Interference underlies attenuation upon relearning in sensorimotor adaptation

Abstract

Savings refers to the gain in performance upon relearning a task. In sensorimotor adaptation, savings is tested by having participants adapt to perturbed feedback and, following a washout block during which the system resets to baseline, presenting the same perturbation again. While savings has been observed with these tasks, we have shown that the contribution from implicit sensorimotor adaptation, a process that uses sensory prediction errors to recalibrate the sensorimotor map, is actually attenuated upon relearning (Avraham et al., 2021). In the present study, we test the hypothesis that this attenuation is due to interference arising from the washout block, and more generally, from experience with a different relationship between the movement and the feedback. In standard adaptation studies, removing the perturbation at the start of the washout block results in a salient error signal in the opposite direction to that observed during learning. As a starting point, we replicated the finding that implicit adaptation is attenuated following a washout period in which the feedback now signals a salient opposite error. When we eliminated visual feedback during washout, implicit adaptation was no longer attenuated upon relearning, consistent with the interference hypothesis. Next, we eliminated the salient error during washout by gradually decreasing the perturbation, creating a scenario in which the perceived errors fell within the range associated with motor noise. Nonetheless, attenuation was still prominent. Inspired by this observation, we tested participants with an extended experience with veridical feedback during an initial baseline phase and found that this was sufficient to cause robust attenuation of implicit adaptation during the first exposure to the perturbation. This effect was context-specific: It did not generalize to movements that were not associated with the interfering feedback. Taken together, these results show that the implicit sensorimotor adaptation system is highly sensitive to memory interference from a recent experience with a discrepant action-outcome contingency.

Fig 1.. Experiment 1: Upon relearning a visuomotor rotation, implicit adaptation is attenuated.

(A) Task design and schematics of all trial types in Experiment 1. Using a trackpad or mouse, participants (N=44) moved a cursor from the start location (white circle) to a target (blue disk), with the target appearing at one of four locations (one representative location is depicted). There were 3 types of trials: 1) No feedback, with the cursor disappearing at movement onset; 2) Veridical feedback, in which the direction of the cursor (small white disk) was veridical with respect to the direction of the movement; 3) Clamped feedback, in which the cursor followed an invariant path with respect to the target. (B) Top: Experimental protocol. The -C|CV|-C abbreviation indicates the main block-level structure of the experiment. There were two learning blocks with clamped feedback (-C), each followed by an aftereffect block with no feedback. To reset the sensorimotor map following the first learning block, we used a washout block composed of a reversed-clamp feedback phase (C) and a veridical feedback phase (V). The green oblique lines in the washout block mark the transition between the two phases, with the length of the reversed clamp phase determined on an individual basis (see Methods). Bottom: Time course of mean reach angle, with the data averaged within each cycle of four movements. Light and dark shading signify learning blocks 1 and 2, respectively, with the onset of the clamped feedback marked by the vertical solid lines. (C) Overlaid reach angle functions for the two learning blocks and two aftereffect blocks. Horizontal thick black lines denote clusters that show a significant difference between blocks 1 and 2 (p < 0.05). (D) The left panel (pair of bars) presents the aftereffect data (mean ± SEM) for each learning block, measured as the averaged reach angle across all cycles in each aftereffect block. The right panel shows the within-participant difference (Aftereffect 2 – Aftereffect 1; dots and violin plot-distribution of individual difference scores, bar- mean difference and 95% CI). SEM, standard error of the mean. CI, Confidence Interval.

Fig 2.. Experiment 2: Implicit adaptation is not attenuated upon relearning when feedback is eliminated during the washout block.

(A) Experimental protocol and learning functions. Top: Participants (N=44) experienced 110 cycles of trials without feedback (N) during a washout block that separated the two learning blocks (-C|N|-C design, purple). Bottom: Time course of mean reach angle averaged over cycles. Light and dark shading signify learning blocks 1 and 2, respectively, with the onset of the clamped feedback marked by the vertical solid lines. The design and learning functions from Experiment 1 are reproduced here to provide a visual point of comparison (gray). (B) Overlaid reach angle functions for the two learning blocks and two aftereffect blocks in Experiment 2 (significant clusters based on p<0.05) (C) The left panel presents the aftereffect (mean ± SEM) for each learning block and the right panel the within-participant difference (Aftereffect 2 – Aftereffect 1; dots and violin plot- distribution of individual difference scores, bar- mean and 95% CI). (D) Scatter plot showing no relationship between the reach angle at late washout and change in relearning (Aftereffect 2 – Aftereffect 1). SEM, standard error of the mean. CI, Confidence Interval.

Fig 3.. Experiment 3: Attenuated adaptation does not require experience with salient, opposite signed error at the beginning of washout.

(A) Experimental protocol and learning functions. Top: At the beginning of the washout block in Experiment 3, participants (N=44) experienced a rotated cursor that was contingent on the direction of their hand movement, with the magnitude of the perturbation set to their final adaptation level at the end of the first learning block; in this way, the cursor position would be near the target. The size of the rotation was gradually decreased until reaching 0°, at which point it was veridical and remained so for the rest of the washout block (-C|-R_GV|-C design, blue). Bottom: Time course of mean reach angle averaged over cycles (4 movements). Light and dark shading signify learning blocks 1 and 2, respectively, with the onset of the clamped feedback marked by the solid vertical lines. The design and learning functions from Experiment 1 are reproduced here to provide a visual point of comparison (gray). (B) Distribution of errors experienced during the non-zero rotation phase of the washout block. These errors were small in magnitude (mean=2.9°, dark blue solid line) and in the opposite direction of the error experienced during the initial learning block (black solid line). Presumably, these opposite errors are the signals that drive the washout of the initial adaptation. The dotted line represents zero error. (C) Overlaid reach angle functions for the two learning blocks. Horizontal thick black lines denote clusters that show a significant difference between blocks 1 and 2 (p < 0.05). (D) The left panel presents the aftereffect (mean ± SEM) for each learning block and the right panel the within-participant difference (Aftereffect 2 – Aftereffect 1; dots and violin plot-distribution of individual difference scores, bar- mean and 95% CI). SEM, standard error of the mean. CI, Confidence Interval.

Fig 4.. Experiment 4: Adaptation is attenuated for movements that were previously associated with either washout after learning or an extended baseline experience with veridical feedback.

(A) Experimental protocol and learning functions. Top: The target appeared at one of four locations, with two locations falling within one 30°-wedge and the other two falling within a same-size wedge on the opposite side of the workspace. Participants (N=44) experienced a -C|CV|-C schedule (cyan) in one wedge and a V₈₅|-C design schedule in the other wedge (orange). For the latter, the number of veridical feedback cycles (85) matched the total number of cycles before relearning at the other wedge (excluding the no-feedback trials). Bottom: Time course of mean reach angle averaged over cycles, with each cycle consisting of 2 movements for each wedge. Light and dark cyan signify learning blocks 1 and 2 in the C|CV|-C condition. (B) Overlaid reach angle functions comparing the first learning block in -C|CV|-C to that of the second learning block in the same wedge (left panel) and to the post long-baseline learning block at the other wedge (right panel). Horizontal thick black lines (B) and (C) denote clusters that show a significant difference between the functions (with p < 0.017 as a significance criterion, see Methods). (C) Left panel (bars) presents the aftereffect data (mean ± SEM) for each learning block and the right panel shows the within-participant differences for three contrasts: 1) Aftereffect 2 – Aftereffect 1; 2) Aftereffect after long baseline – Aftereffect 1; 3) Aftereffect after long baseline – Aftereffect 2. Dots and violin plots show the distribution of individual difference scores; bar- mean and 95% CI. SEM, standard error of the mean. CI, Confidence Interval.

Fig 5.. Experiment 5: Interference from veridical feedback is context-specific.

(A) Experimental protocol and learning functions. Top: During the learning block, participants (N=60) experienced rotated clamped feedback while reaching to six targets, with two targets falling within each of three wedges distributed across the workspace. For each wedge, participants experienced a different number of cycles with veridical feedback prior to the learning block: 5 (V₅|-C, light red); 45 (V₄₅|-C, medium red); 85 (V₈₅|-C, dark red). Bottom: Time course of mean reach angle averaged over cycles (2 movements) for each wedge. The vertical solid lines at cycles 45 and 85 mark the onset of movements to an additional location, and the vertical solid lines at cycles 90 mark the onset of the task-irrelevant clamped feedback. Horizontal thick black lines denote clusters of cycles that show a significant relationship between the reach angle and the number of veridical cycles (p < 0.05). (B) Left panel presents the aftereffect results (mean ± SEM) for each learning condition with the fixed effect regression line obtained using a linear mixed model. Right panel shows the distribution (dots and violin plots) of individuals’ slopes (random effect); bar- mean slope and 95% CI. SEM, standard error of the mean. CI, Confidence Interval.