Voluntary modification of rapid tactile-motor responses during reaching differs from its visuomotor counterpart

Abstract

People commonly hold and manipulate a variety of objects in everyday life, and these objects have different physical properties. To successfully control this wide range of objects, people must associate new patterns of tactile stimuli with appropriate motor outputs. We performed a series of experiments investigating the extent to which people can voluntarily modify tactile-motor associations in the context of a rapid tactile-motor response guiding the hand to a moving target (previously described in Pruszynski JA, Johansson RS, Flanagan JR. Curr Biol 26: 788–792, 2016) by using an anti-reach paradigm in which participants were instructed to move their hands in the opposite direction of a target jump. We compared performance to that observed when people make visually guided reaches to a moving target (cf. Day BL, Lyon IN. Exp Brain Res 130: 159–168, 2000; Pisella L, Grea H, Tilikete C, Vighetto A, Desmurget M, Rode G, Boisson D, Rossetti Y. Nat Neurosci 3: 729–736, 2000). When participants had visual feedback, motor responses during the anti-reach task showed early automatic responses toward the moving target before later modification to move in the instructed direction. When the same participants had only tactile feedback, however, they were able to suppress this early phase of the motor response, which occurs <100 ms after the target jump. Our results indicate that while the tactile motor and visual motor systems both support rapid responses that appear similar under some conditions, the circuits underlying responses show sharp distinctions in terms of their malleability.

NEW & NOTEWORTHY When people reach toward a visual target that moves suddenly, they automatically correct their reach to follow the object; even when explicitly instructed not to follow a moving visual target, people exhibit an initial incorrect movement before moving in the correct direction. We show that when people use tactile feedback, they do not show an initial incorrect response, even though early muscle activity still occurs.

INTRODUCTION

Many everyday actions, like reaching for a bottle’s cap while holding the bottle, require fast and precise bimanual coordination. We recently described a tactile-motor response, in which a tactile stimulus delivered to one hand drives rapid online correction of reaching with the other hand, that can support such coordination (Pruszynski et al. 2016). This tactile-motor response appears similar in timing to fast, visually guided responses that help guide the reaching hand toward a moving target (Day and Lyon 2000; Franklin and Wolpert 2008; Goodale et al. 1986; Gu et al. 2016; Paulignan et al. 1991; Pélisson et al. 1986; Prablanc and Martin 1992; Pruszynski et al. 2010; Saijo et al. 2005).

One hallmark of rapid visual responses is that they are tightly coupled to the position of the visual target. When people are instructed to respond to a target jump by moving in the opposite direction of the jump (“anti-reach”), they exhibit a delay in production of the correct movement compared with that observed when executing a movement in the same direction of the jump (“pro-reach”). This early response is likely of subcortical origin (Alstermark et al. 1987; Day and Brown 2001), and people have difficulty suppressing this sort of movement when they are instructed to do so: they first exhibit an erroneous pro-reach toward the target (Day and Lyon 2000; Gu et al. 2016; Pisella et al. 2000) similar to the erroneous prosaccades observed at low latency during antisaccade tasks (Everling et al. 1999; Fischer and Weber 1992; Gottlieb and Goldberg 1999). This sequence of events is attributed to the automaticity of the pro-reach and the additional time it takes to prepare a voluntary action in the opposite direction owing to the necessity of processing stimulus location and remapping this stimulus, with respect to the instruction, to a specific motor output.

While rapid reaching corrections have been observed in response to tactile stimulus, stimulus-response mappings for tactile stimuli are not necessarily similar to those for visual stimuli. Subcortical circuits play an important role in producing visually guided corrections (although cortical structures are also implicated in planning the reach), and it is possible that subcortical circuits are also responsible for trajectory changes during tactile-guided reaching: orientation of tactile stimuli can be extracted from the activity of first-order tactile neurons (Hay and Pruszynski 2019; Pruszynski and Johansson 2014), so higher centers may not be required for reaching with tactile guidance. Unlike visual information, however, tactile information is not as directly related to the position of objects in external space. In fact, relationships between tactile information and spatial coordinates are highly variable across everyday objects, or even on the same object when it is held in different orientations. Consider using a pool cue to precisely strike the cue ball or a fly rod to cast a fishing lure to a specific location: owing to the different physical properties of the cue stick (rigid) and fly rod (flexible), maneuvering them to specific spatial locations requires mapping tactile information in vastly different ways.

Here, we investigated the flexibility of this tactile-motor response to remap stimuli to outputs by using an anti-reach task. In our first experiment, participants performed anti-reaching with one hand while guided by an edge they felt with the contralateral hand. We predicted that the tactile-motor response would be able to reverse its action without a time cost. However, we found that while participants did not first move in the incorrect direction, they did not make the correct response as quickly as during a tactile pro-reach task. In our second experiment, we investigated whether visual and tactile feedback conflict with one another during the anti-reach task; we predicted that the initial (incorrect) portion of the visual response would be larger when participants had only visual feedback compared with when they had both visual and tactile feedback. Actually, participants’ responses were not reliably different when they had visual feedback compared with when they had both visual and tactile feedback. Finally, we tested whether a geometric tactile stimulus is required to change reaching trajectories by replacing the oriented edge stimulus with a textured stimulus under the contralateral finger; we found that participants performed similarly in both reach and anti-reach tasks when they did not have an oriented stimulus. Together, these results indicate that the tactile-motor response is indeed more flexible than the classic visual stimulus-locked response but that it is not completely malleable and does not result in faster anti-reach kinematic responses, at least on the 1-h timescale of this study.

MATERIALS AND METHODS

Participants

A total of 56 unique participants (18–35 yr old; 34 women) participated in three experiments. One cohort of 20 individuals participated in Experiments 1a and 1b, 16 participants performed Experiment 2, and another cohort of 20 participated in Experiments 3a and 3b. All participants provided written informed consent in accordance with methods approved by the Health Sciences Research Ethics Board at Western University.

Procedures

Participants sat at a table in front of the experimental apparatus. Each participant used their dominant hand to reach from a start position to a spherical target (4-cm diameter) mounted on a 30-cm-long carbon fiber rod (Fig. 1A) such that distance from the start position to the center of the target was 28 cm, and the closest aspect of the target was 26 cm from the start position. Participants used a finger of their nondominant hand (thumb or index finger) to feel a tactile stimulus (which varied depending on experiment) mounted in line with the rod (Fig. 1, B and C). A high-speed stepper motor rotated the rod, stimulus, and target on some trials. On trials where the target moved, its movement was triggered when the participant lifted their finger from the start position (measured with a resistive pressure sensor) to begin reaching. The latency between liftoff and the initiation of movement was measured at ~30 ms, and the rotation movement took 50 ms to complete. Participants received an auditory “ready” cue, after which they could initiate a reach toward the target at a self-selected time. Participants also received an auditory cue 300 ms after the target movement (or time when the target would have moved); they were instructed to try to finish reaching by the time they heard this second cue as a pacing method.

Fig. 1.

Experimental setup. A: illustration of the experimental setup as viewed from the top, illustrating the position of the target in its initial position (gray), after clockwise (CW) jump (red), and after counterclockwise (CCW) jump (blue). B: illustration of participant’s starting position for all trials. C: illustration of the configuration of edge stimulus viewed from underneath the fingertip following CW jump (red), no jump (black), and CCW jump (blue); arrows indicate the relative motion of different parts of the edge about the center of rotation. D and E: across-participant mean muscle electromyographic (EMG) activity traces. Thick line represents the mean and shaded area represents SE. Colors represent the CW (red), CCW (blue), and no (black) jump conditions. Data aligned on rotation onset. Gray box illustrates the time during which EMG data were analyzed: a 25-ms epoch following median time of discrimination via receiver-operator characteristic (ROC) for reach instruction for a given participant (D, gray area; see materials and methods); the same time was then used to analyze EMG for the anti-reach instruction (E). F: illustration of the “crossover” metric on an exemplar participant during the anti-reach instruction with visual and tactile feedback (see materials and methods).

Experiment 1a: Reaching with visual and tactile feedback or tactile feedback only.

Participants (n = 20; 13 women; 1 left-handed) reached toward the target using either visual and tactile feedback or tactile feedback only. Tactile feedback was provided using a raised edge oriented in line with the rod holding the target ball and spanning the entire surface of the finger (Pruszynski et al. 2016). During the touch only condition, we occluded participants’ vision using LCD shutter glasses (PLATO; Translucent Technologies, Toronto, ON, Canada) that closed before participants received the ready cue and opened 300 ms after the target moved. Participants completed 240 trials, blocked according to whether visual feedback was available (120 per feedback condition); in one-third of trials in each block (40), the target did not move; in the other trials, the target rotated 15° clockwise (CW; right) or counterclockwise (CCW; left). Target movement was quasi-randomized such that all participants performed 40 trials with each movement (left, right, or none), but the target movement on a given trial was unpredictable.

Experiment 1b: Anti-reaching with visual and tactile feedback or tactile feedback only.

The same participants who participated in Experiment 1a also performed Experiment 1b. The procedure was identical to that of Experiment 1a, except participants were instructed to reach in the opposite direction of target movement using either vision and touch or touch only such that, in these trials, participants were not moving to any target. Individuals participated in Experiment 1a and Experiment 1b one after the other for a total of 480 trials, but the order of participation was balanced such that half of the participants first performed Experiment 1a and the other half first performed Experiment 1b.

Experiment 2: Anti-reaching with visual and tactile feedback, visual feedback only, or tactile feedback only.

Another cohort of participants (n = 16; 10 women) performed an anti-reaching experiment very similar to Experiment 1b. However, whereas participants in Experiment 1b who received visual feedback also felt the edge with their nondominant finger, we added a condition in Experiment 2 in which participants did not touch anything with their nondominant finger. In each of the three feedback conditions, participants again performed 40 trials with target movements in each direction, performing a total of 360 trials during the experimental session. Participants performed all trials in a given feedback condition as a block; the order of these blocks was balanced across participants.

Experiment 3a: Reaching with texture tactile feedback.

A third cohort of participants (n = 20; 10 women; 1 left-handed) reached to a target using visual and tactile feedback or tactile feedback only. The procedure was identical to Experiment 1, except that tactile feedback was provided with a sheet of fine sandpaper (320 grit; ~50 μm) on a flat plate, rather than a raised edge as in Experiment 1.

Experiment 3b: Anti-reaching with texture tactile feedback.

The participants from Experiment 3a also performed Experiment 3b. The procedure was identical to Experiment 3a, except participants were instructed to reach in the opposite direction of target movement. Individuals participated in Experiments 3a and 3b during the same visit; half of the participants first performed Experiment 3a, while the other half first performed Experiment 3b.

Analysis

We recorded kinematics at 240 Hz using a magnetic tracker (Liberty; Polhemus, Colchester, VT) attached to each participant’s dominant index finger. We recorded muscle activity using wireless electromyography (Trigno; Delsys Inc., Natick, MA) with electrode modules placed on the anterior deltoid, long head of the biceps, posterior deltoid, and long head of the triceps. Electromyography (EMG) data were digitized at 2,000 Hz using a 16-bit analog-to-digital converter (ADC; USB-6225; National Instruments, Austin, TX). Kinematics data were logged using the tracker’s native software (Pimgr), while the EMG data were logged using a program written in Python (Python Software Foundation, Beaverton, OR), which also monitored the resistive pressure sensor to move the target and sent a synchronization signal to the kinematics logging software. We imposed a number of kinematic criteria on each individual trial to ensure that participants did not wait for the target to jump before moving toward the target, moved toward the initial target position rather than guessing the location to which the target would jump, and ended up reaching the target. We full-wave rectified muscle activity and filtered it using a fourth-order forward and reverse Butterworth filter with a passband of 40–250 Hz, and normalized to the average maximum value observed in that muscle over all trials in a given instruction and feedback combination.

We analyzed muscle activity in 25-ms windows defined by the median time at which muscle responses diverged to track leftward (CCW) and rightward (CW) target jumps (shaded regions in Fig. 1, D and E). We assessed the time of divergence for each participant individually using a receiver-operator characteristic (ROC); we define the time of divergence as the last local minimum or maximum preceding the time at which the ROC could successfully differentiate between 60% of trials. We tested a variety of ROC criteria to determine the divergence, which yielded similar temporal results. We use the assessment window computed during reach trials for anti-reach trials because we are interested in comparing the muscular response at similar times following target jump depending on the reaching instruction. While we recorded from four muscles of the upper arm, we report findings from anterior deltoid because it is a prime agonist during the reaching movement and shows direction-dependent activity when participants move to different targets in the medial-lateral plane.

We illustrate kinematic behavior (see Fig. 3, A and B) using the heading of the velocity vector. From our kinematic recordings, we analyzed two aspects of kinematic behavior: correct response latency and kinematic “crossover.” To assess the time when a given trial showed a correct response (see Fig. 4, B and C), we computed the medial-lateral velocity of the hand trajectory for every trial in which the target jumped. The time of correct response was defined as the last time the velocity trace crossed 0 in the direction of the target jump (or in the opposite direction during anti-reach). To obtain crossover (see Fig. 6), we first computed each participant’s average position trajectory toward CW and CCW target jumps under each instruction (reach and anti-reach). We then computed crossover as the maximum width of the area enclosed by overlapping CW and CCW traces; if the traces diverged without overlapping, we define crossover as 0.

Fig. 3.

Muscle activity in Experiment 1. A: average across-participant heading angle during Experiment 1a (pro-reach) for vision and touch (solid lines) and touch only (dashed lines). Color represents the target jump condition as indicated. Shaded areas represent SE. B: average across-participant heading angle during Experiment 1b (anti-reach) for vision and touch (solid lines) and touch only (dotted lines). C: muscle activation during vision and touch trials for pro-reach. D: muscle activation during vision and touch trials for anti-reach. E: muscle activation during touch-only trials for pro-reach. F: muscle activation during touch-only trials for anti-reach. CCW, counterclockwise; CW, clockwise; EMG, electromyography; n = 20 participants for all plots.

Fig. 4.

Analysis of divergence times. A: times of divergence (see materials and methods) for anterior deltoid electromyographic (EMG) activity in clockwise (CW) and counterclockwise (CCW) trials. Solid lines represent Experiment 1 (edge stimulus; n = 20 participants); dashed lines represent Experiment 3 (texture stimulus; n = 20 participants). Black lines are for vision and touch trials; gray lines are for touch-only trials. B and C: cumulative number of trials performed by all participants in Experiment 1 (B) and Experiment 3 (C) showing correct kinematic response at a given time following target movement. Black lines are for vision and touch trials, and gray lines are for touch-only trials. Dashed lines represent trials under the anti-reach instruction, and solid lines represent trials under the reach instruction. Total number of trials shown in B and C is truncated compared with the number of trials analyzed because the onset of correct kinematic response could not be algorithmically identified in all trials.

Fig. 6.

Kinematic crossover. Maximum overlap between average trajectories to clockwise (CW) and counterclockwise (CCW) target jumps (in cm) for each participant, compared between feedback modalities, in Experiment 1 (A; n = 20 participants), Experiment 2 (B; n = 16 participants), and Experiment 3 (C; n = 20 participants). Dark markers represent the reach instruction; light markers represent the anti-reach instruction. Participants performed only anti-reach trials in Experiment 2.

We carried out paired t tests in Python using the ttest_rel command in the Scipy library (Virtanen et al. 2020); for independent t tests, we used the ttest_ind command, and for correlations we used the pearsonr command in the same library. We ran repeated measures ANOVAs using linear mixed models analysis in R (R Core Team, version 3.6.1) with individual participants treated as random variables using independent intercepts, but not slopes, and reduced maximum likelihood fitting. We obtained P values for effects using the Kenward–Roger (Kenward and Roger 1997) approximation for denominator degrees of freedom as implemented in the packages lmerTest (Kuznetsova et al. 2017) and pbkrtest (Halekoh and Højsgaard, 2014), as recommended by Luke (2017) for small sample sizes. We used the package emmeans (Lenth 2020) for post hoc testing on effects more than two levels, utilizing Bonferroni corrections for multiple comparisons. We also used the Python libraries NumPy (Oliphant 2006), Matplotlib (Hunter 2007), and Pandas (McKinney 2010) during data analyses.

RESULTS

Participants’ success varied depending on the instruction (reach/anti-reach) and type of feedback available. Under the reach instruction, the median number of accepted trials (the number of trials in which kinematics met the aforementioned criteria) was 39 (of 40) for vision and touch, and 35 for touch only. Under the anti-reach instruction, the median number of accepted trials was 34 for vision and touch and 32 for touch only. Average reach trajectories in three dimensions (3D) and in the anterior-posterior/medial-lateral plane for a single participant and for all participants together are shown in Fig. 2.

Fig. 2.

Basic behavioral analysis. Data represent average kinematic behavior of a single participant (A, B, E, F, I, J) and across-participant averages (C, D, G, H, K, L) during Experiment 1 (n = 20 participants). Markers provide timing information (in ms) relative to target jump. A and C: 3-dimensional (3D) average traces for counterclockwise (CCW; blue), no jump (black), and clockwise (CW; red) target jumps under the pro-reach instruction. Solid lines are from vision and touch trials, while dashed lines are from touch-only trials. B and D: 3D average traces from feedback under the anti-reach experiment. E–L: average position in the anterior-posterior/medial-lateral plane for vision and touch trials under reach (E, G) and anti-reach instruction (F, H) and for touch-only trials under reach (I, K) and anti-reach instruction (J, L).

Participants moved somewhat faster when they had visual and tactile feedback than when they had tactile feedback only. Under the reach instruction, participants accelerated to an average 3D instantaneous velocity of 49.6 ± 7.7 cm/s (mean ± SE) by the time the target jump occurred and attained an average movement velocity of 112.0 ± 2.4 cm/s over the 300 ms following the target jump; when they did not have visual feedback, they were moving at an average of 37.4 ± 5.2 cm/s when the target jump occurred and moved at an average of 104.2 ± 3.0 cm/s over the next 300 ms. Under the anti-reach instruction, participants with visual feedback were moving at an average of 43.3 ± 6.8 cm/s when the target jumped and moved at 100.0 ± 4.1 cm/s over the next 300 ms, whereas participants with only tactile feedback moved averaged 39.7 ± 6.0 cm/s at target jump and 103.6 ± 3.4 cm/s over the next 300 ms. The aforementioned velocities are all in 3D; in the anterior-posterior direction only, the difference in speed between feedback conditions was marginal under the reach instruction when the target jumped (T₁₉ = 2.19; P = 0.04), and there was no consistent difference over the 300 ms following target jump (T₁₉ = 1.33; P = 0.2). Under the anti-reach instruction, there was no consistent difference in average instantaneous speed between feedback conditions during either analysis epoch.

Reaches toward a jumping target are rapidly updated via tactile information.

Under the reach instruction, when the target moved during an ongoing reach, participants rapidly updated their reaches when the target jumped (Fig. 3A). We used a receiver-operator characteristic (ROC) technique to find a time for each participant when muscle activity in anterior deltoid for rightward (CW) and leftward (CCW) target jumps diverged from one another. The median time of divergence when participants had visual and tactile information was 77 ms, and it was 93 ms (Fig. 4A, solid lines) when they only had tactile feedback; the divergence time was consistently faster in the former versus the latter as assessed by paired t test (T₁₉ = 4.14; P = 5.5e⁻⁴). The median time of correct response across all reach trials with a target jump under vision and touch was 138 ms, while the time of correct response under touch only was 175 ms (Fig. 4B, solid lines). EMG activity over the 25 ms following median time of divergence was modulated according to direction for both vision and touch trials (one-way ANOVA, direction: F_2,38 = 83.76; P = 1.18e⁻¹⁴) and trials with only tactile feedback (F_2,38 = 27.56; P = 4.02e⁻⁸); under vision and touch, EMG activity was different for each possible target movement outcome (CW, CCW, no movement), whereas for touch there was no significant difference between CW and no movement responses during this analysis epoch. We subsequently compared vision-and-touch and touch-only EMG activity by assessing the average difference in EMG activity between CW and CCW reaches during 25-ms bins beginning with the median onset time. The difference between CW and CCW EMG activity was greater for vision and touch compared with touch only (paired t test; T₁₉ = 3.97; P = 8.24e⁻⁴). Participants who showed more EMG activity under visual and tactile feedback across all target directions tended to show more EMG with only tactile feedback, as well (r = 0.76; P = 1.4e⁻¹²; Fig. 5A). Overall, the onset of corrections with only tactile feedback was marginally slower than reported by Pruszynski et al. (2016) and consistent with other reports of automatic visually guided behaviors (Day and Lyon 2000; Franklin and Wolpert 2008; Gu et al. 2016; Veerman et al. 2008).

Fig. 5.

Comparison of muscle activity across feedback and instruction in Experiment 1. A: average magnitude of electromyographic (EMG) activity under the reach instruction for each participant in trials with visual and tactile feedback (x-axis) and tactile feedback only (y-axis). B: average magnitude of EMG activity for each participant in trials under the reach instruction (x-axis) and under the anti-reach instruction (y-axis) for trials with visual and tactile feedback. C: average magnitude of EMG activity for each participant under the reach instruction (x-axis) and anti-reach instruction (y-axis) for trials with tactile feedback only. For A–C, markers are color coded according to direction of target jump (CW, clockwise; CCW, counterclockwise). Trials with no target jump (None) are included only in A. Norm, normalized; n = 20 participants for all plots.

Movement toward an anti-reach target is suppressed, but early EMG activity is evident.

When the same 20 participants performed anti-reach movements with visual and tactile feedback, they made initial movements in the direction of the target jump (consistent with classical stimulus-locked response) before moving in the instructed direction. In contrast, when participants had only tactile feedback, they generally did not make these initial incorrect movements before reaching in the instructed (anti) direction (Fig. 2, F and H, and Fig. 3B). We analyzed the extent to which anti-reach trajectories to CW target jumps overlapped with trajectories to CCW target jumps and showed that this overlap was significantly larger during anti-reach than reach for visual and tactile feedback trials (T₁₉ = 5.04; P = 7.3e⁻⁵); this finding indicates an initial reach in the incorrect direction (see Fig. 2, F and H, for examples of this behavior). In contrast, the instruction did not significantly modulate crossover for trials with only tactile feedback (T₁₉ = 0.55; P = 0.59). The crossover observed during anti-reach was larger for visual and tactile trials compared with tactile feedback-only trials (T₁₉ = 4.57; P = 2.1e⁻⁴), as shown in Fig. 6A.

Importantly, although participants did not move in the incorrect direction when they had only tactile feedback, their movements in the correct direction were delayed with respect to the timing observed under the pro-reach instruction. Across all trials, median time of the correct response increased from 138 ms to 208 ms with visual and tactile feedback, and from 175 ms to 229 ms for tactile feedback only (Fig. 4B).

To understand the patterns of muscle activation underlying this behavior, we analyzed anterior deltoid EMG in anti-reach trials in a 25-ms bin following the median divergence time in reach trials (Fig. 1E). When participants had only tactile feedback, a one-way ANOVA showed a significant effect of direction of target jump (F_2,38 = 15.18; P = 1.43e⁻⁵). Post hoc tests revealed elevated EMG activity compared with no-jump trials for both CW and CCW target jumps, whereas there was no significant difference between EMG activity for CW and CCW target jumps at this latency, indicating a change in the early tactile response from direction specific in reach trials to non-direction specific in anti-reach trials (Fig. 3F; see Fig. 7D). The early response in trials with visual and tactile feedback remained direction specific, as shown by post hoc tests on one-way ANOVA for direction (F_2,38 = 15.18; P = 1.43e⁻⁵; Fig. 3E; see Fig. 7B), which results in an initial movement in the incorrect direction during the early response phase. We tested the difference between CW and CCW muscle activation using t tests against 0 for each feedback condition; although the difference was >0 for vision and touch (T₁₉ = 6.39; P = 3.94e⁻⁶), indicating direction specificity, the difference was not different from 0 for touch only (T₁₉ = 0.67; P = 0.51).

Fig. 7.

Average electromyographic (EMG) activity in 25-ms window following divergence time. A–D: average EMG activity in Experiment 1 (edge stimulus; n = 20 participants). E–H: average EMG activity in Experiment 3 (texture stimulus; n = 20 participants). A and E: average EMG activity under the reach instruction with visual and tactile feedback. B and F: average EMG activity under the anti-reach instruction with visual and tactile feedback. C and G: average EMG activity under the reach instruction with only tactile feedback. D and H: average EMG activity under the anti-reach instruction with only tactile feedback. Marker colors refer to direction of target jump (CW, clockwise; CCW, counterclockwise) or trials with no target jump (None). Exp, experiment.

Finally, we checked whether participants who showed larger EMG responses during pro-reach movements showed larger EMG responses during anti-reach movements. Under visual and tactile feedback (Fig. 5B), reaches to CW target jumps were positively correlated with anti-reaches to CW target jumps (r = 0.46; P = 0.042) and reaches to CCW target jumps were positively correlated with anti-reaches to CCW target jumps (r = 0.46; P = 0.045), indicating the initial incorrect component of anti-reaches was related to behavior during the reaches when participants had visual and tactile feedback. When participants had tactile feedback only (Fig. 5C), EMG during reach movements to CCW target jumps was positively correlated to anti-reach movements to CCW target jumps (r = 0.63; P = 0.003), but EMG during reach movements to CW target jumps showed no significant correlation to anti-reach movements toward CW target jumps (r = 0.31; P = 0.19) as a result of the non-direction-specific EMG response during anti-reaches with tactile feedback only.

Vision and touch do not interfere during anti-reach.

Given the differing responses to anti-reach between visual and tactile feedback and tactile feedback only, we tested whether vision and touch were interfering with one another during anti-reach. If this were the case, we would expect the early movement in the incorrect direction to be even larger when participants had only visual feedback. We recruited an additional 16 participants to perform anti-reach tasks using vision only, vision and touch (as in Experiment 1), and touch only. We observed qualitatively similar kinematic responses to the previous participants in both touch-only and vision-and-touch conditions. There were no obvious differences in crossover between trials with visual feedback only and those with visual and tactile feedback (Fig. 6B). We analyzed the difference in anterior deltoid EMG between CW and CCW target jumps for all three feedback conditions using the same 25-ms analysis windows used during pro-reaches in Experiment 1 for visual and tactile (for vision only and vision and touch) or tactile only (for touch only) using a one-way ANOVA on feedback. This test was significant (F_2,30 = 15.31; P = 2.62e⁻⁵), but post hoc analyses showed that both visual and visual-and-tactile feedback trials were different from tactile feedback trials, while visual feedback trials were not significantly different from visual-and-tactile feedback trials.

The tactile-motor response does not depend on persistent orientation information.

The previous experiments described above and in Pruszynski et al. (2016) used an edge spanning the finger pad as a tactile stimulus. The edge provides information about the position of the target both while it is moving and during steady state because the edge remains in line with the target at the end of the movement (Fig. 1C). We therefore tested whether this persistent steady-state information was necessary to engage this cross-effector tactile-motor response by replacing the edge stimulus with a sandpaper sheet (320 grit; average particle size < 50 μm). When participants use the sandpaper, they receive a cue over the 50 ms when the target is moving but no information about the location of the target after the target stops moving.

We recruited 20 participants (who had not participated in Experiments 1 or 2) to perform reaching and anti-reaching tasks using visual and tactile feedback or tactile feedback only. Experiments were methodologically identical to the first experiment, except the tactile feedback was modified as previously described. When participants performed the pro-reach, we found median divergence times for EMG activity that were similar to those observed in Experiment 1: 76.8 ms when participants had visual and tactile feedback, and 96.5 ms when they had tactile feedback only (Fig. 4A, dashed lines). During the reach instruction, participants showed EMG responses that were specific to the direction of the target jump (one-way ANOVA on direction for visual and tactile: F_2,38 = 60.15; P = 1.68e⁻¹²; for tactile only: F_2,38 = 28.79; P = 2.44e⁻⁸; Fig. 7, E and G). As in Experiment 1, post hoc tests indicated that there was no consistent difference between CW and no-movement trials for tactile feedback only, while all directions were significantly different from each other with visual and tactile feedback. Comparing the differences in muscle activity for CW and CCW target jumps revealed a larger difference in vision-and-touch trials than in touch-only trials (T₁₉ = 3.54; P = 0.0022).

When participants were instructed to perform anti-reaches with the new tactile stimulus, they again showed the kinematic patterns we saw previously. Using both visual and tactile cues, participants made early movements toward the target before moving in the instructed direction away from the target. When they had only tactile stimuli, however, early movements toward the target were not evident, although they again showed a delayed reaching response compared with the pro-reach (Fig. 4C). When participants had both visual and tactile feedback, the median time it took to produce a movement in the correct direction across all trials increased from 146 ms to 225 ms, and when they had tactile feedback only, median time increased from 192 ms to 250 ms. When participants only had the tactile stimulus, we still observed EMG activity that varied depending on the direction of the target jump in the early epoch defined by the pro-reach divergence (one-way ANOVA: F_2,38 = 6.038; P = 0.0053). Post hoc tests indicated that EMG response did not differ during this epoch between CW and CCW target jumps; while CCW target-jump EMG was different from no-jump EMG, the difference between CW and no jump was also not significant (Fig. 7H). The difference in EMG activity between CW and CCW target jumps was significantly different from zero for tactile and visual feedback (T₁₉ = 5.49; P = 2.68e⁻⁵), but it was not significantly different from zero for tactile feedback only (T₁₉ = 1.52; P = 0.15).

DISCUSSION

The study of tactile sensation is heavily influenced by our knowledge of vision. The tactile and visual systems are capable of extracting similar stimulus features (Pack and Bensmaia 2015), and parallels between visual and tactile processing yield interesting results (Hsu et al. 2019; Pei et al. 2008; Pruszynski et al. 2018). Our results caution against general inferences about the properties of the tactile motor system based on the visual motor system. That is, we show that although tactile and visual stimuli elicit rapid motor responses that appear similar at the outset (Pruszynski et al. 2016), they are categorically different in their susceptibility to voluntary modification.

We had participants perform reaching movements to a physical target, making online corrections when the target moved during their reach using either tactile feedback provided by an edge under their nonreaching thumb or index finger, or visual feedback in addition to tactile feedback. We also instructed the same participants to reach in the opposite direction of the target movement to perform a classical anti-reach paradigm. When participants had access to visual feedback during the anti-reach, they erroneously initiated movements in the direction of the target jump, consistent with previous studies (Day and Lyon 2000; Gu et al. 2016; Pisella et al. 2000; Saijo et al. 2005), which suggests that there is something automatic and difficult to suppress about such action (Goodale et al. 1986; Paulignan et al. 1991; Pélisson et al. 1986, Prablanc and Martin 1992; Veerman et al. 2008). When participants made reaches with tactile feedback only, however, they usually did not initiate movement in the wrong direction. Although tactile feedback did not result in earlier onset of movement in the instructed direction, these kinematic changes occurred as a result of qualitative changes in muscle activation patterns at low latencies (<100 ms) following target movement. Both of these results held when another cohort of participants used a rotating texture stimulus rather than the oriented edge.

Based on the heterogeneity of tactile-motor interactions present in everyday behavior, we predicted that this response could be remapped during an anti-reach instruction to react as quickly as observed during a normal reach toward the new target location (i.e., pro-reach). In contrast to our predictions, rapid reaching corrections elicited by a tactile stimulus are not entirely flexible. During anti-reach tasks using tactile stimuli, kinematic corrections play out on a timescale similar to that observed in visually guided anti-reaching. In contrast to these visually guided reaches that show minimal malleability in early responses during anti-reach, however, when participants performed anti-reaches with tactile feedback, their early muscle activity changed from a pattern of agonist excitation and antagonist inhibition (as in reaching toward the target) to generalized excitation. Together, these results suggest that the circuits involved in updating reaching trajectories in response to tactile inputs do implement spatial remapping but that this implementation is different from that in circuits integrating visual stimuli for the same purpose.

The tactile-motor response relies on relative motion cues rather than edge orientation.

The previously described rapid tactile-motor response (Pruszynski et al. 2016) could be supported by multiple features extracted early in the periphery. One possibility is that this response is based directly on recognizing the orientation of edges, which can be extracted very early in the tactile periphery (Pruszynski and Johansson 2014). The edge used in Experiments 1 and 2 was long enough that it could be harnessed for rapid voluntary movements with high angular precision to drive the behavioral response (as shown in Pruszynski et al. 2018). If this were the mechanism used, the anti-reach response could also be remapped at the periphery (rather than requiring a coordinate transformation for the response) by attending to only one part of the edge: in Experiments 1 and 2, the center of rotation of the edge was in the middle of the fingertip, so some parts of the stimulus (those on the finger pad distal to the center of rotation) moved in the same direction as the target, while others (proximal to the center of rotation) moved in the opposite direction (Fig. 1C). If the central nervous system were able to selectively attend to these patterns of tactile stimulation, it would provide a mechanism by which tactile-motor responses could rapidly implement changes in sensorimotor transformations via unambiguous tactile input. This mechanism would obviate the delay in producing an appropriate response observed in visually mediated anti-reach and anti-saccade tasks, which has classically been attributed to increased processing times required to identify the stimulus direction and then reverse it.

Another possibility is that the tactile-motor response relies on relative motion cues, similar to the low-latency responses observed when objects slip under the fingertips (Johansson and Westling 1988). Such responses can be directionally tuned and modified with respect to gravity (Häger-Ross et al. 1996) and are observed bimanually and across fingers in some tasks (Cole et al. 1984; Ohki and Johansson 1999). If this were the case, remapping sensory inputs to behavioral outputs, as required in the anti-reach task, is more likely to require higher executive function similar to that posited as the reason for delayed responses in anti-saccade tasks. Our results suggest that a circuit utilizing relative motion cues and subsequently performing coordinate transformation is more likely to underlie the tactile-motor response. The most convincing reason to favor this explanation is Experiment 3, in which we show that an edge is unnecessary to elicit the tactile-motor response and that similar reach and anti-reach behavior are elicited when participants receive only a relative motion cue. While this leaves open the possibility that the tactile motor response can use edges when they are available, little appears to change in terms of response latency when participants receive only the motion cue (Fig. 4). Furthermore, the edge extraction hypothesis would suggest that remapping of the anti-reach behavior should be faster than we observed (because the relevant part of the edge stimulus could be extracted immediately), potentially being as fast as pro-reach. In contrast to this hypothesis, we observed delayed anti-reach behavior when participants had tactile feedback only. Our experiments do not speak to whether the tactile-motor response could be elicited via edge orientation alone; this would require the delivery of edge orientation without relative motion, potentially using a tactile matrix display.

Flexibility of touch-guided sensorimotor associations.

When participants performed an anti-reach task using touch, they were not able to immediately reverse their pro-reaching behavior to reach in the opposite direction at a similar latency. However, unlike anti-reaching with visual feedback, participants showed categorically different muscle responses under the anti-reach instruction. Instead of the direction-specific activation of agonist musculature and inhibition of antagonists they showed during the reach, participants showed a global increase of muscle activity during anti-reach with tactile feedback only. This contrasts with the muted direction-specific response toward the target that participants showed during anti-reach with visual and visual and tactile feedback (similar to that observed by Day and Lyon 2000). The behavior we observed when participants had access only to touch could be a building block of flexible motor responses to tactile stimuli and indicates that the tactile motor response is more malleable than visually guided updates to reaching. Critically, while we observed the tactile-motor response shift from a direction-specific response to a response that was not direction specific, even this amount of modulation is beyond that afforded by visuomotor responses of the same latency. Pisella et al. (2000) used a visual no-go task where participants were instructed to interrupt their reach if a visual target moved; participants were still unable to suppress the initial correction toward the new target location, indicating that circuits implementing visually guided reaching are not able to support a non-direction-specific response even under explicit instruction.

The limited short-term flexibility we observe in associating tactile stimuli with motor outputs could be related to understanding the dynamics of the system linking tactile stimuli to those outputs. Baugh et al. (2012) showed that reaction-time responses to a visuomotor rotation are lowered, and the initial direction of the correction is more frequently correct, if a virtual tool (a pivoting link) is shown to link a participant’s hand position to a cursor. The mapping between tactile stimulus and motor output is highly variable as we use different objects or even readjust our grip on the same object. As such, a relatively safe “default” response to novel sensorimotor association involving tactile stimulus could be increased generalized muscle activity. This is consistent with responses to rapid changes in fingertip loading (Johansson and Westling 1988), which is observed across hands if both are active in the task (Ohki and Johansson 1999) and across some uninvolved fingers (Cole et al. 1984). This generalized muscle response could then be further reshaped as we learn the dynamics of the object we are handling; for example, when fingertip skin is stretched, objects feel stiffer and predictive grip force slowly changes over the course of practice (Farajian et al. 2020). Further study of whether (and how completely) new tactile-motor associations can be developed over prolonged training and whether they induce aftereffects could help elucidate how tactile-motor associations are built and retained.

Cortical contributions and automaticity.

A major difference between the visual and tactile tasks used here is the visual task is unimanual while the tactile task is bimanual. Although there are many bimanual tasks that use tactile stimuli from one hand to guide action by the other, like attaching a leash to a dog’s collar, it is not obvious how similar this is to visually guided reaching. Rapid corrections to visually guided reaches appear to be mediated by subcortical circuits (Alstermark et al. 1987; Day and Brown 2001), and crossed tactile responses might easily be assumed to require cortical input. In fact, the rapidity with which the crossed tactile response occurs (<100 ms) suggests that it might also have a subcortical basis, and crossed responses to stimuli delivered to the arms (Muraoka and Kurtzer 2020) and fingers (Khong et al. 2020) are observed at similar and even lower latencies. In our experiments, trajectory updates when participants used tactile information were slightly slower than when they also had visual information, which could support the idea that additional supraspinal processing is required for tactile signals. However, this could also reflect a myriad of differences between the tactile and visual processing pathways that our study cannot address.

Aside from the neural origin, however, there remains a question as to whether some aspect of the rapid tactile-motor response studied here can properly be called a “reflex.” In the case of visually guided reaching, there is strong evidence that the early response is automatic in the sense that it cannot be suppressed. Is the tactile-motor response we investigated here similarly automatic? The flexibility of the response, both as described here and the sensitivity to displacement described in previous work (Pruszynski et al. 2016), could be taken as an argument for it not being automatic but rather the outcome of rapid sensory integration (cf. Crevecoeur et al. 2017). On the other hand, numerous behaviors nominally called reflexes across a number of sensory modalities show extensive flexibility. The delay we observed during the anti-reach with tactile information could be taken as evidence that one epoch of the response is automatic and the later one is not, but we have no firm empirical data to address this point, nor is there clear agreement on what such data would look like or even whether such a distinction is meaningful (Prochazka et al. 2000).

Limitations

One difficulty in comparing tactile-motor and visuomotor behaviors is that it is unclear how to match the salience of tactile and visual stimuli. Our experiments made no concerted effort to match salience, and participants generally reported more confidence in their abilities when they had visual feedback; we observed no difference in Experiment 2 when participants had only visual feedback compared with when they had visual and tactile feedback. In addition, while our results in Experiment 3 indicate that relative motion is sufficient to evoke this tactile-motor response, they do not speak to whether it is necessary, as we did not apply an oriented edge stimulus without relative motion. Participants also performed all experiments with complete knowledge of how the movement of the tactile stimulus was related to target movement. It is possible that the tactile motor response could be more completely remapped if participants did not know how the stimulus and target movement were related (e.g., by hiding the apparatus). Similarly, participants might be able to adapt more quickly or completely to a real machine that dissociates stimulus and target position (for example, using a hinge). Finally, our experiment did not test the effect of prolonged training on the remapping of the tactile-motor response.

GRANTS

This work was funded by a Canadian Institutes of Health Research Foundation grant (to J.A.P.). The Polhemus tracker was purchased by a National Sciences and Engineering Research Council (NSERC) Research Tools and Instruments grant (to Dr. Brian Corneil). S.R. received a salary award from Western University’s BrainsCAN program through the Canada First Research Excellence Fund (CFREF). J.A.P. received a salary award from the Canadian Research Chair Program.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

S.R. and J.A.P. conceived and designed research; S.R. performed experiments; S.R. analyzed data; S.R. and J.A.P. interpreted results of experiments; S.R. prepared figures; S.R. drafted manuscript; S.R. and J.A.P. edited and revised manuscript; S.R. and J.A.P. approved final version of manuscript.