Proprioceptive Deficits in Inactive Older Adults are not Reflected in Discrete Reaching Performance

During normal healthy ageing there is a decline in the ability to control simple movements, characterised by increased reaction times, movement durations and variability. There is also growing evidence of age-related proprioceptive loss which may contribute to these impairments. However this relationship has not been studied in detail for the upper limb. We recruited 20 younger adults (YAs) and 31 older adults (OAs) who each performed 2 tasks on a 2D robotic manipulandum. The first assessed dynamic proprioceptive acuity using active, multi-joint movements towards visually presented targets, with movement constrained by the robot to a predefined path. Participants made perceptual judgements of the lateral position of the unseen arm. The second was a rapid motor task which required fast, accurate movements to the same targets in the absence of hand position visual feedback, and without constraint by the robot. We predicted that the variable proprioceptive error (uncertainty range) from Task 1 would be increased in physically inactive OAs and would predict increased movement variability in Task 2. Instead we found that physically inactive OAs had larger systematic proprioceptive errors (bias). Neither proprioceptive acuity nor bias was related to motor performance in either age group. We suggest that previously reported estimates of proprioceptive decline with ageing may be exaggerated by task demands and that the extent of these deficits is unrelated to discrete, ballistic movement control. The relationship of dynamic proprioceptive acuity with movement control in tasks which emphasise online proprioceptive feedback for performance is still unclear and warrants further investigation.


Abstract 21
During normal healthy ageing there is a decline in the ability to control simple movements, 22 characterised by increased reaction times, movement durations and variability. There is also 23 growing evidence of age-related proprioceptive loss which may contribute to these 24 impairments. However this relationship has not been studied in detail for the upper limb. 25 We recruited 20 younger adults (YAs) and 31 older adults (OAs) who each performed 2 tasks 26 on a 2D robotic manipulandum. The first assessed dynamic proprioceptive acuity using 27 active, multi-joint movements towards visually presented targets, with movement 28 constrained by the robot to a predefined path. Participants made perceptual judgements of 29 the lateral position of the unseen arm. The second was a rapid motor task which required 30 fast, accurate movements to the same targets in the absence of hand position visual 31 feedback, and without constraint by the robot. We predicted that the variable 32 proprioceptive error (uncertainty range) from Task 1 would be increased in physically 33 inactive OAs and would predict increased movement variability in Task 2. Instead we found 34 that physically inactive OAs had larger systematic proprioceptive errors (bias). Neither 35 proprioceptive acuity nor bias was related to motor performance in either age group. We 36 suggest that previously reported estimates of proprioceptive decline with ageing may be 37 exaggerated by task demands and that the extent of these deficits is unrelated to discrete, 38 ballistic movement control. The relationship of dynamic proprioceptive acuity with 39

Introduction 43
As we get older there is a general decline in motor system physiology which affects the 44 ability to perform simple movements. This includes degradation of musculature through loss 45 and remodelling of muscle motor units (Lexell, 1995 frequently observed in this population (Helsen et al., 2016;Ketcham et al., 2002). 59 In addition to motor physiology, loss of proprioception has also been suggested as a 60 contributing factor to the presentation of these age-related motor deficits. Specifically, 61 there is growing evidence to show decline of this sensation through a range of different 62 measurement techniques (see Goble to make active, multi-joint reaching movements constrained to a tight, pre-defined 105 trajectory, before making instantaneous judgements of their unseen limb relative to a 106 visually presented reference position. These two-alternative forced choice responses were 107 then gathered and used to estimate both systematic (bias) and variable (uncertainty range) 6 proprioceptive errors; only the latter showed age-related increase, with marginal statistical 109 significance. Variants of this task have been reported elsewhere (Cressman & Henriques,110 2009; Ostry, Darainy, Mattar, Wong, & Gribble, 2010), but this was the first report of its use 111 with an ageing population. Critically, since this type of task reduces dependence on working 112 memory and utilizes active movements, it may be more suited for the investigation of age-113 related proprioceptive loss and voluntary movement control. Moreover, if it is indeed the 114 case that proprioceptive uncertainty increases with ageing, then this elevated sensory noise 115 could make the sensory consequences of motor commands unpredictable (Miall & Wolpert, 116 1996) and thus lead to more variable movement characteristics, which are frequently 117 reported for the older adult population (Darling et al., 1989;Ketcham et al., 2002;Seidler et 118 al., 2002). As such, the proprioceptive uncertainty estimate derived from this type of task 119 makes for a compelling predictor of motor performance in the ageing population. 120 The aim of this experiment was therefore to assess, in groups of older and younger adults, 121 the extent to which dynamic, multi-joint proprioceptive acuity of the upper limb could 122 predict performance on a fast, targeted reaching movement task. We predicted that 123 physically inactive older adults would exhibit larger proprioceptive uncertainty ranges and 124 that this would predict greater variation in motor performance. Conversely, since a 125 systematic perceptual error (assessed as proprioceptive bias), may be easier to predict and 126 account for during motor control, we predicted bias would be unrelated to motor 127 performance for either age group. 128

Methods 130
Participants 131 Thirty one older adults (OAs) aged 65 years or older (11 male, 71.2 ± 4.5 yrs), and 20 132 younger adults (YAs) aged 18-25 years (11 male, 20.4 ± 2.0 yrs) participated in the 133 experiment after giving informed consent; the University of Birmingham ethics panel 134 approved the study. All participants were right-hand dominant as defined by a laterality 135 quotient of 30 or higher on the 10-item Edinburgh Handedness Inventory (Oldfield, 1971). 136 Participants were excluded if they had any history of neurological illness, or carpal tunnel 137 syndrome, arthritis or similar movement pains or limitations in the arm, wrist or fingers. OAs 138 also completed the Montreal Cognitive Assessment (MoCA) and were only included in the 139 analysis if they scored 26 or above out of 30, which is considered to indicate normal 140 cognitive functioning (Nasreddine et al., 2005). position were sampled at 1kHz with any applied forces updated at the same rate. In both 152 the dynamic proprioceptive and rapid motor reaching tasks, participants made reaching 153 movements from a white 1cm radius start position located 8cm into the workspace 154 (approximately 28cm from the participant's torso). Participants made reaching movements 155 to one of three positions, shown by a 1cm radius grey target, which were located 20cm from 156 the start position at 30°, 90° and 150° elevation ( Figure 1B). When made available, hand 157 position feedback was provided on a real-time basis by a 0.5cm radius white cursor that was 158 always spatially congruent with the vBOT handle. In all cases targets were presented in a 159 pseudorandomised order. 160

Experimental Design 161
All participants performed the dynamic proprioceptive task first. Hence there was no 162 possibility for the feedback associated with the rapid motor reaching task to alter or 163 improve proprioceptive acuity to the same spatially located targets. 164

Task 1: Dynamic Proprioception 165
Procedure 166 Participants made reaching movements towards 1 of the 3 targets with visual feedback of 167 hand position occluded throughout, and target position occluded after the initial 5cm 168 outward movement (see Figure 1C). These movements were constrained to a pre-defined 169 minimum jerk path using stiff virtual walls (see Ostry et al. 2010) that steered the hand  Workspace locations and relative distances of the 3 targets (T1-T3) used in both the dynamic proprioception and rapid motor tasks C. Illustration of minimum jerk channel for the dynamic proprioception task. At termination, a circle and square are displayed to prompt a verbal response ("Circle" would be correct in this example). Target is visible for first 5cm before it disappears for remainder of trial, hand positon cursor remains occluded for all channel trials in a given block D. Illustration of rapid reaching task. Visual feedback of hand position was occluded once the cursor left the home position and remained so for the entire trial. Coloured feedback was provided at the target location on trial termination to indicate the endpoint accuracy of the movement. Both the experimental tasks in C. and D. are performed at target T2 (T1 and T3 not shown)

PEST Sequences 182
The size and direction of the lateral deviation imposed by the virtual channels was dictated 183 by two randomly interleaved PEST sequences (Taylor & Creelman, 1967) spanning across all 184 TPV was expressed as a percentage of total MT (time between movement initiation and 245 termination) to examine the speed profile of the movement independently of its actual 246 duration. Accuracy was quantified both by the absolute error (AE) at endpoint (the 247 Euclidean distance from trial termination position to the target location) and by the lateral 248 deviation at endpoint (LE). LE was calculated as the orthogonal distance from the linear path 249 between start position and target, to endpoint and was included to improve the validity of 250 the association with the proprioceptive measures, which also use an orthogonal deviation 251 measure. Within participants variability in motor accuracy was assessed using the standard 252 deviation of the accuracy measure across trials for each participant, separately for each 253 target. 254 The rapid motor task was preceded by 9 practice trials (3 per target), with main task 255 performance consisting of 3 blocks of 20 trials such that there were a total of 20 movements 256 to each target. 257

Physical Activity Measures 258
Older Adults 259 After completing the experiment, OAs were given wrist-worn accelerometers (Philips 260 Actiwatch 2) to wear for 5 days (120 hours), where "activity counts" were logged in 30 261 second epochs. If an epoch had less than 40 counts it was deemed to be inactive 262 (intermediate activity threshold defined by Philips Actiware software version 6.0.2). 263 The sum of all counts in the surviving active epochs over the 5 days provided a physical 264 activity (PA) metric for each older participant. The median value of the scores between participants was then used as a threshold to define "Inactive" and "Active" sub-groups of 266 OAs for further analysis (demographic details for these groups are detailed in the Results 267 section). 268

Younger Adults 269
We were unable to use accelerometer data to sub-group the YA participants. Hence self-270 reported PA measures were recorded for YAs using the IPAQ-Short questionnaire (Craig et 271 al., 2003), with participants scoring in the highest "Health Enhancing Physical Activity" 272 category being excluded from participation, in order to decrease heterogeneity. 273

Working Memory 274
To test if working memory capacity influenced our proprioceptive measures, working 275 memory was measured before participation in the experiment by using the backward digit 276 span test, following previous reports of its use in proprioceptive ageing studies (Adamo et 277 al., 2009;Goble et al., 2012). In this task, participants were required to memorise a 278 sequence of random numbers (ranging 1-9; read out to them at a rate of approximately 1 279 number per second), and then recite them in reverse order. The task began with two trials 280 at a sequence length of 2. If participants could correctly recite the sequence on at least 1 281 out of the 2 attempts at that sequence length level, the sequence length would increase by 282 one. The task then incremented in this fashion until both attempted recitals were incorrect. 283 The highest sequence length which the participant could correctly recite at least 1 out of the 284 2 attempts was recorded as their verbal working memory score. 285 Statistical and Cross-Task Analysis 287 All data are presented as group means ± standard deviation unless otherwise stated, with 288 values greater than 2.5 standard deviations away from the group mean at each target 289 removed as outliers (approximately 5% of data). The remaining data were analysed in 290 separate 3 x 3 mixed-design ANOVAs, with a between subjects factor of Group (inactive 291 OAs, active OAs and YAs) and repeated measure of Target (T1-T3). A Greenhouse-Geisser 292 correction was used in all cases where the sphericity assumption was violated, and 293 significance was assessed at the α < .050 level. Statistically significant ANOVA effects and 294 interactions were followed up with post-hoc t-test pairwise comparisons, and assessed for 295 significance using a False Discovery Rate (FDR) analysis (Benjamini & Hochberg, 1995 To assess the relationship between motor performance and proprioceptive acuity, a series 307 of linear regression models were calculated. Since proprioceptive judgements were made along an axis orthogonal to the start-target vector, we assume that if either measure was 309 related to motor control this would be most apparent with motor errors along a similar 310 orthogonal axis. Thus, average lateral error (LE) and within-subject variation of LE (LE Var) 311 were chosen as the motor performance measures to include in the regression models. 312 Specifically, we hypothesize that proprioceptive noise could predict motor accuracy 313 variation and so used uncertainty range to predict LE Var. We then examined the 314 association between systematic proprioceptive and motor errors by using bias to predict LE. 315 PA level was used as an additional predictor in the models which allowed us to collapse data 316 across the inactive and active OA groups. Separate regression models were calculated for 317 each of the 2 proprioceptive-motor relationships of interest for both OAs and YAs 318 separately, with an FDR-adjusted α-threshold used to control for multiple tests. 319 Results 320 Physical Activity Grouping 321 The 31 OAs were divided into either a physically inactive or physically active sub-group 322 according to a threshold median value of 1.68 x 10 6 activity counts from the 5-day 323 accelerometer data. This left 16 OAs in the inactive group (1.29 ± .31 x 10 6 counts; 7 male, 324 72.9 ± 5.1 yrs) and 15 in the active group (1.96 ± .26 x 10 6 counts; 4 male, 69.3 ± 2.7 yrs). 325 The inactive group were found to be significantly older than the active group (t [ Activity Grouping) we correlated age and bias (averaged across all 3 targets) for the entire 339 OA sample. The correlation was non-significant (r = .005, p = .977) and we conclude that the 340 group effect on bias is indeed due to the physical inactivity of OAs. 341 Contrary to our predictions, there was no effect of Group on uncertainty range (F[2, 45] = 342 .31, p = .733). There was an overall effect of Target (F[2, 90] = 4.8, p = .011, η 2 p = .10), such 343 that uncertainty range was larger at T3 than T2 (t[47] = -2.9, p = .006; αFDR = .017). There 344 was no Group x Target interaction (F[4, 90] = .51, p = .730). 345

Kinematic Measures 349
Due to an unforeseen technical error, for 4 OAs in the physically inactive group we had only 350 partial kinematic data which was non-analysable; the perceptual judgement data remained 351 valid for all participants. For this reason kinematic data here was analysed as n = 12 for 352 inactive OAs; the perceptual data for this sub-group did not differ from the others, tested 353 with a mixed-ANOVA between the excluded and retained participants (bias p = .99, 354 uncertainty range p = .16). YAs made the fastest movements (20.2 ± 5.9 cm/sec) followed by 355 active OAs (16.1 ± 4.7 cm/sec) and inactive OAs who moved slowest (14.6 ± 5.4 cm/sec).  Results for uncertainty range where there were no significant differences observed between any of the 3 groups Movement speed might influence perceptual performance in this task since the lateral 363 acceleration through channel deviation ( Figure 1C) would be greater for faster movements. 364 We therefore tested if bias and uncertainty range were correlated with average movement 365 velocity for each of the 3 different groups. We found that none of the correlations were 366 significant for the bias (|r| < .34, pmin = .045; αFDR = .017); however the inactive OAs showed 367 a significant, positive correlation between average movement velocity and uncertainty 368 range (r = .46, p = .008; αFDR = .017; all others |r| < .31) indicating faster movements were 369 related to lower perceptual acuity. There were no significant relationships observed 370 between bias and mean force exerted against the final section of the channel wall for any of 371 the 3 groups (|r| < .294, pmin = .096; αFDR = .017). This shows that systematic perceptual 372 errors were independent of direction of effort exerted during the verbal reporting stage. 373

Rapid Motor Reaching Performance 374
Performance Accuracy Measures 375 Results for the LE and LE Var motor accuracy measures are shown in Figure 3A and 3B 376 respectively. All motor accuracy data (LE and AE parameters) are shown in Table 1 There was also no effect of Group on AE (F[2, 44] = 1.  between groups. This therefore shows endpoint accuracy in this motor task was maintained 395 with advanced age, and was independent of PA. 396 Since participants were provided with accuracy feedback during the motor task, an 397 additional ANOVA was performed on the accuracy measures in the early vs. late parts of the 398 task (first vs. last 10 trials) to assess whether any motor learning occurred. We focus on, and 399 report only, the factors of Time (early or late in the task) and Group x Time interaction (all p > .050). This shows that although there were improvements in performance over the 406 duration of the task, the extent of these improvements did not differ between the 3 groups. 407

Kinematic Performance Measures 411
The data for RT and PV are summarised in Figure 4A and 4B respectively, with all kinematic 412 measures for the rapid motor task shown in Table 2 1 (± 3.2) ***51.2 (± 3.9)   Together, the results from these kinematic measures shows that there were target-specific 445 common kinematic features across all three groups, but overall, the OAs tend to react and 446 move more slowly than YAs, regardless of their PA level. However, the shape of velocity 447 profiles of movements were similar between all groups. 448

Speed-Accuracy Trade-off 449
Since there were significant differences in peak hand velocity between older and younger 450 groups, we wanted to test for a potential speed-accuracy trade-off. We therefore divided 451 both LE and AE values by corresponding PV on a trial-by-trial basis to create lateral and 452 absolute error indices controlled for movement speed (LEPVCont and AEPVCont respectively), 453 then analysed by 3 x 3 mixed-design ANOVAs: (Group) x (Target), as above. 454 There was no effect of Group on LEPVCont (F[2, 46] = .19, p = .826) but the main effect of 455 Target was significant (F[1.6, 73.7] = 58.1, p < .001, η 2 p = .56; see Figure 5A Collectively, this additional analysis of the speed-accuracy trade-off shows that the 474 maintenance of absolute endpoint accuracy in OAs may be partially explained by movement 475 Figure 5 -Group average motor accuracy measures controlled for by peak hand velocity (means ± standard error). A. Lateral error divided by peak hand velocity (LEPVCont) where more positive values represent errors to the clockwise (or "Circle" from proprioceptive task) side. B. Absolute errors divided by peak hand velocity (AEPVCont). Pairwise comparisons which were significant (p < .05) but did not survive corrections for multiple comparisons are indicated by † slowing. However, the lateral errors appear to be similar between age groups even when 476 controlling for movement speed, suggesting they may be less susceptible to a speed-477 accuracy trade-off in this context. 478 Working Memory Capacity 479 All groups had similar working memory capacity scores, as indicated by a non-significant 480 one-way ANOVA (F[2, 48] = .16, p = .854). YAs had the highest score (5.8 ± 1.6 numbers 481 recalled) followed by active OAs (5.7 ± 1.4) and inactive OAs with the lowest score (5.5 ± 482 1.3). To test if working memory was related to proprioceptive performance, we correlated 483 the bias and uncertainty range, averaged across all 3 targets, with working memory score. 484 There were no significant relationships found (all |r| < .38, pmin = .106; αFDR = .008), showing 485 proprioceptive performance was independent of working memory. 486 Predicting Motor Performance from Proprioceptive Acuity 487 To allow visual comparison of the reaching performance with the proprioceptive measures, 488 the spatial distribution of individuals' average end-positions and the 95% confidence 489 interval ellipses in the motor reaching task are shown in Figure 6 for each target, with the 490 bias and uncertainty range from the proprioceptive task shown in bar-format. 491 We generated 2 regression models for each proprioceptive-motor performance pairing, 492 collapsing data across all 3 targets, giving 4 models overall. Neither the bias and LE (OAs, R 2 493 = .002; YAs, R 2 = .020) nor the uncertainty range and LE Var (OAs, R 2 = .060; YAs, R 2 = .035; 494 pmin = .090; αFDR = .013) models were significant (see Table 3 for summary). We did observe that uncertainty range was a significant, negative predictor of LE Var for OAs only (β = -.245; 496 p = .030), however, this did not survive corrections for multiple comparisons and the overall 497 model still accounted for only 6% of the variance in the data. The lack of relationship 498 between proprioceptive uncertainty and motor error in advanced age contradicts our 499 original prediction, and no consistent positive association was seen in any group. 500 501 502 30 504 505 506 507 Figure 6 -Individual participant average end-positions from rapid motor task (coloured 'X' markers) and 95% confidence ellipses for each of the different groups and targets. Group average data from dynamic proprioceptive task is scaled and superimposed over targets as coloured bars. The central thick coloured line in each bar represents the bias and on average shows participants perceived their hand to be more towards the clockwise ("Circle") side of the target. The length of the coloured bar represents the uncertainty range and was similar between groups (figure generated for visualisation purposes only) Discussion 508 This experiment aimed to determine the relationship between dynamic proprioceptive 509 acuity and movement control in the upper limb with advanced age. Although we found 510 stereotypical features of ageing in motor kinematics, we also found that proprioceptive bias, 511 and not uncertainty range, was larger for physically inactive OAs, contrasting to our 512 predictions. While we did observe a trend towards higher uncertainty range predicting 513 lower variability in motor accuracy for OAs, the direction of this relationship and its limited 514 strength (R 2 = .06) lead us to conclude a negligible association overall. Ultimately, 515 proprioceptive uncertainty was not consistently related to variability in movement accuracy; 516 thus, we find no evidence to link proprioception and movement control in either older or 517 younger adults in this experiment. 518 Our results replicate the findings of Helsen et al. (2016), who showed a dissociation of 519 proprioceptive acuity and rapid motor performance, but we extend beyond their results to 520 show this is true when proprioception is measured via an active movement task, which 521  concluded that OAs were able to overcome a decline in sensory acuity through increased 523 reliance on predictive control mechanisms in a "play-it-safe" strategy (Elliott et al., 2010). 524 We also saw evidence that OAs tend to emphasise accuracy over speed, exemplified by their 525 increased reaction times and reduced peak velocities. These speed differences may partially 526 explain the comparable endpoint accuracy seen between groups ( Figure 5B); a finding which 527 has also been reported elsewhere (Helsen et al., 2016;Lee et al., 2007;Seidler-Dobrin & 528 Stelmach, 1998). We note that the utility of online proprioceptive feedback in fast, discrete, 529 movements is likely reduced compared to slower, guided movements, and the reliance on 530 predictive mechanisms may therefore already be high in our reaching task ( Contrary to our predictions and to prior literature, we showed that physical inactivity did 557 not increase proprioceptive uncertainty in OAs. We suggest this novel finding reflects the 558 steps we took to remove confounds when measuring proprioception. Namely, we used 559 active instead of passive movements (Smith et al., 2009)  differences which influence the presentation of these errors, or the mechanism by which 575 they may occur. Here, we have shown that physical inactivity in ageing is a contributing 576 factor. Although the cause is as yet unclear, a reduction in physical activity could lead to 577 everyday limb movements being made within a more concentrated volume, ipsilateral to 578 the limb (Howard, Ingram, Körding, & Wolpert, 2009), biasing sensory experience to this 579 region. Increased sensory uncertainty upon removal of vision (as in the proprioceptive 580 assessment task) may therefore lead to greater reliance on prior experience during the 581 optimal estimation of limb position (Gritsenko, Krouchev, & Kalaska, 2007;Körding & 582 Wolpert, 2006). We also note that spindle afferents are directionally tuned to specific 583 movements (Bergenheim, Ribot-Ciscar, & Roll, 2000; Jones, Wessberg, & Vallbo, 2001) and 584 loss of intrafusal fibres with age has been shown to be muscle specific (Kararizou, Manta, 585 Kalfakis, & Vassilopoulos, 2005). Therefore if movements are indeed limited to a smaller 586 range in physically inactive adults, a selective loss of intrafusal fibres which are directionally 587 tuned to the less frequent movements, might result. Collectively, these effects could lead to 588 the increase in proprioceptive bias we observed in the physically inactive OAs.
Unfortunately, the wrist-worn accelerometers we used do not provide spatial information, 590 and this suggestion remains to be tested. An alternative could be that the perceptual bias 591 arose from proprioceptive drift (Brown, Rosenbaum, & Sainburg, 2003b, 2003aDesmurget, 592 Vindras, Gréa, Viviani, & Grafton, 2000). However, drift is typically observed during 593 repetitive, unconstrained movements and has been attributed to the persistence of motor 594 errors rather than to proprioceptive fading (Brown et al., 2003b). In addition, the extent of 595 proprioceptive drift has been associated with movement speed (Brown et al., 2003b), and 596 we found no association between bias and movement velocity. 597 We do, however, report a positive correlation of average movement speed and uncertainty 598 range in the proprioceptive task for the inactive OAs. This observation may further reflect a 599 speed-accuracy trade-off where insufficient sensory information is accumulated to make 600 reliable perceptual judgements as movement speed increases (Bogacz, Wagenmakers, 601 resources in the inactive OAs might therefore impair their ability to process sensory 607 feedback for perceptual judgements. However, we found no relationship between verbal 608 working memory score and perceptual acuity for any group, suggesting this is not a factor in 609 our inactive elderly group. 610 In conclusion, we found systematic differences in movement kinematics in OAs compared to 611 YAs, as expected from previous reports. We also found an age-dependent increase in 612 proprioceptive bias measured in active, multi-joint movement, but not of uncertainty range. 613 This finding is novel and may reflect our careful task design which aimed to remove 614 methodological confounds for testing with an ageing population. However, we did not find 615 any evidence to suggest that proprioceptive acuity is related to performance in rapid, goal-