Modulation of proprioceptive integration in the motor cortex shapes human motor learning

Abstract

Sensory and motor systems interact closely during movement performance. Furthermore, proprioceptive feedback from ongoing movements provides an important input for successful learning of a new motor skill. Here, we show in humans that attention to proprioceptive input during a purely sensory task can influence subsequent learning of a novel motor task. We applied low-amplitude vibration to the abductor pollicis brevis (APB) muscle of eight healthy volunteers for 15 min while they discriminated either a small change in vibration frequency or the presence of a simultaneous weak cutaneous stimulus. Before and after the sensory attention tasks, we evaluated the following in separate experiments: (1) sensorimotor interaction in the motor cortex by testing the efficacy of proprioceptive input to reduce GABA(A)ergic intracortical inhibition using paired-pulse transcranial magnetic stimulation, and (2) how well the same subjects learned a ballistic thumb abduction task using the APB muscle. Performance of the vibration discrimination task increased the interaction of proprioceptive input with motor cortex excitability in the APB muscle, whereas performance in the cutaneous discrimination task had the opposite effect. There was a significant correlation between the integration of proprioceptive input in the motor cortex and the motor learning gain: increasing the integration of proprioceptive input from the APB increased the rate of motor learning and reduced performance variability, while decreasing proprioceptive integration had opposite effects. These findings suggest that the sensory attention tasks transiently change how proprioceptive input is integrated into the motor cortex and that these sensory changes drive subsequent learning behavior in the human motor cortex.

Figure 1.

Experimental design. Each of the subjects took part in five different experiments, which were separated by several weeks (>2). Experiments 1–3 were behavioral experiments that tested motor learning with or without preceding sensory attention task. Experiments 4 and 5 were neurophysiological experiments that tested the change of SMO as induced by the sensory attention tasks. The order of experiments was randomized for each subject.

Figure 2.

Behavioral data of the sensory attention tasks. Group average of the behavioral data of the sensory attention tasks. A shows the mean error per minute training (±SEM) for AttVIB, and B shows the mean reaction time per minute (±SEM) for AttCUT. The performance did not change significantly in either of the sensory attention tasks over the 15 min period.

Figure 3.

SMO before (baseline) and after the AttVIB and AttCUT tasks. The three graphs plot the SICI measured without vibration (novib) and with vibration of either APB (vibAPB) or FDI (vibFDI) in the APB (A), FDI (B), and ADM (C). SICI data are shown as percentage conditioned/test MEP (±SEM). The two baseline SMO were pooled for purpose of display. At baseline, SICI is reduced (columns going up) in the vibrated muscle, and increased (columns going down) in the nonvibrated muscles. After AttVIB, vibAPB induced an even stronger reduction of SICI in APB, and now reduced SICI in FDI, too. After AttCUT, the effect of vibAPB on APB and FDI was abolished. Neither intervention had an effect on the SMO recorded in ADM, probably because its cortical representation is functionally more separated from the APB and FDI. The asterisks indicate significant differences from corresponding baseline values (paired t tests; *p < 0.005; **p < 0.0001).

Figure 4.

Motor learning. The mean peak acceleration (A) normalized relative to the mean acceleration during the first minute of training (±SEM), and the mean CoVar (±SEM) (B) for the thumb abduction movement per minute of training are shown for the three experiments, without (no intervention) or with either preceding AttVIB or AttCUT. The mean acceleration and CoVar during the first minute of training were not significantly different between the experiments (p = 0.56). A, In the “no-intervention” condition, the mean acceleration increased over the training time (p < 0.00001), indicating a learning effect. This was significantly enhanced by the preceding AttVIB task and significantly decreased by the AttCUT task (p < 0.0001). B, After AttVIB, the variability of performance was significantly reduced with training compared with the condition without preceding intervention (p < 0.001), whereas after AttCUT the performance variability did not change.

Figure 5.

Behavioral-neurophysiological correlation. The two graphs plot the correlation of the peak thumb acceleration and the change to SMO induced by the AttVIB (A) and AttCUT (B) tasks. The linear regression is calculated on the mean peak acceleration data per minute during thumb abduction training (x-axis) versus the difference of SICI (after − before intervention) recorded during vibration of APB in all three hand muscles (y-axis). Both sensory attention tasks induced a correlated effect on SMO and motor learning. After AttVIB, the more APB vibration reduced SICI in APB and FDI (increase in percentage conditioned/test MEP: less SICI), the faster was the increase in thumb acceleration. After AttCUT, the more SICI increased in APB (decrease in percentage conditioned/test MEP: stronger SICI) and decreased in FDI, the slower was the increase of acceleration. For the significant correlations, r² is given.

Figure 6.

Model: modulations of proprioceptive-motor integration influence subsequent motor learning. A, Abduction movements of the thumb evoke proprioceptive movement feedback from the APB muscle. This feedback signal is intrinsically noisy so that, when compared with the position predicted on the basis of efferent copy, the “error” signal is equally noisy. This estimate of error is then used to optimize the motor command for the next movement. Visual feedback of the peak acceleration is provided to the subjects and contributes to this optimization process. B, AttVIB directs the attentional focus of the subjects on proprioceptive input from the APB, potentially by top-down controlled sensory gating processes. Thus, at the start of the motor learning, the noise of the sensory estimate is reduced. Consequently, during learning, proprioceptive input gains easier access to the motor cortex since the variability of the error signal is reduced, which consequently leads to a higher learning gain. C, AttCUT directs subjects' attention away from proprioceptive input. Consequently, during subsequent learning, the sensory estimate and the error signal are not easily modified, which prevents motor learning.