscle stretch in the agonist muscles when vibration is administered at a frequency of at least 40 Hz [41]. Participants completed precision grip trials with the tendon vibrator turned on at a frequency of 80 Hz based on prior research suggesting multiple motor behaviors are reliably disrupted at 80 Hz [42]. Participants also completed trials while wearing the tendon vibrator turned off (no proprioceptive illusion) keeping wrist position consistent across conditions.
Participants completed blocks of 5 trials at each gain level and tendon vibration frequency using their dominant hand (5 trials x 3 gain levels x 2 vibration conditions = 30 trials). Trials were 15s in duration and alternated with 15s rest periods. Each block was separated by 30s of rest. The target force was set to 45% of the participant's MVC for all trials. The tendon vibration off condition was always administered prior to the on condition as vibration effects on motor control can persist for at least 20 min after the tendon vibration is turned off [43]. The order of gain levels was randomized across participants.

Data processing
Force traces for each trial were low-pass ltered via a double-pass fourth-order Butterworth lter at a lowpass cutoff of 15 Hz in MATLAB (MathWorks, Inc., Natick, Massachusetts). Data were processed using a custom MATLAB scoring program previously developed by our lab [15]. To account for variability in the rate at which participants reached the target force, a minimum of 8s and a maximum of 12s of the 15s trial data (from start cue to stop cue) were used for analyses. Trials were excluded if they had less than 8 seconds of sustained force output, the load cells were not properly re-zeroed between trials, or if there were indications that the participant was not following instructions (e.g., the mean force exceeded twice the target force, the mean force was less than half of the target force, there was evidence that the participants used ngers other than their index nger and thumb to press). Based on these criteria, 10.0% of trials were excluded. Force data were linearly detrended to account for drift in participants' force output over the duration of the trial. The mean force of the trial divided by the target force was used as a measure of force accuracy. To assess force variability, the standard deviation (SD) of the force time series was examined. To test the time dependent regularity of the force time series, sample entropy (SampEn) was calculated for each trial [44,45]. SampEn is de ned as the natural logarithm of the conditional probability that two similar sequences of m data points in a timeseries of a given length (N) remain similar within a tolerance level (r) at the next data point in the series. SampEn returns a value between 0 and 2. Lower values of SampEn indicate greater regularity of the timeseries (e.g., a sine wave, with its predictable oscillating pattern, would have a SampEn value near 0). SampEn has been shown to be stable with as few as 200 data points in the timeseries. Parameter settings for SampEn calculations were m = 2 and r = .2 x SD of the timeseries. The timeseries length ranged from 800 to 1200 data points (8-12sec sampled at 100 Hz

Statistical Analysis
Force accuracy, SD, and SampEn were analyzed using separate linear multilevel mixed effects models (MLM) [48,49]. MLM allows for the analysis of within-and between-subjects xed effects while allowing within-subjects effects to vary randomly and is robust to missing data. Gain level (Low, Medium, High) and vibration condition (On, Off) were included as level 1 predictors. Group (ASD, TD), age, sex, and perceptual IQ (PIQ) were included as level 2 predictors. Random intercepts of participant also were included in our models.

Initial models included three-way interactions of Group x Gain Level x Vibration Condition, Group x Gain
Level x Age, and Group x Vibration Frequency x Age, as well as all relevant 2-way interactions and main effects terms. To maintain the most parsimonious models possible, other 3-way and 4-way interactions were not included. Sex and perceptual IQ (PIQ) effects also were tested in the models, as these variables signi cantly differed between groups. Models were tted using the maximum likelihood approach to allow for model comparisons. Terms were removed systematically, and model t was compared between the previous model and the model with the removed term using likelihood ratio tests. Terms that did not signi cantly improve model t (p < 0.05), based on the model comparisons, were not included in the nal models. Satterthwaite's method was used to calculate degrees of freedom for the nal model and post hoc comparisons [50]. Due to the inherent challenge in determining denominator degrees-of-freedom and calculating p-values for MLMs, we treated the t-value as a z-value and used a z > 1.96 threshold as an additional guideline for determining whether terms explained signi cant variance in the model [50]. Pearson correlations were used to assess the relation between experimental variables and ASD symptom severity measured using the ADOS Composite Severity Score (ADOS-CSS). To assess associations between precision force outcomes and sensory issues, the Visual Processing and Movement Processing subscales of the SP-2 and Adolescent/Adult SP were examined. Analyses for SP-2 (N = 29) and Adolescent/Adult SP (N = 11; three participants did not complete the Adolescent/Adult SP) were conducted independently as scores are not standardized across the two versions of this measure. Force variability and regularity also were examined in relation to the Fine Motor Control Subscale of the BOT-2. Three participants with ASD did not complete the BOT-2 (N = 40). P-values were adjusted using false discovery rate (FDR) to limit Type I error for each set of correlations; however, due to small sample sizes and the exploratory nature of these analyses, interpretation of results focuses on effect sizes (r values).

Force Variability
The results of the model for force SD are summarized in Table 2. Group differences in force SD varied as a function of age ( Fig. 2; β = -0.573, R 2 = .168, t 65.6 = -4.054, p = .0001) and tendon vibrator condition  Force Regularity Force regularity varied as a function of age, but the strength of this relationship differed between groups and was dependent on visual gain level (Table 3;

Correlations with Symptom Severity
Correlations between force SD and clinical ratings are shown in Table 4. The Movement Processing subscale of the SP-2 was positively trending with force SD in the tendon vibrator off condition (r = .38, p = .09) and the low (r = .41, p = .07) and medium visual gain conditions (r = .39, p = .06). Force SD was not correlated with the SP-2 Movement Processing subscale for any other conditions, and SD did not correlate with the SP-2 Visual Processing subscale for any visual gain or tendon vibration conditions. The BOT-2 Fine Motor Control Subscale showed negative trends with force SD in the tendon vibrator on (r = − .41, p = .06) and medium visual gain (r = − .38, p = .06) conditions. Force SD correlations with the ADOS-CSS and the Movement and Visual Processing subscales of the Adolescent/Adult SP did not survive FDR corrections, though effect sizes indicated moderate associations (r > 0.3) for some sensory conditions, including tendon vibration off, and all visual gain conditions (Table 4). Force SampEn correlations did not survive FDR corrections for any clinical measures or sensory conditions, though effect sizes, reported in Table 5 indicted moderate correlations for some conditions including tendon vibration on and medium visual gain.

Sensory feedback processing during motor behavior in ASD
Our ndings that only TD controls showed changes in force control during proprioceptive feedback interference suggests that the ability to integrate sensory feedback information from multiple sensory modalities is de cient in ASD. Multisensory feedback integration during motor behavior involves modulating the weighting of feedback from separate sensory modalities to optimize motor output [51]. Vision is dominant for visually-guided upper limb and precision motor behaviors [29][30][31], though secondary sources also contribute to the re nement of behavioral output [52,53], consistent with our nding that TD controls showed increased force variability when proprioception was inaccurate. Individuals with ASD and TD controls showed similar changes in force variability when visual feedback was manipulated demonstrating that both groups used the primary feedback source during precision gripping. Our previous studies of a similar precision gripping test indicated that individuals with ASD show more severe deteriorations in their ability to limit variability of force output when visual feedback is perturbed, further supporting the hypothesis that they are highly reliant on visual input (i.e., the dominant source of sensory feedback) for precision gripping [14,54]. In the present study, individuals with ASD did not show elevations in force variability in ASD that varied as a function of visual gain, perhaps re ecting the narrower range of visual gains and ages studied here relative to our prior work [14].
Our ndings of decreased integration of non-primary sensory feedback processes in ASD is consistent with prior studies of separate sensorimotor behaviors. For example, a study of postural control in ASD documented an over-reliance on proprioceptive feedback, which is the dominant sensory input for maintaining postural stability [55]. Speci cally, Morris et al. [13] showed that disrupting proprioceptive feedback resulted in increased center of pressure (COP) variability in individuals with ASD regardless of whether visual feedback was available; however, TD controls only showed increased COP variability when both visual and proprioceptive feedback were disrupted. These results suggest that TD controls were able to compensate for disrupted proprioceptive feedback by up-weighting secondary sources of feedback (e.g., visual), whereas individuals with ASD continued to rely on the primary source of feedback (proprioceptive) even though it was unreliable. Combined with our ndings, these results indicate that, individually, visual and proprioceptive feedback mechanisms are relatively intact in ASD, but the ability to integrate and optimally weight feedback across multiple sensory modalities during motor behavior is compromised.
Motor learning studies also have demonstrated that persons with ASD are better at adapting to induced proprioceptive errors than TD controls during upper limb reaching, but they were less effective at adapting to visually induced errors [6,16,17]. On the surface, these studies seemingly contradict our nding that participants with ASD were under-reliant on proprioceptive feedback. However, the prior motor learning studies assessed adaptation in response to external sensory perturbations, which is a fundamentally different behavioral process than monitoring and adjusting ongoing behavior during precision grip force and likely requires a different weighting of sensory feedback inputs. These studies provide evidence that de cits across diverse sensorimotor behaviors in persons with ASD re ect atypical weighting of sensory inputs and an inability to integrate multiple sources of feedback.

Development of sensorimotor control in ASD
We found that individuals with ASD show stronger age-associated gains in precision force control (decreased variability, increased entropy) relative to TD peers across all visual gain and tendon vibrator conditions. These results indicate that the development of precision sensorimotor control is delayed in ASD, and that sensorimotor de cits (increased SD, reduced entropy) may represent important markers of neurodevelopmental dysfunction in childhood. Our ndings are consistent with considerable evidence from infant sibling and early childhood studies that show sensorimotor de cits are some of the earliest indicators of ASD [56, 57] and may be most severe during the rst years of life. While our data suggest sensorimotor impairments may be attenuated or even normalize by adolescence/early adulthood in ASD, their disruption early in life likely interferes with the maturation of cognitive, social, and language processes that are known to rely on early ontological progression of reaching and grasping behaviors [58][59][60][61]. Tracking the early childhood development of precision manual variability and regularity will be an important next step in characterizing key behavioral indicators of ASD, and in de ning neurodevelopmental mechanisms contributing to the range of clinical issues associated with ASD.
We also found that differences between individuals with ASD and TD peers in age-associated gains in force control varied across sensory feedback conditions suggesting distinct timing of separate sensory feedback control mechanisms. More speci cally, age-related gains in motor variability (decreases) and irregularity (increases) were stronger in the ASD group during conditions in which visual feedback was most precise (higher gains). These ndings are consistent with prior studies of normative development showing that while motor variability decreases and entropy increases with age, the rates and timing of these changes are dependent on the quality and nature of sensory feedback [62-64]. For example, no age-associated differences are seen in precision grip force variability and entropy across childhood and into adulthood (ages 6-22 years) when visual feedback is occluded, suggesting the ability to dynamically and precisely adjust motor behavior in response to sensory feedback improves with age due, at least in part, to a greater capacity to integrate multiple sensory inputs [62-64]. The stronger age-related improvements in force control that we observed in the ASD group relative to the control group suggest delayed maturation of sensory feedback processing for re ning motor output. Unlike controls, age-related decreases in force regularity in the ASD group were similar across proprioceptive feedback conditions indicating age-related improvements in the ASD group were dependent on the ability to utilize the dominant (visual) source of sensory feedback rather than the integration of multiple sensory modalities.
The age-associations observed in the present study differ from our prior precision gripping study, which found that TD individuals show greater improvements in motor regularity with age than individuals with ASD [14]. These opposing trends may be due to the age distributions in the samples. The prior study (range: 5-35 years, median: 13 years) likely captured a period of rapid maturation in TD children that also may represent an epoch of relatively slowed sensorimotor development in ASD. The present study restricted the age distribution to later childhood and early adulthood (range: 10-20 years, median 13.6 years) during a period in which typical motor development is relatively stable. The present ndings, in addition to studies showing that motor de cits in ASD are more severe in early childhood and improve over the course of adolescence [65,66], indicate that individuals with ASD experience a delayed trajectory of motor development.
Implications for understanding neurodevelopmental processes associated with ASD Our ndings of sensorimotor impairment in ASD and reduced integration of multisensory feedback implicate dysfunction of cortical-cerebellar sensorimotor networks. Posterior parietal cortex, including superior and inferior parietal lobules, integrate multiple sensory inputs during motor behavior [67-69] and innervate premotor and primary motor cortices to generate reactive motor adjustments based on feedback error information [70][71][72]. Parietal-cerebellar circuits also form a faster subcortical pathway for translating sensory error information into corrective motor commands relayed to motor cortex [73,74].
During motor behavior, cerebellar circuits critically compare the expected sensory consequences of motor output (received from primary motor cortex) to the actual consequences of the behavior (processed initially by primary and association sensory cortex) to correct errors in the motor command, which are relayed to the primary motor cortex though the thalamus [75,76]. Our ndings that persons with ASD relied almost exclusively on visual feedback during precision motor control suggest de cits in parietalcerebellar networks that are responsible for integrating feedback from multiple sources to accurately update motor commands. Additionally, stronger age-related improvements in force regularity at higher visual gains in the ASD group suggest delayed development of cortical-cerebellar circuits involved in rapid visual feedback processing and feedback error processing.
Our prior fMRI studies have found that increased motor variability and regularity in ASD during precision gripping are associated with increased activation and functional connectivity of cerebellar-parietal networks and decreased activation and functional connectivity of intra-cerebellar networks [23,24]. These prior studies also showed that increased force variability and regularity in ASD are associated with reduced activation and functional connectivity of frontal-parietal networks involved in the executive control of movement. Speci cally, persons with ASD showed increased activation of putamen and cerebellum relative to TD controls during precision gripping behavior, indicating greater reliance on subcortical sensorimotor processes [23]. Unlike controls, individuals with ASD showed no association between force variability and premotor activation, indicating that they do not modulate cortical motor planning circuits in response to sensory feedback [23]. At rest, individuals with ASD showed increased functional connectivity between cerebellum and superior occipital and parietal gyri, which are involved in visual and sensorimotor processing [24]. Persons with ASD also showed reduced resting functional connectivity relative to TD controls between cerebellum and frontal (superior and medial frontal gyri) and temporal (Heschl's and superior temporal gyri) cortices, which are involved in cognitive and multisensory processing [24]. An independent study similarly found increased intrinsic functional connectivity between cerebellum and sensorimotor regions of cortex (superior temporal, primary somatosensory, pre/primary motor, and occipital) and reduced intrinsic functional connectivity between cerebellum and cognitive regions of cortex (prefrontal, superior frontal, anterior cingulate, medial temporal gyrus), indicating that persons with ASD rely on basic sensory processing rather than complex multisensory or executive processing for sensorimotor control [77]. The only known prior fMRI study of precision visuomotor behavior found reduced activation of cerebellum as well as parietal and frontal eye elds, but increased activation of prefrontal-striatal-thalamocortical circuits suggesting increased reliance on non-motor regions during visuomotor control [78]. These ndings implicate reorganization of cortical and subcortical sensorimotor networks in persons with ASD potentially resulting from delayed maturation and specialization.

Sensorimotor behavior and clinical impairments
We found that force variability and regularity explained 9 to 15% of variability in clinically rated ASD symptom severity suggesting that sensorimotor feedback de cits may contribute to core symptoms or share common developmental pathways. For example, learning and interpreting social gestures requires early advances in sensorimotor behavior that facilitate both action representations, imitation, and reciprocal social interactions. More speci cally, early developing sensorimotor processes involve integration of visual information regarding the timing and intention of others' movement and mapping this information onto internal sensorimotor representations to estimate the expected visual and somatosensory consequences of the movement [79,80]. Di culties integrating visual and proprioceptive feedback for motor control in ASD may not only impact self-generated movements, including socially relevant behaviors, but also compromise the developing child's ability to interpret and predict others' behaviors [81]. Further, our ndings that more severe force control impairments in ASD are associated with clinical measures of motor ability indicate de cits of multisensory feedback control may contribute to functional motor issues in ASD.

Limitations and Future Directions
Several limitations of the present study should be noted. First, the inclusion of younger children in future studies will be important for characterizing key epochs of sensorimotor dysmaturation in ASD. Second, while our ndings of under-reliance on proprioceptive feedback for precision gripping and prior ndings of over-reliance on proprioception during postural control each suggest reduced integration of nonprimary sensory inputs during motor behavior in ASD, studies testing manipulations of multiple sensory inputs across multiple behaviors are needed to further clarify sensory feedback mechanisms of distinct behavioral impairments in ASD. Additionally, our study did not include a sham vibration condition (i.e., vibration on at a frequency that does not induce a proprioceptive illusion), so it is possible that our nding that persons with ASD were not affected by altered proprioception may re ect a reduced susceptibility to the proprioceptive illusion rather than reduced reliance on proprioceptive feedback.

Conclusions
The present study demonstrates that individuals with ASD show a reduced ability to integrate proprioceptive feedback during visually guided manual motor behavior implicating de cits integrating multiple sources of sensory feedback to guide precision motor behavior. We also show evidence for atypical development of sensorimotor abilities in ASD characterized by delayed maturation of precision sensorimotor control. These results help clarify the sensory feedback processes contributing to de cits in online motor control in individuals with ASD and provide new insights into important neurodevelopmental processes that contribute to the disorder. Figure 1 Task Design. A) Participants rest their arm on a custom arm rest with a tendon vibrator secured to their wrist with a Velcro strap. They place their thumb and index nger on the load cells of the force transducer. The tendon vibrator is either turned on to induce a proprioceptive illusion, or it is turned off so that there is no proprioceptive illusion. B) Participants view two bars on the computer screen. Participant force output is represented by the white bar, which moves up with increased force. The target bar is red during rest periods, and it turns green to indicate the start of the trial. Participants are instructed to press on the force transducers as quickly as possible when the target bar turns green and try to keep the white force bar at the same level as the green target bar. The gain of the visual feedback is presented at three different gain levels, such that the white force bar moves more per Newton of force at higher gain levels. At rest, the force output bar is at the 0N position, which changes as a function of the gain condition (shown here at medium gain).

Figure 3
Condition Effects on Force variability. Effects of tendon vibration (Off: dark, On: light) and gain level on the log of force SD for the ASD (red circles) and TD (blue triangles) groups. Error bars represent standard error.
Page 26/27 Figure 4 Force regularity vs. Age. Age associations with the force SampEn for the ASD (red circles) and TD (blue triangles) groups. Columns represent tendon vibration off (left) and on (right). Rows represent low (top), medium (middle) and high (bottom) gain levels. Age is in years. Shaded areas represent the 95% con dence intervals.