The KINARM Exoskeleton Lab (BKIN Technologies, Kingston, ON, Canada) was used to collect kinematic data of the upper limbs (Figure 1A; Scott, 1999). Participants were seated in the robotic exoskeleton with their shoulders at ~85° of abduction and their upper and lower arm secured within arm troughs. The proximal and distal segments of the robot were adjusted to each participants’ respective limb proportions. Once the exoskeleton was fit, participants were wheeled into the horizontally-mounted virtual reality display with their head resting on a padded fabric affixed to the virtual reality display. Participants had two degrees of freedom within the robot: horizontal flexion/extension at the shoulder and flexion/extension at the elbow. The exoskeleton is capable of producing mechanical loads at both the shoulder and elbow joints to passively move the participant’s arm with a bell-shaped velocity profile given a desired location and duration. A calibration procedure preceded each participants’ experiment to create the most accurate virtual reality display. Vision of the arms was occluded with a metal shutter and a bib.

The robot passively moved one arm (passive arm) and participants were instructed to mirror-match the movement with their opposite arm (active arm) without the use of visual feedback. Specifically, individuals were instructed to mirror-match the speed, direction, and length of each movement with their active arm as soon as they perceived their passive arm being moved (Supplementary material; Figures 1B,C). For the purposes of experimental protocol, a trial began when the robot started to passively move one arm and ended when the participant verbally indicated they were matched, and the robot operator clicked a button. The button click revealed vision of the hands’ current positions (1 cm white cursor at fingertip of both hands) and the desired positions (2 cm cyan circles mirrored across body midline), and participants were then instructed to move their active arm (white cursor) into the cyan circle in order to move to the next trial. After their active hand was in the target for a randomized amount of time (400–1,000 ms), all visual information was extinguished, and the next trial began. The purpose of this visual feedback was to ensure that all participants began each trial with both limbs in a mirror-matched position. No data from time periods with visual feedback was used in subsequent analyses. The passive arm (i.e., arm moved by robot) was counterbalanced within each group to account for effects of handedness. In the younger (≤30 years old) control group, 10 participants had their dominant arm passively moved by the robot and 10 participants had their non-dominant arm passively moved by the robot. In the older (≥45 years old) control group, 10 participants had their dominant arm passively moved by the robot and 11 participants had their non-dominant arm passively moved by the robot.

We assessed five distances and seven peak speeds with a total of 10 reaches per combination of peak speed and distance, resulting in a total of 350 reaches to 350 targets within a 20 × 20 cm workspace (Figure 1D). The assessed distances ranged from 7.5 to 17.5 cm with steps of 2.5 cm, for a total of 5 distances tested. The assessed peak speeds ranged from 0.1 to 0.4 m/s with steps of 0.05 m/s, for a total of 7 speeds tested. Together, there were a total of 35 speed x distance combinations. Each passive robotic movement had a desired movement profile, initialized with a randomized start and end target. The angle of this movement was randomly chosen from a uniform distribution from 0° to 359° with consideration of workspace boundaries. Within these two targets, movement speed was commanded to generate the passive movement implemented via a bell-shaped velocity profile. A custom MATLAB algorithm was designed to create a schedule of target locations and displacement durations, such that (1) speed and distance combinations were randomly ordered (2) each speed and distance combination was sampled equally, and (3) direction of the movement was random. For this experiment, the same randomized schedule of target locations and displacement durations were used for all participants.

The data were analyzed to determine the magnitude of spatial and temporal proprioceptive matching errors. Data was initially parsed from the time of trial start (passive movement begins) to end of trial (participant indicates they are matched/operator button click), which is prior to when vision of hands and targets is revealed. Peak speed was defined as the maximum hand speed within this time frame. Movement onset, for each arm, was defined as the time when hand speed was greater than or equal to 10% of the respective peak hand speed for 50 consecutive milliseconds. Movement onset of the active arm was additionally required to be at least 150 ms after the movement onset of the passive arm to ensure movements were not anticipated. Movement end, for each arm, was defined as the time when the respective hand speed decreased below 10% of the respective peak speed for 50 consecutive milliseconds or when the trial ended, whichever occurred first. All analyses and figures were done using custom MATLAB code.

Kinematic data of the active limb in the x-dimension was mirrored (sign-flipped) across the workspace to make direct comparisons between passive and active movements (Figure 1E). End Point Error was calculated as the Euclidean distance between the passive limb and the active limb at their respective movement end locations (Polit and Bizzi, 1979; Messier et al., 2003; Sarlegna and Sainburg, 2009; Dukelow et al., 2010) (Equation 1). A larger End Point Error indicates that the active arm was farther away from the “ideal” end point of the passive arm. Initial Direction Error was calculated as the angle between vectors connecting the start location and the locations at peak speed for the passive arm and the mirrored active arm (Sarlegna and Sainburg, 2009; Semrau et al., 2013) (Equation 2). A larger Initial Direction Error indicates that the direction of movement of the active arm was greater than the “ideal” trajectory of the passive arm.

Equation 1. End Point Error. Offset is defined as the hand position during movement offset. Movement offset was calculated as the first time the hand speed fell below 10% of the peak speed or the trial ended. The offset of the active arm was reflected across the x-axis for the calculation.

Equation 2. Initial Direction Error. Initial movement vectors were determined for both the passive and active arms. This vector began at the starting x,y position and stopped at the x,y position when peak speed occurred. Initial Direction Error was then calculated by determining the angular difference between the two vectors.

Response Latency was calculated as the difference in time between movement onset of the passive arm and movement onset of the active arm (Vercher et al., 1997; Semrau et al., 2013) (Equation 3). A larger Response Latency indicates that the participant took longer to initiate their movement in response to the passive movement of the robot. Peak Speed Ratio was calculated as the quotient of peak speed of the active arm and peak speed of the passive arm (Semrau et al., 2013) (Equation 4). A Peak Speed Ratio greater than 1 indicates that the active arm had a greater peak speed than the passive arm.

Equation 3. Response Latency. Movement onset was calculated as the first time the hand speed fell below 10% of the peak speed when looking backwards in time from the time of peak speed.

Equation 4. Peak Speed Ratio. Peak speed of the arm was defined as the peak of the hand speed between movement onset and movement offset. Movement onset was defined as in Response Latency (Equation 3). Movement offset for Peak Speed Ratio was defined the same as in End Point Error (Equation 1).

To analyze group differences, permutation tests and common language effect size (CLES) were utilized (Good, 2005). Directional permutation tests (H₀: younger controls ≥ older controls) were used to compare between groups for the following parameters: End Point Error, Initial Direction Error, and Response Latency, and to statistically confirm age differences between the older control group and the younger control group. Non-directional permutation tests (H₀: younger controls = older controls) were used for comparisons between groups for the fourth (Peak Speed Ratio) proprioceptive matching outcome. Non-directional tests were used for Peak Speed Ratio because error increased as the value deviated from 1 in either (positive or negative) direction. All permutation testing was completed by using averaged parameter (e.g., Response Latency) data from individual subjects, shuffling group assignments, and implementing 1,000,000 iterations to compare the resultant distribution to test for differences in the averages of the resampled groups. CLES values were calculated exact (i.e., permute all possible combinations). This method was used to examine (1) group level performance without consideration of speed or distance, (2) group level differences for speed within a single parameter (e.g., Response Latency), (3) group level differences for distance within a single parameter. For example, in examining effects of speed on Initial Direction Error performance, for each of the seven speed values tested [0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4 m/s], a permutation test was utilized to determine if the error distribution of Initial Direction Error was significantly greater for older controls compared to younger controls. This process was then repeated for all parameters (End Point Error, Initial Direction Error, Response Latency, and Peak Speed Ratio) described above for each of the seven speed values and each of the five distance values.

To examine overall patterns of proprioceptive matching error between groups for differences in speed or distance, for each parameter (e.g., Response Latency), we fit individual participants averaged data (5 data points for distance analyses, 7 data points for speed analyses) to a line using ordinary least squares. This yielded an intercept ( β 0 ) and slope ( β 1 ) coefficient for each participant for each proprioceptive matching outcome to examine parameter-based differences between groups for error magnitude (intercept) and error scaling (slope). To determine if these distributions were significantly different between groups, we used permutation tests to compare the distributions of intercept and slope coefficients between groups for each proprioceptive matching outcome by speed/distance. Lastly, we utilized a two-way ANOVA with averaged group data to quantify interaction effects between speed and distance within each group for each proprioceptive matching parameter (alpha = 0.05).

To compare overall proprioceptive matching accuracy between our groups, we compared group performance for each proprioceptive matching measure, regardless of speed and distance (Figure 2). We found significant differences between groups for all our spatial (End Point Error and Initial Direction Error) and temporal (Response Latency and Peak Speed Ratio) proprioceptive matching measures. Specifically, older adults had significantly higher End Point Error (p = 0.007, CLES = 71.90), Initial Direction Error (p = 0.008, CLES = 70.95), and Response Latency (p = 0.013, CLES = 67.38) and less accurate Peak Speed Ratio (p = 0.031, CLES = 67.62) compared to younger adults.

To understand which movement speeds were contributing to the differences seen in overall proprioceptive matching error, we examined the difference in group performance within each speed tested (Figure 3). We found that older adults had significantly worse proprioception compared to younger controls for End Point Error [p < 0.05 for 0.15, 0.20, 0.25, 0.30, 0.35, 0.40 m/s], Initial Direction Error [p < 0.05 for 0.15, 0.20, 0.25, 0.30, 0.35, 0.40 m/s], Response Latency [p < 0.05 for 0.10, 0.15, 0.20, 0.25, 0.30, 0.35, 0.40 m/s], and Peak Speed Ratio [p < 0.05 for 0.20, 0.25, 0.30, 0.35, 0.40 m/s] (Full statistical results included in Table 1).

To further understand the influence of age on proprioceptive matching accuracy as function of speed, we examined how each proprioceptive matching outcome changed as a function of speed (Figure 4). When we examined the intercept ( β 0 ) and slope ( β 1 ) coefficients of the proprioceptive matching outcomes by speed, we found that for End Point Error and Peak Speed Ratio, older controls had significantly greater (End Point Error: p = 0.005, CLES = 74.29) and significantly different (Peak Speed Ratio: p = 0.047, CLES = 69.76) slopes compared to younger controls. We also found that for Response Latency, intercept values for older controls were significantly greater than younger controls indicating a higher degree of error for older adults (Response Latency: p = 0.009, CLES = 72.86). This shows that for End Point Error and Peak Speed Ratio, older adults had increased error as a function of passive movement speed. Additionally, for Response Latency, older adults had a higher level of proprioceptive matching error regardless of movement speed. We also examined the influence of handedness for task performance for younger and older adults. Here, we separated those individuals where the robot moved the dominant arm (Younger Adults: N = 10, Older Adults: N = 10) and where the robot moved the non-dominant arm (Younger Adults: N = 10, Older Adults: N = 11) into two groups. We found that across the different reference speeds tested, younger adults’ dominant arm showed slight improvement in proprioceptive matching error for End Point Error [p < 0.05 for 0.10, 0.15, 0.20, 0.25, and 0.30 m/s, and intercept] and Initial Direction Error [p < 0.05 for 0.10, 0.15, 0.20, and 0.25 m/s, and intercept and slope] compared to their non-dominant arm. In contrast, we found no significant differences within our older adult group for dominant vs. non-dominant limbs.

To understand which distances were contributing to differences seen in overall proprioceptive matching error, we examined the difference in group performance within each distance tested (Figure 5). We found that older adults had significantly worse End Point Error [p < 0.05 for 7.5, 10.0, 12.5, 15.0, 17.5 cm], Initial Direction Error [p < 0.05 for 7.5, 10.0, 12.5, 15 cm], Response Latency [p < 0.05 for 7.5, 10.0, 12.5, 15.0, 17.5 cm] and significantly different Peak Speed Ratio [p < 0.05 for 7.5, 10.0, 15 cm] at a majority of distances for these measures when compared to younger controls (Table 1).

To further understand how proprioceptive matching accuracy changed as a function of distance, we examined the intercept and slope coefficients of the proprioceptive matching outcomes by individual distance (Figure 6). We found that for older adults, Initial Direction Error, Response Latency, and Peak Speed Ratio (Initial Direction Error: p = 0.002, CLES = 74.29, Response Latency: p = 0.014, CLES = 63.57, Peak Speed Ratio: p = 0.016, CLES = 70.71) had intercepts that indicated greater overall error across distances compared to younger controls. This suggests that for multiple parameters, older adults had a higher error magnitude compared to younger controls. As described above, to examine effects of handedness, groups were split into individuals where the robot moved the dominant arm and where the robot moved the non-dominant arm. We found that across the different reference distances tested, younger adults’ dominant arm showed slight improvement in proprioceptive matching error for End Point Error [p < 0.05 for 12.5, 15.0, and 17.5 cm, and slope], Initial Direction Error [p < 0.05 for 12.5 and 15.0 cm], and Path Length Ratio [p < 0.05 for intercept and slope] compared to their non-dominant arm. In contrast, we found no significant differences within our older adult group for dominant vs. non-dominant limbs.

To gain an understanding of the interaction of speed and distance on proprioceptive matching accuracy, we examined the interaction term of a two-way ANOVA on each groups average data. We found that younger controls had a significant speed x distance interaction effect for Peak Speed Ratio (F = 7.080, p = 0.012), while older adults had a significant speed x distance interaction effect for Response Latency (F = 16.901, p < 0.001).

Previous work has detailed the presence of age-related declines in proprioception (Kokmen et al., 1978; Stelmach and Sirica, 1986; Adamo et al., 2007; Wright et al., 2011; Ingemanson et al., 2016; Rinderknecht et al., 2017; Acosta-Sojo and Martin, 2021). This work, and others, has typically examined participants’ ability to replicate movement speed or distance (e.g., reference movement of 20 degree elbow flexion) (Adamo et al., 2007, 2009; Wingert et al., 2009; Dukelow et al., 2010; Semrau et al., 2013; Li and Wu, 2014; Contu et al., 2017; Watkins et al., 2020; Acosta-Sojo and Martin, 2021) and has found that aging-related increases in proprioception are common across different testing mechanisms including bilateral proprioceptive testing (Stelmach and Sirica, 1986; Adamo et al., 2007, 2009; Kalisch et al., 2012; Herter et al., 2014; Acosta-Sojo and Martin, 2021) and unilateral or within arm proprioceptive testing (Ferrell et al., 1992; Adamo et al., 2007, 2009; Wright et al., 2011; Wingert et al., 2014; Ingemanson et al., 2016; Rinderknecht et al., 2017; Acosta-Sojo and Martin, 2021). However, it is important to note that these studies, as well as our own previous work (Semrau et al., 2013), have generally used methods that test a relatively narrow range of movement types. This has provided limited interpretation on whether the proprioceptive system is sensitive to particular movement characteristics or whether it exhibits a similar level of error across all movement characteristics.

In the current study, we have used methodology that allows us to examine proprioceptive error responses that consider a broad range of movement types, ranging from short, slow reference movements to long, fast reference movements (see Methods). Here, we have used this method to understand two main aspects of proprioception: (1) whether the magnitude of proprioceptive error changes as a function of the characteristics of the reference movement, and (2) whether aging negatively impacts proprioceptive estimations and if so, are proprioceptive error patterns different for younger and older adults. Importantly, we first observed that older adults show considerable decline in nearly all proprioceptive parameters compared to younger adults and that when we examine specific speeds and distances, these differences are consistent. Secondly, we find that while overall patterns of error are significantly higher for older adults compared to younger adults, there were also significant differences in proprioceptive error scaling for changes in reference speed for older adults. Here, we found that for End Point Error, older adults scaled their error more rapidly as reference speed increased, as indicated by a much steeper slope (Figure 4A). Additionally, we observed that older adults were able to similarly match movement speed when the reference movement was slow (0.1 m/s), but as the reference speed increased, the magnitude of speed matching error as measured by the Peak Speed Ratio parameter also increased, as demonstrated by older adults greatly underestimating the reference speed at the fastest reference movements. In contrast, while we observed overall higher levels of proprioceptive error when the reference movement changed distance, we did not observe differences in error scaling for changes in distance as we did for changes in speed (Figure 6). Work by Goble and Brown (2009) examined upper limb proprioceptive matching behavior at two different reference speeds (30°/s and 60°/s) in younger adults. They found that proprioceptive matching errors were significantly greater when participants experienced the faster reference speed. They concluded that this increase in proprioceptive error at the faster speed was due to sensory attenuation that occurs during the movement. Previous work from Collins et al. (1998) found that faster movements of the wrist resulted in a greater amount of sensory attenuation, as demonstrated by significant reductions in muscular sense (i.e., twitch amplitude). We believe that our current results demonstrate similar sensory attenuation at faster movement speeds for older adults, and that the differences observed between younger adults and older adults demonstrate an amplification of this attenuation as a result of the aging process. However, we must note that various control properties of motor units change with aging, such as decreased average firing rate (Erim et al., 1999) and decreases in the number of motor units (Brown et al., 1988), which may contribute to other changes in limb perception with age, such as the sense of effort (Monjo et al., 2018).

While we expected to observe that older adults had higher levels of proprioceptive error compared to younger adults, we did not expect that differences in error patterns relative to speed or distance to be so robust. These differences suggest that as we age, the distribution of our proprioceptive accuracy or sensitivity narrows. There are several contributing factors that may explain these differences. The first is that during normal aging, neuroanatomical changes occur. Here, there is a loss of white matter (Guttmann et al., 1998) over time and diminished white matter integrity has been shown to be predictive of poor proprioception and balance in the older adults (Van Impe et al., 2012); however the impact on upper limb proprioception is unknown (Zhai et al., 2020). In contrast, volumetric changes have not been observed in sensorimotor cortical areas or callosal projections to these areas in older adults (Raz et al., 2004; Michielse et al., 2010). This suggests that aged-related changes to fibers responsible for carrying afferent information may play a role in increases in proprioceptive error. Second, as we age, there is loss of both extrafusal and intrafusal muscle fibers (Brooks and Faulkner, 1994). Previous studies have found that, with age, the number of muscle spindles decrease in number over time and receive less innervation than in “young” muscle (Swallow, 1966; Swash and Fox, 1972; Proske and Gandevia, 2012; Landelle et al., 2021). It is reasonable to suspect that diminished function of sensory receptors leads to more variable proprioceptive estimation. Lastly, it is possible that the increased errors and age-related differences in proprioceptive error patterning that we observe in older adults may be due to decrements in sensorimotor integration that occur with age (Brown et al., 2018; Sorrentino et al., 2021), that may result in higher proprioceptive gains or a noisier system that is less sensitive to changes in perceived body position or movement. A contributing factor to this may be from sensory attenuation that occurs as a result of age. A recent study has suggested that sensory attenuation leads to alterations in the internal estimation of force for older adults (Parthasharathy et al., 2022). While the authors report that they did not observe declines in static proprioception in older adults, they found that older adults significantly over-estimated self-produced forces. This result, in combination with previous work finding increased sensory attenuation during both slow and fast passive movements (Collins et al., 1998), suggests that alterations in afferent signaling from the muscles as well as efference copy signals contribute to diminished proprioceptive accuracy in older adults.

The paradigm used to assess proprioception in this study is not without limitations. One limitation of this paradigm is the inherent motor component of the matched movements. If a participant had a motor deficit in their matching arm, one could confuse the deficits seen in their data as sensory deficits, when in fact they were motor deficits. In this study, we counterbalanced the arm moved by the robot in each of our groups in order to ensure that any interlimb differences in proprioception were accounted for and any sensory declines related to aging would not potentially be masked by improved ability using the dominant limb. When we examined inter-limb differences, we found some differences for when the robot moved the dominant vs. non-dominant arm, but only in younger adults. These results are in line with previous work describing minimal differences in task performance with a bilateral matching task, such as the one we use here (Goble and Brown, 2009), suggesting that the small differences seen in the younger adult group require further exploration to determine if these are true differences in proprioception or variability in performance. Another limitation is the range of movement speeds and distances tested. We set movement speed and distance bounds to fit the robot’s mechanical capacity as well as participants physiological capacity. We could have tested with more granularity between the bounds, but this would have increased the task time beyond a comfortable time for many participants.