Modulating Motor Learning through Transcranial Direct-Current Stimulation: An Integrative View

Motor learning consists of the ability to improve motor actions through practice playing a major role in the acquisition of skills required for high-performance sports or motor function recovery after brain lesions. During the last decades, it has been reported that transcranial direct-current stimulation (tDCS), consisting in applying weak direct current through the scalp, is able of inducing polarity-specific changes in the excitability of cortical neurons. This low-cost, painless and well-tolerated portable technique has found a wide-spread use in the motor learning domain where it has been successfully applied to enhance motor learning in healthy individuals and for motor recovery after brain lesion as well as in pathological states associated to motor deficits. The main objective of this mini-review is to offer an integrative view about the potential use of tDCS for human motor learning modulation. Furthermore, we introduce the basic mechanisms underlying immediate and long-term effects associated to tDCS along with important considerations about its limitations and progression in recent years.


INTRODUCTION
Motor learning entails improving motor actions through practice (Willingham, 1998;Dayan and Cohen, 2011;Wolpert et al., 2011). We make use of this ability when acquiring new motor skills and when adapting our movements to account for predictable changes to our environment. Motor learning plays a critical role in acquiring the motor actions necessary for high-performance sports (Nielsen and Cohen, 2008) and for motor recovery after brain lesions (Kitago and Krakauer, 2013). Applying weak direct current through the scalp induces polarity-specific changes in the excitability of cortical neurons Brunoni et al., 2012). This effect of transcranial directcurrent stimulation (tDCS) was first demonstrated in the human motor cortex Paulus, 2000, 2001), but has also been described for other brain regions such as visual (Antal et al., 2001(Antal et al., , 2004, somatosensory (Rogalewski et al., 2004;Dieckhöfer et al., 2006), prefrontal (Fregni et al., 2005;Mulquiney et al., 2011) and cerebellar cortices Grimaldi et al., 2014).
The modulatory effects and simplicity of tDCS have caught the attention of both basic and clinical neuroresearchers for its potential to modulate motor learning Nitsche et al., 2003;Antal et al., 2004;Reis et al., 2008;López-Alonso et al., 2015). Most studies using tDCS deliver a low-current intensity (1-2 mA) between two rubber electrodes (25-35 cm 2 ) placed on the scalp for 10-20 min (Stagg and Nitsche, 2011). For this montage, the stimulating electrode is placed over the region of interest while the reference electrode is placed over either the contralateral supraorbital, the mastoid or shoulder. Following this procedure, researchers have utilized tDCS to enhance motor learning in healthy individuals (Reis et al., 2008) and for motor recovery due to brain lesions or pathological states linked to motor deficits (Demirtas-Tatlidede et al., 2012;Grimaldi et al., 2014). tDCS has also been proposed to improve motor capacities and muscle endurance of high-performance sport athletes (Cogiamanian et al., 2007;Banissy and Muggleton, 2013;Williams et al., 2013). Although tDCS application in the motor domain is vast, the main objective of this review is to offer an integrative view of the main findings from studies using cerebral and cerebellar tDCS application in healthy human participants.

BASIC MECHANISMS UNDERLYING tDCS
Although there is increasing interest for using tDCS as a noninvasive neuromodulation technique, little is known about the molecular and/or cellular mechanisms underlying its effects (Márquez-Ruiz et al., 2012). Since Nitsche and Paulus (2000) described the impact of transcranial low current over the human primary motor cortex (M1), excitatory/inhibitory effects have been broadly associated to anodal/cathodal current stimulation, respectively. However, the net effect of tDCS depends on the stimulated brain region (Dieckhöfer et al., 2006), the number of tDCS sessions (Monte-Silva et al., 2013), the applied current intensity (Batsikadze et al., 2013), and the brain state (Silvanto and Pascual-Leone, 2008;Krause and Cohen Kadosh, 2014) among other parameters. To understand the physiological mechanisms underlying these effects, it is important to disassociate: a) the immediate tDCS effects observed in cells exposed to simultaneous exogenous electrical fields and b) effects mediated by protein modifications requiring longer stimulation periods, lasting for several minutes after tDCS application. The immediate effects are elicited when an external electric field causes displacement of intracellular ions, thus altering the internal charge distribution and modifying the neuronal membrane potential (Ruffini et al., 2013;Márquez-Ruiz et al., 2014). Moreover, animal studies have shown both neuronal morphology (Radman et al., 2009) and axonal orientation (Kabakov et al., 2012) are critical to consider when explaining tDCS-induced responses, since the maximal effects occur when electric fields are applied parallel to the somatodendritic axis (Bikson et al., 2004). Beyond these somatic changes, animal studies have also demonstrated the importance of presynaptic effects during current application (Kabakov et al., 2012;Márquez-Ruiz et al., 2012;Bikson et al., 2013). The longterm effects, measured indirectly in human studies (recording motor evoked potentials, MEPs, elicited by transcranial magnetic pulses over M1) are mediated by N-methyl-D-aspartate (NMDA) and γ-aminobutyric acid type A (GABA A ) receptors (see for review Stagg and Nitsche, 2011). Animal studies have confirmed the involvement of NMDA receptors and brainderived neurotrophic factor (BDNF) (Fritsch et al., 2010) for the long-term effects observed after anodal direct-current stimulation (atDCS), and adenosine A1 receptors (Márquez-Ruiz et al., 2012) after cathodal direct-current stimulation (ctDCS).

MODULATING MOTOR LEARNING PROCESSES THROUGH tDCS
Motor learning encompasses various forms of learning, including, but not exclusive to error-based, reinforcement, use-dependent plasticity, and cognitive strategies (Krakauer and Mazzoni, 2011), each likely involving different neuronal substrates. It becomes more complicated given that these forms of learning likely all contribute to the learning process when acquiring a new skill (Kitago and Krakauer, 2013). Therefore, for better comprehensibility, we grouped publications based on different motor learning paradigms and not the different forms of learning, to explore the impact of tDCS on specific motor behaviors (see Table 1). We included adaptation, skill, and use-dependent repetition (i.e., repeated practice of simple movements) tasks. Undoubtedly, the number of positive findings described below, highlight the potential of tDCS for (1) modulating new behavior acquisition and retention, (2) identifying the underlying learning processes, and (3) studying the role of different brain regions.

Modulating Skill Learning
Skill learning refers to a process that results in improving the trade-off between speed and accuracy (Reis et al., 2009), typically achieved by reducing movement variability (Smuelof et al., 2012). Investigations have used tDCS to either modulate learning or to better understand the underlying learning processes (Orban de Xivry and Shadmehr, 2014;Savic and Meier, 2016). However, the number of brain regions involved in skill learning is vast  which has led to various targeted brain regions for tDCS application, electrode montages, and types of motor tasks. The leading paradigms combined with tDCS are motor sequence tasks, including serial reaction time task (SRTT), sequential finger tapping tasks (SFTT), and sequential visual isometric pinch task (SVIPT) (see Table 2 for details).
Several studies have reported enhanced SRTT performance and retention with simultaneously applying atDCS over M1. This is shown by reduced reaction times (RTs), a common way to quantify sequence acquisition Kang and Paik, 2011;Kantak et al., 2012;Ehsani et al., 2016). Comparably reduced RTs were found during the recall of a sequence task when tDCS was applied over premotor (PM) cortex throughout REM sleep (Nitsche et al., 2010). A few studies, however, have presented null effects of tDCS on RTs, specifically when stimulation was not applied during training

Motor task Description
SRTT Participants respond to visual cues presented on a screen by pressing an associated keyboard response. The position of the visual cue is either presented in a repeating sequence or random.
SFTT A specific order of sequence elements is presented on a screen that present specific finger movements. Participants are instructed to make the representative key-presses as fast and accurate as possible.
SVIPT Participants control the movement of a cursor displayed on a computer screen by squeezing an isometric force transducer using the thumb and index finger. The aim is to move the cursor as quickly and accurately as possible between the start position and a numbered order of target zones. The magnitude of pinch force applied to the sensor is non-linearly (usually a logarithmic transduction is applied) related to the displacement of the cursor.
VPFT Similar to the SVIPT, participants match their own pinch force visually displayed by a force bar on a computer screen with the height of a moving reference bar by squeezing a force transducer.
VAT Participants make hand-reaching movements with a pen over a horizontal digitizing tablet to respond to a target displayed on a vertical screen. Vision of the hand was not visible to participants, but a cursor on the screen was given to participants to represent the position of their hand. Participants are instructed to make rapid and straight uncorrected movements throughout training. After some practice, a perturbation is introduced by applying a visual rotation (e.g., by 30 • counterclockwise) of the cursor. Participants adapt incrementally their movements to the new position and show large and prolonged after-effects once the perturbation is removed.
Force fields Participants hold a robotic arm handle in order to make reaching movements to a specific target displayed on a screen. Vision of the hand was obstructed, however, visual feedback of hand position is provided on the screen. After baseline performance, reaching is perturbed by a force field that pushes the hand perpendicular to the direction of movement. After participants adapt to the force field perturbation, participants show large after-effects when the perturbation is removed.
SFTT, sequential finger tapping task; SRTT, serial reaction time task; SVIPT, sequential visual isometric pinch task; VAT, visuomotor adaptation task; VPFT, visual pinch force task. , or when tDCS was combined with singlepulse TMS, causing a potential reduction of tDCS' efficacy (Ambrus et al., 2016). Moreover, when tDCS was applied over PM during SRTT, neither acquisition nor consolidation was modulated , but instead interfered with the retention of learned sequences (Kantak et al., 2012). In contrast, when PM-tDCS was applied while participants watched a video of a hand performing key-press sequences prior to training, RTs were reduced in comparison to sham stimulation. This suggests that increasing excitability of a region involved in action observation promotes skill acquisition (Wade and Hammond, 2015). Additional studies have revealed significant benefits of tDCS on SFTT learning. Interestingly, the number of correctly executed sequences increased both when M1-tDCS was applied concurrently with performance (Saucedo Marquez et al., 2013), and when tDCS was applied during motor imagery of sequences (Saimpont et al., 2016). When individuals received M1-atDCS during performance, RTs decreased during training , whereas when M1-atDCS was applied between two training sessions, reduced execution time of correct sequences was found during early consolidation (Tecchio et al., 2010), i.e., stabilization of the motor memory rapidly after its initial acquisition (Brashers-Krug et al., 1996). Together, this suggests M1 as an important site for storage of motor sequences. On the other hand, the role of the cerebellum, a structure critical for motor adaptation (Tseng et al., 2007;Donchin et al., 2012;Izawa et al., 2012), is not well understood for procedural sequence learning (Jenkins et al., 1994;Doyon et al., 2002;Shimizu et al., 2016). Only a few studies have addressed the effects of cerebellar atDCS on sequence learning. For example, cerebellar stimulation applied during SRTT performance reduced the error rate (Ehsani et al., 2016), whereas it reduced RTs when applied prior to a follow-up session (Ferrucci et al., 2013). Interestingly, both M1 and cerebellar atDCS showed enhanced retention of SRTT performance (Ehsani et al., 2016). In a different type of sequence learning which relies on lateral cerebellar function, atDCS over cerebellum reduced tapping movement errors in follow-up sessions. Thus, it appears cerebellar tDCS may facilitate retention of complex motor skills (Wessel et al., 2016). Simultaneously applying M1-atDCS during SVIPT learning facilitated skill acquisition over several consecutive days of training (Reis et al., 2009;Schambra et al., 2011;Saucedo Marquez et al., 2013). Specifically, stimulation promoted between-session (Reis et al., 2009) or long-term retention processes (Saucedo Marquez et al., 2013). Interestingly, when atDCS was applied over the cerebellum, skill acquisition was enhanced within-session (online) rather than between-session gains. Here, skill improvement was marked by lower errorrates rather than movement time (Cantarero et al., 2015). In a slightly different task (visuo-motor pinch force task, see Table 2 for details), tDCS over secondary motor areas such as the supplementary motor area (SMA) showed to increase participants' spatial accuracy, providing new insights about the role of SMA during skill performance (Vollmann et al., 2013).
Beyond the SRTT, SFTT, and SVIPT tasks, there are additional investigations with varying tasks that have explored tDCS effects during skill learning. For instance, atDCS applied either over M1 or an extrastriate visual area during a visuo-motor coordination task improved early performance of correctly tracked movements (Antal et al., 2004), whereas performance was enhanced for both tDCS polarities when stimulation was applied prior to training . Moreover, both uni-lateral and bi-lateral M1-tDCS applied concurrently with skill tracing tasks showed enhanced target-tracking accuracy (Shah et al., 2013;Prichard et al., 2014;Naros et al., 2016), an effect similarly found when pairing training with anodal and cathodal cerebellar tDCS (Shah et al., 2013). Furthermore, combining mirror visual feedback with M1-atDCS improved performance of a manual ball-rotation task with the untrained hand, likely due to additive effects on motor performance (von Rein et al., 2015). Accordingly, when the anode electrode was placed over SMA and cathode over right prefrontal cortex (PFC) performance of a dynamic whole body task was impaired (Kaminski et al., 2013). On the other side, PFC-ctDCS improved performance of a golf-putting task during acquisition and retention, highlighting a promising application of tDCS toward everyday motor activities (Zhu et al., 2015).

Modulating Motor Adaptation
Another type of learning studied in laboratory settings is motor adaptation, or a reduction of errors in response to environmental changes via generating an internal model to predict the consequences of actions. Adaptation is generally tested in a variety of error-based tasks (prisms, rotations, force fields), where quickly accounting for perturbations leads to large behavioral changes (Krakauer and Mazzoni, 2011). In relation to brain stimulation, a recent study applied tDCS to distinct brain regions while participants learned a visuomotor rotation (see Table 2 for details). Specifically, they found cerebellar atDCS resulted in faster reduction of errors caused by a consistent visuomotor-rotation (Galea et al., 2011;Block and Celnik, 2013), whereas atDCS over M1 showed a marked increase in retention of the newly learned rotation (Galea et al., 2011). By using tDCS, this study was able to show an important dissociation in acquisition and retention processes related to motor adaptation and further highlighted the distinct roles of the cerebellum and motor cortex. Furthermore, tDCS over these regions did not enhance intermanual transfer of visuomotor rotation learning (Block and Celnik, 2013) suggesting that these structures do not play as critical of a role for this process.
Another study tested tDCS over cerebellum and M1 during force-field adaptation (see Table 2 for details) and consistent with the results reported by Galea et al. (2011), the authors found that cerebellar atDCS enhanced the rate of acquisition (Herzfeld et al., 2014). This study also showed that cerebellar ctDCS delayed the feedback response to the introduced perturbation and decreased the learning rate. Taubert et al. (2016) observed impaired adaptation and re-acquisition of a force-field perturbation with cerebellar atDCS, while no effect was found for ctDCS. It is possible that the experimental design differences of these studies may explain the inconsistent findings.
Regarding the role of M1 in force-field adaptation, M1-tDCS did not alter the rate of adaptation learning during reaching movements (Orban de Xivry et al., 2011;Herzfeld et al., 2014) similar to visuomotor adaptation. While most studies have reported that motor adaptation is not affected by M1-tDCS, one study showed atDCS over M1 biceps brachii representation led to greater overshooting errors in force-field learning once the field was removed, suggesting a possible role of M1 in the adaptation process of reaching movements (Hunter et al., 2009). While these results remain inconclusive, M1-tDCS showed a clear increase of generalization in intrinsic coordinates for joints and muscles during force-field adaptation, without changing extrinsic generalization patterns. In contrast, tDCS tested over posterior parietal cortex had no effects on learning or generalization (Orban de Xivry et al., 2011).
A few studies have also used tDCS to examine functions of the cerebellum outside of visuomotor and force-field adaptation. One study showed that cerebellar excitability plays a crucial role in saccadic adaptation (Panouillères et al., 2015), as well as in all stages of prism adaptation, i.e., in flexible motor adjustments in response to changes of the visual field (Panico et al., 2016). Moreover, Jayaram et al. (2012) were able to modulate locomotor adaptation by applying tDCS over the cerebellum while participants walked on a split-belt treadmill at two different speeds. They found atDCS ipsilateral to the fast leg accelerated adaptation (i.e., promoted faster gait step-symmetry), whereas ctDCS slowed adaptation. Interestingly, atDCS effects primarily affected spatial, rather than temporal components of walking (Jayaram et al., 2012).

Modulating Use-Dependent Learning
Use-dependent learning (UDL) describes a phenomenon where short-term motor memories are formed and retained due to repeatedly trained motor actions, thus inducing representational changes in the motor cortex (Classen et al., 1998). Rosenkranz et al. (2000) first addressed the effects of tDCS over M1 on UDL by comparing the directional variation of TMS-induced thumb movements (opposite to the trained direction) before and after tDCS application. They found that applying tDCS during the last 5 min of 30-min thumb-movement training resulted in smaller TMS-induced angular deviation compared to controls. In other words, anodal or cathodal tDCS with training produced a movement direction similar to the pretraining direction, whereas movements of the control group were biased to the trained direction. The authors concluded that tDCS preserves pre-training cortical movements by interfering with the mechanisms of UDL and the formation of motor memories (Rosenkranz et al., 2000). In contrast,  demonstrated enhanced retention effects of repetitive thumb training when atDCS over M1 was applied throughout the 30 min training period. Importantly, cathodal and sham group responses did not show significant changes. The inconsistencies between these two studies could potentially be explained by the different stimulation periods of tDCS (5 vs. 30 min). On the other hand, the prior state of the system (i.e., 25 min of training vs. no training) may not be the same when tDCS is applied at training onset vs. at the end of training . A recent study aimed to determine whether M1-tDCS applied before, during, or after motor training enhances UDL. The authors found larger MEP amplitudes (first dorsal interosseous muscle) only when atDCS was applied before motor training. This suggests tDCS prior to training benefits optimization of UDL (Cabral et al., 2015). However, these results are inconsistent with other studies.  showed a significant effect on training by applying tDCS during the training, an effect that is similarly found with sequence-learning . Furthermore, recent results showed enhanced retention of ballistic thumb movements when M1-atDCS was applied during training when evaluating both peak velocities and accelerations of thumb movements (Koyama et al., 2015;Rroji et al., 2015).

CONSIDERATIONS ON MOTOR LEARNING MODULATION AND NEW PERSPECTIVES
Overall, the results summarized in this review highlight the need for new stimulation paradigms based on more natural and individualized stimulation protocols, aiming to optimize the desired stimulation effects. Variability and contradictions between studies need to be considered, however, this is frequently caused by methodological differences (Paulus, 2011;Horvath et al., 2014Horvath et al., , 2015. When considering that different brain regions are likely involved in distinct motor learning processes (Shmuelof and Krakauer, 2011;Penhune and Steele, 2012), the simultaneous (or sequential) electrical stimulation of these areas on the proper polarity and intensity could potentially optimize tCS effects. In this regard, bilateral M1 combined with PFC stimulation has been successfully applied (Vines et al., 2008;Mordillo-Mateos et al., 2012;Leite et al., 2013;Naros et al., 2016). However, the characterization of the effects associated to concomitant stimulation of different brain regions is nearly absent in the literature (Kaminski et al., 2013;Minichino et al., 2015) due to the low focality inherent to the technique and the inability from traditional tDCS devices to simultaneously control multiple stimulation electrodes. Indeed, there has been some progression in recent years. Thus, multifocal tDCS devices using several small-size electrodes , High-Definition tDCS (HD-tDCS) scalp montage (4 × cathode, surrounding a single central anode, Edwards et al., 2013), or concentric electrodes (Bortoletto et al., 2016) provide evidence for more focal tDCS.
On the other hand, new devices allowing for EEG recording during simultaneous tDCS also present an excellent tool for the development of individualized stimulation protocols based on the observed individual brain activity (Schestatsky et al., 2013).
Although more investigations are needed to provide a better understanding of the effects induced by tDCS, its impact on motor learning and use for exploring neural substrates underlying motor learning have been successfully demonstrated. In other words, the potential of this technique for basic studies and future clinical treatments seems promising. However, ethical considerations using tDCS for high-performance sports are still a matter of discussion (Reardon, 2016).

AUTHOR CONTRIBUTIONS
CA and JM contributed to the initial draft, CA, DS, and JM edited the text and wrote the final version of the mini-review.