Online control of prehension predicts performance on a standardised motor assessment test in 8-12 year-old children

Goal-directed hand movements are guided by sensory information and may be adjusted ‘online’, during the movement. If the target of a movement unexpectedly changes position, trajectory corrections can be initiated in as little as 100ms in adults. This rapid visual online control is impaired in children with developmental coordination disorder (DCD), and potentially in other neurodevelopmental conditions. We investigated the visual control of hand movements in children in a ‘centre-out’ double-step reaching and grasping task, and examined how parameters of this visuomotor control co-vary with performance on standardised motor tests often used with typically and atypically developing children. Two groups of children aged 8-12 years were asked to reach and grasp an illuminated central ball on a vertically oriented board. On a proportion of trials, and at movement onset, the illumination switched unpredictably to one of four other balls in a centre-out configuration (left, right, up, or down). When the target moved, all but one of the children were able to correct their movements before reaching the initial target, at least on some trials, but the latencies to initiate these corrections were longer than those typically reported in the adult literature, ranging from 211 to 581 ms. These later corrections may be due to less developed motor skills in children, or to the increased cognitive and biomechanical complexity of switching movements in four directions. In the first group (n=187), reaching and grasping parameters significantly predicted standardised movement scores on the MABC-2, most strongly for the aiming and catching component. In the second group (n=85), these same parameters did not significantly predict scores on the DCDQ-07 parent questionnaire. Our reaching and grasping task provides a sensitive and continuous measure of movement skill that predicts scores on standardized movement tasks used to screen for DCD.

trajectory corrections can be initiated in as little as 100ms in adults. This rapid visual online 23 control is impaired in children with developmental coordination disorder (DCD), and 24 potentially in other neurodevelopmental conditions. We investigated the visual control of 25 hand movements in children in a 'centre-out' double-step reaching and grasping task, and 26 examined how parameters of this visuomotor control co-vary with performance on 27 standardised motor tests often used with typically and atypically developing children. Two 28 groups of children aged 8-12 years were asked to reach and grasp an illuminated central ball 29 on a vertically oriented board. On a proportion of trials, and at movement onset, the 30 illumination switched unpredictably to one of four other balls in a centre-out configuration 31 (left, right, up, or down). When the target moved, all but one of the children were able to 32 correct their movements before reaching the initial target, at least on some trials, but the 33 latencies to initiate these corrections were longer than those typically reported in the adult 34 literature, ranging from 211 to 581 ms. These later corrections may be due to less developed 35

Introduction 44
Almost from the moment able-bodied people wake up, they begin reaching and grasping for 45 objects with their handsbed covers, a cup of coffee, a toothbrush. Coordinating and 46 controlling accurate, goal-directed reaching and grasping movements is done many times a 47 DCDQ'07 (Wilson et al., 2000(Wilson et al., , 2009) -The DCDQ'07 is a brief parent questionnaire 183 designed to screen for motor problems associated with DCD in children aged 5 to 15 years. 184 Parents are asked to compare their child's motor performance to that of his/her peers 185 depending on the child age band (5:0-7:11, 8:0-9:11, 10:0-15:0). The DCDQ'07 consists of 'perturbed' target conditions, this distribution is not universal (e.g., Prablanc and Martin, 1992, used 33% unperturbed and 67% perturbed), and neither perturbation probability nor 211 perturbation expectation have strong effects on the latency to initiate movement corrections 212 (Cameron et al., 2013). The hand position was analysed online, and the target change 213 occurred as soon as the tangential velocity of the thumb reached 15 cm/s. This criterion was 214 subsequently changed (after the first 48 children), to a velocity towards the central target of 215 15 cm/s in order to counter the strategy of some children opting to make a short initial 216 movement (e.g., upwards or sideways), before making a second movement towards the 217 target location, which may since have changed. The balls in the up, down, left, and right 218 positions each lit up on 10% (12.5%) of the trials in a pseudorandom order. Children were 219 instructed to start each trial with their thumb and index fingers closed in a pincer grip and 220 placed on the starting point, then to reach and grasp the illuminated ball as accurately and 221 as quickly as possible, but in a controlled manneras natural a reaching movement as 222 possible. Children were instructed to interrupt their movement to the central ball and grasp 223 instead the eccentric ball when the illumination switched locations. There were 10 practice 224 trials before the main data collection to familiarise the children with the task and this was 225 followed by one testing block of 40 trials. Motion trackers were attached over the thumb and 226 index fingers (i.e., the grasp 'opposition axis', Holt et al., 2013) of a 'NASA' astronaut's glove 227 (this did not appear to affect children's hand movements, see also Hyde and Wilson, 2011b), 228 to record the position (3 degrees of freedom) of these digits with a Polhemus Fastrack 229 (Polhemus, Colchester, VT, USA) magnetic tracking system. The system has a spatial 230 accuracy of 0.08 cm, and a precision of 0.0055 cm (for the average location sampled in the 231 current study), sampling the two receivers, each at 60 Hz. We opted for two trackers 232 sampling at 60Hz as the ideal trade-off between trackers (1-4) and frequency (120-30Hz) -233 an additional third tracker on the wrist would have entailed a reduction of sampling frequency 234 to 40Hz. Since human hand and finger movements cannot move or oscillate at much more 235 than 30 Hz (Raethjen et al., 2000), and the visual online control of movement takes a 236 minimum of 100 ms, 60 Hz sampling is more than adequate to capture the relevant 237 information required to test our hypotheses. 238 Cognitive assessments -Children in the first, MABC-2 (Age band 2, 7-10 years old), group 240 were assessed with the Reading, Verbal Similarities, and Matrices tests of the British Ability 241 Scales (BAS) (Elliot, 1996), and the Conners 3-AI (Conners, 2008). Children in the second, 242 DCDQ'07, group were assessed with the British Picture Vocabulary Scale (Dunn et al., 243 1997)

Data analysis 259
The experiments were run and the data analysis was performed using Matlab (Mathworks, 260 Natick, USA), and SPSS for the factor analysis. All the programs and all raw data are or will 261 be freely available from the last author or his website (http://neurobiography.info/). All data 262 analysis was fully automated and scripted, using procedures developed during previous 263 work (for full methods and discussion, see Holmes and Dakwar, 2015). A summary of the 264 analytical approach is provided here. 265 Raw data -Six degree-of-freedom position and orientation data from the index finger and 267 thumb were acquired at 60 Hz for 2 s per trial. Data were re-sampled to 120 Hz then filtered 268 with a 2 nd order, zero-lag (dual-pass) Butterworth filter with a 15 Hz low-pass cut-off. Two raw 269 data channels (index finger and thumb), as well as the mean (used for many kinematic are provided for all variables in supplementary data. All raw and summary data were 300 inspected visually, in order to set criteria and adjust analytic parameters and procedures. 301 The final analysis is fully-automated and repeatable. The only human intervention in the final 302 analysis was to exclude two clear outlying participants, following plotting of the factor 303 analysis datafactor analysis is sensitive to outliers (Flora et al., 2012). To visualise the data, the velocities, accelerations, and jerks across trials in the same 324 condition were averaged by aligning the movement onsets. Each trial was also resampled to 325 120 data points, from RT-5 to MT+5 samples. These resampled, standardised, data were 326 then re-scaled to a maximum height of 1, averaged across conditions per participant, and re-327 scaled again to a maximum of 1. This re-scaling compensates for between-participant 328 differences in movement velocity, duration, and variability. The final average movement 329 profiles ( Figure 4) are then useful to assess the overall 'quality' or 'shape' of movement.  likely also reach peak velocity early; higher acceleration leads to higher velocity; these 346 parameters are correlated. Rather than examine a series of kinematic parameters 347 independently, reducing these highly-correlated variables to a smaller number of more 348 independent factors helps resolve problems with multiple comparisons across different 349 dependent variables. We extracted 87 reaching and grasping parameters from each of 262 participants who had valid reaching and grasping data, and reduced this to 17 factors using 351 principal components analysis in SPSS 21 with oblique (direct oblimin) factor rotation in 352 order to minimise the number of variables loading heavily onto each factor. A criterion of 353 eigenvalues >1 was used for factor selection; factor scores were estimated using Bartlett's 354 method. While researchers may disagree over whether and when to use orthogonal or 355 oblique factor rotation, the underlying mathematics is identical, the total variance explained 356 remains the same, and only with criteria external to the factor analysis itself can the 357 usefulness of any particular rotation method be judged (Comrey & Lee, 1992). We assessed Eighty-seven variables derived from the kinematic data were entered into an exploratory 435 factor analysis with oblique factor rotation. Seventeen resulting factors had eigenvalues >1, 436 accounting for 85.7% of the variance (Table 1). The first three factors had loadings of ≥0.3 437 on 50, 40, and 31 original variables respectively, and as such were hard to describe, but 438 likely account for general between-participants' differences in movement speed and 439 variability, or body size, which affects multiple variables. The remaining 14 factors loaded 440 strongly onto between 0 and 18 original variables. An attempt to describe these factors is 441 presented in Table 1, along with the correlation between scores from each factor and the  Table 1, and the whole model fits in Figure 5.

Group analyses 486
The relationships between reaching and grasping and standardised movement coordination 487 scores seem to be continuous, rather than containing any discontinuities at particular scores 488 or ranges. Nevertheless, following a reviewer's request, the continuum was divided into 489 discrete groups on the basis of both clinical diagnoses (e.g., DCD diagnosis) and the MABC-490 2 and DCDQ'07 scores relating to clinically significant cut-offs. In our sample, 11 children 491 had formal diagnoses of DCD; 25 children (13.8% of our sample) were at or below the 5 th 492 percentile of the MABC-2; 35 (19.3%) were between the 6 th and 16 th percentile inclusive; and 493 121 (66.8%) had scores above the 16 th percentile. For the DCDQ'07, a large proportion 494 (31.7%) of parents rated their children as having movement coordination below the cut-off. 495 These groups were compared on factors 6 and 15 from the factor analysis, and on the model 496 prediction scores (summary data in the supplementary table). 497 498 Children at or below the 5 th percentile on the MABC-2 had significantly different (p≤.025, 1-499 tailed, correcting for 2 comparisons) scores from children above the 16 th percentile on both 500 factors 6, t139=3.08, p=.001, and 15, t139=-2.37, p=.01, and children between the 6 th and 16 th 501 percentiles inclusive also differed from the >16 th percentile group on factor 6 (t151=2.19, 502 p=.015), but not factor 15, t151=-1.25, p=.11. Factor scores of the groups at or below the 5 th 503 and between the 6 th and 16 th inclusive did not differ significantly. Regarding the linear model 504 predictions of the MABC-2 component scores and total scores, the same pattern was found, 505 with the two lower-scoring MABC-2 groups differing significantly (p≤.0125, 1-tailed, 506 correcting for 4 comparisons) from the higher-scoring group on manual dexterity, aiming and 507 catching, and total scores, while only the comparison between the ≤5 th percentile group and 508 the >16 th percentile group was significant for the balance scores. All the differences were in 509 the expected directions, which is not surprising as the models were set up to predict these 510 scoresdividing the range into bins and re-testing is a statistical 'double-dip'. There was no 511 evidence for significant differences between the 11 children with a formal diagnosis of DCD 512 and the rest of the sample, either on factor 6, t255=0.663, p=.51 or factor 15, t255=-1.26, 513 p=.21, or on the aiming and catching, balance, or total scores (|t174|s<1.68, ps>.095. Again, 514 this may not be surprising, as the model was set up to predict MABC-2 scores, rather than 515 DCD diagnosis. Factor 12, however, did show a relatively large difference between children 516 with DCD and those without, t255=-2.72, p=.007the 11 children with DCD had larger, earlier 517 peak grip apertures, and made more additional acceleration on perturbed trials, as 518 compared to children without DCD. 519 520 By contrast to the MABC-2, children with low versus high parent ratings of movement 521 coordination (DCDQ'07) did not differ significantly in their factor scores, although the 522 direction of effects was equivalent to those in the MABC-2 groups. Finally, while responding 523 to reviewers' comments, we discovered several significant differences in the factor scores for 524 participants who performed the task in the dark versus in the light, who used their dominant 525 versus their non-dominant hand to reach, and based on their gender. Analysis of these 526 categorical variables, along with age and other participant-specific predictors, is beyond the 527 current scope and will be dealt with fully elsewhere (Blanchard et al., in preparation). Performance of our reaching and grasping task requires accurate planning, generation, and 541 visual online control of reach-to-grasp movements, including the coordination of reaching 542 and grasping phases. From our results, the strongest predictor of MABC-2 scores (especially 543 the aiming and catching component) was factor 6, which loaded heavily on measures of the 544 additional acceleration and jerk on perturbed compared to unperturbed trials, movement path, curvature, the latency to initiate movement corrections, and peak grip aperture. This 546 factor may represent the key sensorimotor processes required in visual online control. 547 Following a change in target location, the ideal movement correction would comprise a 548 change of direction towards the new target, but without an overall increase in movement 549 speed (i.e., no additional acceleration or jerk), and with a minimum overall increase in 550 movement path length and duration. Efficient corrections will thus have lower overall jerk, 551 path, movement time, curvature, and correction latency. Factors 6 and 15 also loaded on the 552 variability and relative timing of peak grip aperture. An ideal correction to the reaching 553 component of the movement should not also require a correction to the grasping component. 554 Children who correct their reaching movement optimally would not need to adjust their 555 grasping movement -the time to peak grasp aperture could stay relatively constant relative 556 to overall movement time. By contrast, children who fail to adjust their reaching movement 557 efficiently might close their grasp onto the central target, then require an additional opening 558 of the grasp for the peripheral target. On some trials, the initial grasp will be detected as the 559 peak grip aperture, and on other trials peak aperture will occur on the second grasp. This 560 double-grasping movement leads to greater variability in the measured relative time of peak 561 grip aperture. Our result echoes an earlier finding in which children with DCD showed a 562 much greater variability in grasp timing than typically developing children (Astill and Utley, 563 2008). The authors of this previous study suggested that children with coordination disorders 564 may use a decomposition strategy to simplify the control of transport and grasp phases of 565 catching by uncoupling these movement components. 566 567 While aiming and catching scores were best-predicted by the reaching and grasping factors 568 (16% of variance in the MABC-2 explained), manual dexterity, and to a lesser extent balance 569 scores, were also significantly predicted by reaching and grasping, with 10% and 4% of 570 variance explained, respectively. Because scores across the three components of the 571 MABC-2 are correlated (across 225 of our participants, manual dexterity component scores 572 correlate with aiming and catching, r223=.342, and balance, r223=.525; aiming and catching correlates with balance, r225=.389), factors which predict one of the components are also 574 likely to predict the others. This is likely due to general movement coordination ability, to 575 general cognitive, attentional, or motivational factors which are common to the movement 576 tasks, or to the fact that accurate control of the hands and arms also requires postural and 577 balance control, leading to functional links in development of these abilities (Flatters et al., 578 2014). 579

580
The relationship between kinematic factors and the aiming and catching component of the 581 MABC-2 (16% variance explained) was modest, given that, for example, the manual 582 dexterity and balance components shared 28% of variance in our dataset. Nevertheless, we 583 found no significant relationships at all between any of our kinematic factors and the 584 DCDQ'07 parent questionnaire. This negative finding suggests that parents' evaluations of 585 how their child's movement coordination ability compares with others' should be interpreted 586 cautiously. The DCDQ'07 alone may be unlikely to measure movement coordination skill, at 587 least for reaching, grasping, aiming, and catching skills, although note that we did not 588 measure the DCDQ'07 and the MABC-2 in the same children. Second, correction latencies based on differences in total movement time, which confounds 620 correction latency with the post-correction movement time, were just 235 ms, more than 100 621 ms less than that of the individual trial-by-trial analysis. The additional movement time 622 following a movement correction will be lower in children who reach faster or straighter 623 overall, or who execute a faster correction movement. Indeed, the 107 ms difference 624 between our preferred measure of correction latency and the additional movement time 625 suggests that children increase their movement speed substantially after the target change, 626 'catching-up'. While our preferred correction latency measure was longest for the lower 627 target location, the additional movement time required was longest for the upper target 628 location. Moving the arm upwards probably requires more effort than moving downwards, so the post-correction movement direction may well influence overall movement time. 630 Correction latencies from different studies can only meaningfully be compared if identical 631 methods were used to measure them. 632

633
We chose to extract as many variables as possible from the reaching and grasping 634 movements in an attempt to fully capture the differences in movement between children and 635 conditions. With 87 extracted variables, the problem of multiple comparisons and collinearity 636 arises, which we addressed by reducing the data to 17 relatively independent factors (cf 637 Naish et al., 2013). An alternative, preferable, but computationally-expensive approach is to 638 fit a series of low-dimensional models to the raw velocity data, and to analyse only the model 639 parameters across participants and conditions. This analysis of 'sub-movements' is based on 640 the minimum-jerk model, and may account well for online movement corrections (Flash and 641 Henis, 1991). This approach, using constrained non-linear optimisation in Matlab, was 642 investigated for analysis of the current dataset. However, with 262 participants and 40 trials, 643 the computer processor time alone was likely to take several months! We will use this 644 technique for future work. 645 646

Continuous versus discrete groups of movement ability 647
Our approach to data analysis was continuous, in that we did not set out to create two 648 distinct groups consisting of children with DCD and TD children. Rather, we explored motor 649 abilities across the spectrum, eliminating the difficulties that arise when trying to categorise 650 DCD, which is well known for its heterogeneity (Zwicker et al., 2012). We have noted that 651 diagnosis of DCD is incomplete in the local population, and variable between groups of 652 children, for example from different schools or administrative areas. Furthermore, we found 653 that some children with a diagnosis of DCD performed perfectly well on the MABC-2. This 654 could be due either to the wrong diagnosis being made, an intervention having been 655 effective, developmental improvements since diagnosis, or to the inadequacy of the MABC-2 656 as a diagnostic instrument (Venetsanou et al., 2011). In the absence of a diagnosis, then, any division of continuous MABC-2 data into discrete clinical (i.e., ≤ 5 th percentile) or pre-658 clinical (i.e. 6 th < percentile ≤ 16th) categories is arbitrary, and, we suggest, likely to obscure 659 the underlying, probably continuous, relationships between individual movement parameters 660 and performance on the MABC-2. We have also noted that ceiling effects and the non- Our results support the interpretation that impaired visual online control is a strong predictor 668 of performance on standard tests of movement ability, as are often used to diagnose 669 developmental movement disorders. The visual online control task developed for this study 670 provides a continuous and high-resolution measurement, and is directly comparable 671 between adults and children, which makes it a promising task for further study. The present 672 results show that children who are poor at aiming and catching are also particularly poor at 673 the online control of reaching and grasping.  with 95% confidence intervals. Reaching and grasping explains 17% of the variance in 870 aiming and catching scores, 16% of variance in the total scores, 10% of the manual 871 dexterity, and 4% of the balance score variance.