Functional electrical stimulation (FES) can electrically activate a set of muscles selected to address individual movement deficits with a pre-programmed pattern of electrical stimulation 1 2 . Users normally employ a switch to manually trigger each pre-programmed stimulation pattern, but triggering and/or modulation of the electrical stimulation using residual electromyogram (EMG) from the paretic muscle - which is an alternative option to control FES - may encourage sensory-motor integration, where the residual volitional effort is reinforced with FES-assisted functional movement 3 , and thus fostering re-learning of self-initiated movements. Unfortunately the muscle activity in hemiparetic limbs often suffers from a lack of coordination and delays in initiation/termination 4 . These deficits, which likely are controlled by the central nervous system (CNS), might be alleviated with an appropriate adjuvant treatment that improves CNS function. One possibly suited adjuvant treatment at the CNS level to facilitate learning of myoelectric control for brain machine interfaces is transcranial direct current stimulation (tDCS), which induces cortical excitability changes 5 6 7 8 induces neuroplasticity 9 , and has been shown to improve motor learning in healthy humans 10 11 , as well as in stroke survivors 12 13 . Therefore tDCS in combination with rehabilitative therapy has been suggested for stroke rehabilitation 14 15 16 . However tDCS-facilitated motor learning in lower limbs has not been explored systematically. Its effects on initiation and termination of muscle activations need further investigation to determine an appropriate adjuvant treatment with tDCS that may help in facilitating myoelectric control of FES. Tanaka and colleagues 17 found that anodal tDCS of the primary motor cortex representation of the tibialis anterior (TA) muscle (M1) had no significant effects on reaction time, but transiently enhanced maximal leg pinch force. Also, Madhavan and colleagues 18 found that M1 anodal tDCS of the primary motor representation of TA muscle applied to the lesioned motor cortex of moderate to well recovered stroke patients enhanced voluntary control of the paretic ankle. However, Galea and colleagues 19 20 did not observe any changes of reaction time with either M1 or cerebellar anodal tDCS.

In order to understand and further investigate the effects of tDCS on EMG latencies, we followed the general feedback-error-learning model 21 where both, the M1 and the cerebellum, are presumed to mediate generation of force profiles during manual tasks 22 23 , as illustrated in Figure 1. The model incorporates three basic elements: 1. an inverse model that captures the feedforward part, 2. a feedback controller that captures the feedback part, and 3. a learning rule that adapts the inverse model based on motor command errors. The inverse (feedforward) model is primarily associated with the cerebellum and the feedback controller is primarily associated with premotor/motor cortices 23 . In this study, we investigated volitonal control of EMG during isometric conditions that reflects muscle force quite well 24 . Specifically, we investigated the impact of anodal tDCS of M1 and cerebellum on two commonly used myoelectric control paradigms for FES control 25 , which are initiation/termination of muscle activation, i.e., 'ballistic EMG control’ for switching FES on-off with a threshold-based classifier 26 and modulation of EMG for 'proportional EMG control’ of FES 27 . The myoelectric visual biofeedback was presented with proportional system dynamics, where the subjects had to modulate the EMG activity (here, EMG is the system input) from one level to another in a finite time. In this randomized sham-controlled study, we specifically explored two cases: 1. the effects of offline anodal tDCS of M1 and cerebellum on delays in initiation (step-up response) and termination (step-down response) of muscle activity to a visual on/off cue with maximal contraction of isometric TA muscle during 'ballistic EMG control’, 2. the effects of online anodal tDCS of M1 and cerebellum on learning visual pursuit while following a sinusoidal target with EMG from TA during 'proportional EMG control’.

In this study, the motor control task involved visual pursuit of a TARGET signal with EMG-based proportional control of a visual TRACKING signal. Prior work has shown that EMG reflects muscle force quite well during isometric conditions where EMG follows a quadratic increase in its root-mean-square value across force levels 24 . In Experiment 1, the myoelectric step-response task was familiar to the subjects but the trials were presented in an unpredictable temporal manner during baseline and post-intervention to avoid cognitive anticipation, and to identify the delay in initiation and termination of muscle activity during an open-loop 'ballistic EMG control’ task. We first ruled out subject specific effects on the initiation and termination of muscle activity during baseline trials, and then found that termination of muscle activity was significantly (p < 0.05) slower than initiation of muscle activity for all subjects over all baseline trials. We also found that cerebellar anodal tDCS increased the normalized delay in initiation of muscle activity post-intervention while M1 anodal tDCS decreased it, when compared to sham tDCS, as shown in the top panel of Figure 5. Also, M1 anodal tDCS increased the normalized delay in termination of muscle activity post-intervention when compared to cerebellar anodal tDCS and sham tDCS, as shown in the bottom panel of Figure 5. Therefore in this study, off-line anodal tDCS of M1 decreased the delay in initiation while it increased the delay in termination during performance of the 'ballistic EMG control’ task. However Galea and colleagues 20 did not observe any changes in reaction time with either M1 or cerebellar anodal tDCS during a more complex task where the subjects had to move a digitizing pen with their right hand over a horizontal digitizing tablet. The outcome differences could be caused by different EMG recordings in the respective studies. In the present study, EMG latencies were obtained only for the excitation dynamics of the target muscle. Excitation and contraction processes of multiple muscles crossing a joint and subsequent joint mechanics, as recorded in the study by Galea and colleagues 20 , might limit comparability. Moreover, the premotor cortex might have played an important role in the rather complex (fine) motor task movements employed by Galea and colleagues 20 that required controlled contraction of multiple muscles compared to the control of a single muscle in this study. Here, M1 is involved in task performance in part through its reciprocal interaction with the cerebellum 34 . Cerebellar priming of M1 plasticity in shaping the impending motor command by favoring or inhibiting the recruitment of several muscle representations has been shown recently 35 . Prior work has suggested that cerebellar anodal tDCS may increase the Purkinje cells’ excitability and facilitate the inhibitory tone the cerebellum exerts over M1 (cerebellar brain inhibition) 19 , which explains an increase in normalized delay in initiation of muscle activity following cerebellar anodal tDCS and a respective decrease following M1 anodal tDCS. Conversely, cerebellar-caused M1 inhibition should then facilitate sudden termination of ongoing muscle activity. In principal accordance with this concept, cerebellar anodal tDCS trended towards decreasing the normalized delay in termination of muscle activity post-intervention, while M1 anodal tDCS significantly increased it, when compared to sham tDCS.

In Experiment 2, naive subject learned a novel visual pursuit task. Subjects had to minimize spatial ERROR between the TARGET signal and the TRACKING signal using visual feedback. The response latency was high at the start of myoelectric training during 'proportional EMG control’, but reduced with training as the TRACKING response signal temporally shifted with respect to the TARGET signal, as shown in the top panel of Figure 6. The visual pursuit was initially driven primarily with feedback motor command where the feedback motor command also served as the motor command error for developing or modifying the inverse model for this novel visuomotor transformation 23 as the subject learned proportional dynamics of the myoelectric visual pursuit. The feedback control is inherently slow because it uses delayed sensory (e.g., proprioceptive and visual) signals to compute the motor command 36 37 but feedforward control uses the inverse model to predict the (feedforward) motor command necessary to pursue the TARGET signal, which should increase the speed and accuracy during visual pursuit if the inverse model is accurate 21 36 . However, it was found that the absolute value of the power law exponent for the mean absolute ERROR of the cerebellar tDCS group was lower than other tDCS groups (see Table 1), which indicated slower motor learning (bottom panel of Figure 6) during 'proportional EMG control.’ This is in contrast with the results from Galea and colleagues 18 , which may be explained in terms of a computational model of human motor learning. In the computational model of the cerebellar circuit, the simple spikes represent the feedforward motor commands and the parallel fiber inputs represent the desired trajectory (TARGET) as well as the sensory state of the TRACKING signal 21 . The climbing fiber inputs are assumed to carry a copy of the feedback motor commands where the complex spiking of the climbing fibers in the cerebellum are considered to be the biological representation of an error signal. Marko and colleagues 38 recently found that the probabilities of complex spiking declined with increasing error size and therefore they postulated that complex spiking is a representative of the sensitivity to error, and not the error itself. Therefore learning of the inverse model may be dependent on the sensitivity to error, in addition to the magnitude of the error presented during the task performance. From results of Experiment 1, it can be postulated that there was facilitation of cerebello-thalamocortical inhibitory connections at movement initiation 39 with cerebellar anodal tDCS, which might have reduced the sensitivity of the Purkinje cells to errors represented by complex spiking of the climbing fibers that resulted in a slower decrease in the mean absolute ERROR during 'proportional EMG control’ trials. In fact, Galea and colleagues 20 hypothesized that cerebellar tDCS may change Purkinje cells response to the input of the climbing fibers by affecting secondary events such as long-term depression. Moreover, Popa and colleagues recently showed that modulation of the cerebellar cortex by noninvasive brain stimulation affects the response of M1 to a subsequent plasticity induction protocol that involves sensory afferent input but not otherwise 35 .

In terms of clinical applications, the current study on healthy subjects showed a decrease in the normalized delay in initiation of muscle activity following M1 anodal tDCS which may be beneficial for stroke survivors who often suffer from delays in initiation of muscle activity 4 . Therefore an adjuvant treatment with M1 anodal tDCS may facilitate appropriate myoelectric triggering of FES where delays in initiation of muscle activity makes it difficult for EMG-triggered FES to assist time-critical functional tasks such as ankle dorsiflexion during overground walking. However, the optimal positioning of the cerebellar tDCS electrode remains unclear. More studies are required to better define how tDCS over particular regions of the cerebellum affects individual cerebellar sensory-motor functions given its topographical organization 40 . In our future studies, we will investigate optimization of the electrode montage for cerebellar tDCS to target different sensory modalities (e.g., proprioceptive instead of visual), since recent studies in patients with cerebellar damage demonstrated that adaptation to proprioceptive versus visual errors relies on the integrity of different regions of the cerebellum 41 42 . Therefore anodal tDCS-induced changes in excitability of different regions of the cerebellum may differentially affect proprioceptive versus visual sensory modalities 40 .

The preliminary results from healthy subjects showed specific, and at least partially antagonistic effects, of M1 and cerebellar stimulation on motor performance. An appropriate adjuvant treatment with tDCS may help to facilitate myoelectric control for brain machine interfaces, however the neuroprosthetic and neurotherapeutic efficacy of such an adjuvant treatment needs further investigation in stroke survivors. Furthermore, an adjuvant treatment with tDCS may improve muscle recruitment and coordination during post-stroke neurorehabilitation.

A. Dutta and W. Paulus declare that they have no competing interests. M.A. Nitsche declares that he is in the Advisory Board of Neuroelectronics, S.L., Spain.

AD have made substantial contributions to conception and design of the experiments, acquisition, analysis and interpretation of data. AD have been involved in drafting the manuscript. WP and MAN have made contributions to conception and design of the experiments, in revising the manuscript critically for important intellectual content; and have given final approval of the version to be published. All authors read and approved the final manuscript.