## Introduction

Throughout our lifespan, numerous motor skills have to be acquired and retained in memory. Behavioural plasticity, as observed in motor skill learning, is contingent on underlying adaptations in the central nervous system, i.e. mechanisms of neuroplasticity. In humans, neuroplasticity is commonly investigated indirectly e.g. using brain stimulation and neuroimaging techniques (see1 for review of early, seminal papers). The corticospinal tract plays a key role in motor control2, and plastic changes in the corticospinal pathway are important mechanisms for the acquisition of skilled movement3. In humans, the early phase of motor skill acquisition is accompanied by a transient increase in corticospinal excitability (CSE). This is evident as an increase in the size of responses to transcranial magnetic stimulation over the representation of the trained muscle following a single session of motor skill training4,5,6. This initial increase in CSE can increase further with multiple days and weeks of motor training, but it eventually stagnates and decreases as training progresses further without additional challenge7. In contrast to this observation, studies of expert musicians and athletes have found larger cortical representation and increased CSE of their trained muscles and movements suggesting that continued training is accompanied by persistent corticospinal plastic changes8,9,10. Such long-term plastic changes in the human corticospinal pathway accompanying motor training are not well understood, and the role of behavioural determinants in particular has only been studied sparsely7,11,12,13.

In contrast, plastic changes in the primary motor cortex (M1) and the corticospinal pathway following weeks of practice have been extensively reported in murine models of motor skill learning14,15,16,17,18,19,20,21 and M1 has been shown to be a pivotal locus for neuronal changes underlying both associative22,23,24, sequence25 and motor skill25,26,27 learning in primates. Furthermore, evidence from rodents28,29 and primates26 indicates that learning per se rather than use drives long-term plastic changes. In support of this, previous work in humans has shown that training-induced increases in corticospinal excitability reach a plateau and may even decline after weeks of training7. However, the decline in CSE is not associated with a corresponding decrease in performance. This suggests that increased corticospinal excitability is more likely related to the acquisition of new motor skills and does not relate directly to performance in the trained task. In contrast to the longitudinal studies, cross-sectional studies have reported that expert musicians, proficient Braille readers and athletes have larger sensorimotor cortical representation of their trained hand compared to non-experts or their own non-trained hand8,9,30,31,32,33.

One possible explanation for these seemingly contrasting findings may be that training for elite athletes and expert musicians is not characterized by practicing discrete laboratory tasks but rather by deliberate practice in which their skills are developed and expanded through progressive challenge34. The continuous progression in task demands and the tailoring of demands to individual levels of motor proficiency likely augments plastic changes in the corticospinal system, resulting in superior motor performance compared to conventional non-progressive motor practice.

This could explain the differences in cortical representation and CSE that have been observed between novices and experts9,31. In support of this hypothesis, we recently demonstrated that 4 days of training a visuomotor accuracy task with progressively adjusted task difficulty resulted in superior learning accompanied by transient increases in CSE after both the first and last practice session4. In contrast, maintaining task difficulty at the baseline level was only accompanied by an increase in CSE on the first day of practice. This confirms earlier results suggesting that distinct neurophysiological processes are involved in early versus later stages of learning35. Altogether, these results suggest that transient within-session changes in CSE are related to top-down processes involved in skill learning whilst automaticity is low.

Here we aim to extend the our previous findings4 by testing the effects of long-term progressive motor training on corticospinal plasticity and skill acquisition. We hypothesize that individually tailored progressive training (PT) that continuously challenges and engages the learner will potentiate the changes in CSE accompanying nonprogressive training (NTP) and resemble those previously found in cross-sectional studies of virtuosi and elite athletes. Furthermore, we aim to provide proof-of-principle that progressive practice promotes learning. Finally, we explore the effects of PT on long-term retention of skill and its potential relation to corticospinal plasticity.

## Results

Group means for motor performance and electrophysiological measures are presented in Tables 1, 2 and 3. Participants were matched based on their ‘Baseline’ motor performance and randomly allocated to the two training groups, i.e. progressive training (PT) or nonprogressive training (NPT) to ensure similar ‘Baseline’ motor performance between groups. It should be noted that the task difficulty was identical for all participants during the first training session, i.e. on Day 1.

### Changes in motor performance during the first day of training

Motor performance during the initial training session was assessed during the first, middle and last 4 min of training. Likelihood ratio tests of the mixed effects model showed that there was neither an interaction between GROUP and TIME (Chi2(2) = 0.44, p = 0.8) nor a main effect of GROUP (Chi2(1) = 0.01, p = 0.91). There was a robust effect of TIME (Chi2(2) = 18.1, p < 0.001), and post-hoc Tukeys test showed that motor performance was significantly higher in the middle (9.9 ± 1.5) and last (11.6 ± 1.2) compared to the first 4 min of training (8.3 ± 1) (z = 2.62, p = 0.026 and z = 4.68, p < 0.001). There was no significant difference between the middle and last training block (z = 2.06, p = 0.12) . Single subject and group data from the first training session is illustrated in Fig. 1A.

### Long-term changes in motor performance at the ‘Day 1’ task level

The longitudinal effect of training was assessed as motor performance after 6 weeks of practice and reassessed after 8 days of detraining (retention test 1) and again 14 months later in a subsample of participants (retention test 2). Motor performance at these time points was assessed at the same level of task difficulty as ‘Day 1’ and compared to ‘Day 1’ motor performance taken as the average performance in the three blocks of training illustrated in Fig. 1A. Motor performance for the two groups is shown in Fig. 1B and data are presented in Table 1. The likelihood ratio tests on the mixed effect model for long-term motor changes in performance showed neither a main effect of GROUP (Chi2(1) = 0.007, p = 0.93) nor a GROUP × TIME interaction (Chi2(3) = 0.12, p = 0.99). There was however a significant main effect of TIME (Chi2(3) = 69.26, p < 0.001). Post-hoc tests revealed a performance increase from ‘Day 1’ to ‘6 weeks’ (10.04 points increase, z = 9.05, p < 0.001), and a similar increase was present at the ‘Retention 1’ test 8 days after the final training session (10.7 points increase, z = 9.9, p < 0.001). There was no difference in motor performance from the 6-week test to the ‘Retention 1’ test 8 days later (z = 0.84, p = 1). Motor performance 14 months after the intervention was still higher compared to ‘Baseline’ (6.83 points increase, z = 5.4, p < 0.001), but there was a decrease in performance when compared to performance at Retention test 1 8 days after the last training session (z = − 2.965, p = 0.018). Please note that model estimates for ‘Retention 2’ are based on fewer data points. Raw values for the 14 participants tested on ‘Day 1’-task level at all time-points incl. Retention 2 are shown in Table 2.

### Long-term changes in motor performance at the ‘End’-task level

After 6 weeks of training and at the delayed retention tests, motor performance was also assessed at an advanced task level for half of the participants. The task difficulty corresponded to the level that the first batch of participants in the progressive group reached during the 6 weeks of training and is hence referred to as ‘End task level’. Figure 1C shows motor performance at the ‘End task level’ for the 12 participants tested on this level, along with group means. The statistical model showed a significant GROUP × TIME interaction (Chi2(2) = 11.287, p = 0.0035). Post-hoc testing of relevant comparisons showed that the PT group performed better at the ‘End task level’ after 6 weeks of training (z = 4.51, p < 0.001) and 8 days later at Retention test 1 (z = 5.27, p < 0.001). Interestingly, motor performance at the End task level dropped in the PT group between retention test 1 and 2 14 months later (5.76 points decrease, z = 5.872, p < 0.001) where it was not different to that of the NPT group (z = 0.759, p = 1). Data is presented in Table 1 and examples of EMG activity during the first and last day of training are illustrated in Fig. 1D. These results demonstrate that increasing task difficulty for the PT group during the 6 weeks of training allowed them to perform significantly better at a high level of task difficulty compared to the NPT group. Furthermore, this improved motor performance persisted through 8 days but not 14 months of detraining. Please note that the model-estimate for Retention test 2 is estimated from five participants in each group (n = 10). Accordingly, raw values for the ten participants tested at all time points are presented in Table 2.

### Electrophysiological measures of corticospinal plasticity

The individual TMS recruitment curves were obtained before and after the first day of training as well as after 2, 4 and 6 weeks of training and at the two retention tests. The curve fitting procedure resulted in a mean r2 of 0.72. In addition to TMS recruitment curves, measurements were also obtained for resting motor threshold (rMT) and maximal compound muscle potential elicited by peripheral nerve stimulation (Mmax) in each session. All motor evoked potential (MEP) amplitudes were normalized to the corresponding Mmax. There were no differences in Mmax between groups or across time. All electrophysiological data are presented in Table 3. Additionally, recruitment curve parameter estimates were normalized to individual ‘Baseline’ values in order to investigate the relative changes in each parameter. Data normalized to individual ‘Baseline’ values are presented in Fig. 2. For illustrative purposes, all normalized recruitment curve data for all participants in the PT and NPT groups are plotted in Fig. 2A,B. The depicted curves represent global fits for all participants in the two groups in each test. No statistical comparisons were performed on the global fits.

### Changes in electrophysiological measures following the first training session

#### Area under recruitment curve

The mixed effects model for the area under the recruitment curve AURC revealed no significant interaction between GROUP and TIME (Chi2(1) = 2.14, p = 0.14) and no main effect of GROUP (Chi2(1) = 2.05, p = 0.15). There was a significant effect of TIME (Chi2(1) = 11.28, p < 0.001), meaning that AURC increased following the first day, i.e. after 28 min of training.

#### Maximal MEP amplitude

The model comparison for maximal MEP amplitudes MEPmax revealed no significant interaction between GROUP and TIME (Chi2(1) = 1.73, p = 0.19) and no main effect of GROUP (Chi2(1) = 1.67, p = 0.2). There was however a significant effect of TIME (Chi2(1) = 10.28, p = 0.001), indicating that MEPmax increased following the first day of training.

### Long-term changes in electrophysiological measures

We observed no significant effects of GROUP or TIME, and no interaction for the slope or the I50 parameter of the recruitment curves (p > 0.15). Consequently, the following results section focuses on the AURC, MEPmax, and rMT parameters.

#### Area under recruitment curve

The mixed effects model of AURC with GROUP × TIME as fixed effects and PARTICIPANT as a random intercept effect showed a significant GROUP × TIME interaction when tested against the model without the interaction term (Chi2(4) = 10.11, p = 0.039). Post hoc tests of differences in AURC compared to ‘Baseline’ (eight comparisons) showed an increase in AURC after 2 weeks (124.9 ± 34.3%, z = 3.84, p = 0.001), and AURC remained elevated at 4 weeks (124.8 ± 42.2%, z = 3.47, p = 0.004) and at the Retention test 1 (115.2 ± 29.5, z = 3.7, p = 0.002) for the PT group. The 71.5 ± 26.4% increase observed in AURC after 6 weeks of training in the PT group did not reach statistical significance when correcting for multiple comparisons (z = 2.3, p = 0.17). There were no significant long-term changes in AURC for the NPT group (all p = 1). Post hoc tests for differences between groups for the \$ time points (four comparisons) showed a higher AURC for the PT group following 2 (z = 2.84, p = 0.018) and 4 weeks of training (z = 3.49, p = 0.002). The difference between groups did not reach statistical significance after 6 weeks of training and at Retention test 1 (p = 0.43 and p = 0.086, respectively). The development in AURC for the two groups can be seen in Fig. 2C.

#### Maximal MEP amplitude

For MEPmax the analysis revealed a significant GROUP × TIME interaction (Chi2(4) = 9.65, p = 0.047). Post hoc Tukeys tests were conducted to compare the groups (four comparisons) and the effect of time within each group (eight comparisons). The PT group showed a significant increase in MEPmax following the first 2 weeks of training (195.4 ± 35.9% of ‘Baseline’; z = 2.95, p = 0.026). After 4 weeks of training, MEPmax was increased to 219.40 ± 48.7% of ‘Baseline’, which was also significantly higher than ‘Baseline’ (z = 3, p = 0.016). After 6 weeks, MEPmax was no longer significantly higher than ‘Baseline’ (160.76 ± 32.0% increase, z = 1.58, p = 0.91). At the retention test 8 days later, MEPmax increased again to 193.5 ± 30.2% of ‘Baseline’, which again was significantly higher than the at ‘Baseline’ test (z = 3.13, p = 0.014). The NPT group showed no significant changes in MEPmax at any time point (p = 1). Although the PT group generally expressed larger relative increases in MEPmax values after 2, 4 and 6 weeks of motor training as well as Retention test 1, a statistically significant difference was only observed following 4 weeks of practice (z = 3.57, p = 0.001, other p > 0.14) when considering between groups post hoc comparisons. Data for both groups normalized to individual ‘Baseline’ values are presented in Fig. 2D.

#### Resting motor threshold

Analysis of the relative changes in resting motor threshold (rMT) based on normalized data also revealed a main effect of TIME (Chi2(4) = 22.35, p < 0.001), but no main effect of GROUP (Chi2(1) = 0.6, p = 0.38) or interaction between GROUP and TIME (Chi2(4) = 3.13, p = 0.54). Post hoc comparisons for the main effect of TIME (four comparisons) showed a relative reduction in rMT by ~ 6% or more at all time points compared to ‘Baseline’, (z > 3.3, p < 0.004) Group mean values for rMT are presented in Table 3 and data normalized to ‘Baseline’ are illustrated in Fig. 2D.

#### Correlation analysis

Pearson product moment correlation analysis revealed a significant association between changes in MEPmax and end-level task performance at the 6-week test (r2 = 0.48, p = 0.018). In addition, a strong tendency towards a negative association between changes in rMT and relative changes in performance from Day1 to 6 weeks on the ‘Day 1’ task level was observed (r2 = 0.14 and p = 0.057). No correlation was present between either MEPmax at 6 weeks and motor performance at ‘Day 1-level’ after 6 weeks of training (p = 0.73).

## Discussion

Our study demonstrates for the first time that an individually adjusted, progressive training protocol involving 6 weeks of visuomotor skill acquisition results in significantly better motor performance compared to a non-progressive skill learning protocol. Progressive training was also accompanied by pronounced increases in corticospinal excitability suggesting that continuous challenge during motor training drives corticospinal plasticity. Thus, this study confirms previous findings in non-human primates indicating that learning, rather than execution, drives long-term corticospinal plasticity27. These findings also provide a plausible mechanistic background for increased cortical representational and/or CSE observed in expert musicians and athletes. This underlines the importance of continuously challenging motor performance for driving plastic changes and ensuring maximal performance gains. This is relevant not only for musicians and athletes but also as a plasticity inducing tool to improve the outcomes of rehabilitation in stroke survivors.

### Progressive practice leads to superior performance

In the present study, we adapted a task previously used over a similar timespan11. This task enables individually adjusted requirements for motor acuity (i.e. the ability to move faster and more accurately36), while being entertaining enough to engage the participant when training on a basic level of task difficulty. The task entails a partly unpredictable environment with no fixed sequences. This differentiates it from the majority of tasks previously used in long-term studies as discussed below.

Whereas performance measures after 8 days of detraining mirrored those observed after 6 weeks of practice, a long-term retention test conducted after 14 months demonstrated that between-group differences in motor performance were no longer present. Here, both groups showed retention of skill when tested on the ‘Day 1’ task , but the benefits of progressive practice on ‘End level’ performance were no longer present. This emphasizes the necessity of continued training at a high task level in order to maintain skill mastery.

### The motor learning model

High fidelity, individual adjustments of task difficulty are a pivotal part of the deliberate practice necessary when training for skill mastery in elite sports34. The progressive training model provides a readily implementable method for studying the development of expertise in laboratory settings in search of determinants contributing to skill development towards expertise. Furthermore, motor learning is frequently studied by employing models of sequential learning and motor adaptation (see e.g.37) or categorized as either model-free or model-based learning based on the dependence on error versus reward-contingent processes38. However, most real life skills contain both model-based and model-free elements and require learning of motor sequences as well as the capacity to adapt to new sensorimotor environments39. In the present study, the participants practiced a visuomotor task (breakout), in which performance relies on accurate visuomotor transformation and fast, accurate movements. Hence, the task has a prominent motor component and improved performance is contingent on improved acuity, which differentiates it from the sequential tasks deployed in most long-term motor learning experiments (e.g.40,41). This is likely to be reflected in the neural correlates as discussed below.

### Initial within-session increases in corticospinal excitability

As expected, we found a left and upwards shift of the recruitment curve after the first motor practice session i.e. 30 min of visuomotor skill learning. This demonstrates that motor practice at the ‘Day 1’ task level was accompanied by an acute increase in CSE. Facilitation of TMS MEPs accompanying motor learning is supported by ample existing evidence (e.g.4,5,6,42,43). It further supports previous findings of acutely increased CSE with practice of both visuomotor6, ballistic43, and sequential44 motor tasks. The increased response to magnetic stimulation likely reflects processes associated with motor learning when skill automaticity is low, since the fourth4 and fifth45 day of motor practice has been found not to be associated with acute increases in CSE. It is likely that within-session changes mark top-down processes such as attention and motivation leading to successful learning, but it is yet to be investigated how such transient changes in CSE relate to the long-lasting plastic changes observed in the present study.

### Long-term progressive practice leads to increased global measures of corticospinal plasticity

We found that progressive visuomotor practice was accompanied by substantial and long-lasting increases in CSE assessed as AURC. This increase was evident as a left and upwards shift of the recruitment curve i.e. changes in the MEPmax and rMT parameters (see Fig. 2). In contrast, long-term NP training was only accompanied by decreased rMT.

We argue that the increased plateau level of the recruitment curve reflects cortical processes that also enable fast and precise digit control. Direct measurements from primate M1 have revealed that activity in cortical neurons is related to speed and direction of limb movement46 and in humans, contractions with high demands for precision are accompanied by increased corticospinal activity as reflected in increases in MEP size47. Furthermore ballistic learning increases MEP size35,43,48,49,50, and increasing requirements of both speed and accuracy bring about repeated gains in CSE4. Recently, Raffin and Siebner reported increased cortical representational map after 8 days of training on a visuomotor task comparable to the one used in the present study51. Interestingly, they found that changes in measures of excitability explained approximately 20% of the variance in performance gains, which corresponds well to the results of the present study (R2 = 0.48 and 0.14 for the ‘End’ and ‘Day 1’ task level respectively). Based on the existing literature4,35,51, we argue that it takes a minimum of 5 days of visuomotor training to induce learning-related representational changes in M1 that can be captured by single-pulse TMS measures. The 6 weeks of training in the present study should consequently be adequate to induce such representational changes.

In summary, existing evidence from studies in non-human primates and humans lends credence to the hypothesis that changes in the maximal corticospinal output subserve the acquisition of skilled visuomotor performance. This relation is further supported by the current findings of related changes in the MEPmax parameter and skilled performance.

### Long-term motor training leads to decreases in rMT

Motor threshold decreased for both groups during the training period, but no significant differences were observed between groups. The rMT is suggested to reflect the excitability of the most excitable elements in the stimulated pathway52. Recent work combining electric field modelling with modelling of cell morphology suggests that layer 2/3 and 5 pyramidal neurons located superficially at the crown of the precentral gyrus are most sensitive to TMS53. Changes in inputs to these or in the downstream synaptic connections could therefore account for or contribute to the reduced rMT.

It should be noted that since both groups displayed decreased rMT and no passive control group was included, it cannot be entirely excluded that motor use, passing of time, or repeated visual stimuli contributed to this decrease. However, this is unlikely as changes in rMT do not accompany unskilled motor repetition7.

The observed decreases in rMT with long-term motor practice in the majority of participants therefore support the ample evidence from animal research demonstrating a pivotal role of M1 in long-term motor training20,21,25,28,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77.

### Changes upstream and downstream of M1 could contribute to the increased CSE

An increase in resting MEP amplitude may reflect both top-down and bottom-up processes and could in theory reflect changes in inputs from association cortices upstream from M1 engaged in skill learning but independent of the mechanisms leading to sensorimotor refinement (see78 for discussion). In support of this, the majority of previous investigations have failed to demonstrate a relationship between changes MEP amplitude and increased performance (e.g.4,5,6,43,45,48,49,79,80). However, in the present study, the long-term changes in MEPmax were related to end-level motor performance on an individual level after 6 weeks of practice. Accordingly, we suggest that the observed changes in CSE reflect task-specific adaptations in neural ensembles and single cell activity of the trained M1 as demonstrated in rodents after skilled motor practice. However, it cannot be entirely excluded that the present results reflect changes in other cortical areas than M1.

The experimental paradigm of the present study also does not permit us to exclude the possibility of contributions to the observed increase in CSE from neural structures downstream of M1. Synaptic connections between descending pyramidal cells and their spinal targets are malleable81,82 and are a likely substrate for activity-dependent modulation83. It is likely that plastic changes on a spinal level contribute to reshaping motor output during skill learning (see84 for review). Previous findings that short-term ballistic learning potentiates the response to electrical stimulation at the level of the brainstem support the notion that spinal adaptations may contribute to the observed changes in MEP amplitude on the first day of training85. However, training on a visuomotor tracking task similar to the dynamic visuomotor task used in this study does not increase the size of brainstem evoked motor potentials85. This suggests that the observed changes in CSE following the first day of practice are likely cortical in origin.

Due to methodological challenges, measures of subcortical plasticity such as changes in the response to cervicomedullary stimulation were not included in the present study. The position and impedance of the stimulation electrodes on the mastoids are difficult to reproduce throughout 6 weeks of measurements. Furthermore, whereas substantial discomfort would be associated with mapping the full input–output relationship in resting hand muscle, current would spread to C7/C8 cervical motoneurons at intensities below those needed to trigger a maximal compound potential86. Also, we did not asses changes in excitability of the motoneurons or spinal reflexes using e.g. F-waves and H-reflexes. H-reflexes are not readily elicited in ADM and comparisons between days are complicated by the inability to elicit stable reflexes on the ascending part of the H-reflex recruitment curve while an M-wave is present for stimulus control87. In contrast, F-waves are readily obtained from intrinsic hand muscles and have previously been obtained to estimate changes in intrinsic excitability of the motoneuronal pool (see e.g.88). However, the between-day reproducibility of the read-outs i.e. chrono-dispersion, persistence and amplitude89 as well as the interpretation of changes in these are associated with some controversy90. Future investigations into neural underpinnings of the behavioural effects of long-term progressive training should include measures of both intracortical and subcortical excitability to further elucidate central nervous correlates of skill learning.

### Perspectives for neurorehabilitation

Our observations have direct clinical applicability. The overarching aim of sensorimotor rehabilitation following central nervous lesions is to improve motor functions and recover independent living. This, in turn, depends on the underlying neuroplastic changes. The few daily repetitions reported for neurological patients during upper limb training (32 for stroke91, 7 for paraplegic and 42 after tetraplegic spinal cord injured92) are not likely to drive functional changes through use-dependent mechanisms. To achieve extensive central nervous reorganization comparable to those reported in laboratory animals following a > tenfold number of repetitions (e.g.27,28), additional measures are likely needed. Despite substantial attention to deliberate practice both within and outside of the scientific community, an emphasis on deliberate training for expertise has not gained much influence in neurorehabilitation. The finding that a continuously challenging training regime boosts corticospinal plasticity and elicits additional behavioural effects with a similar volume of training has extensive implications for rehabilitation. Based on the present behavioural results and underlying changes in corticospinal transmission, we suggest carefully monitored progressive training protocols as routine choice in neurological rehabilitation.

## Conclusions

We demonstrate that continuously challenging the individual participant by progressively increasing task difficulty during long-term motor practice enhances motor learning and optimizes performance in a complex discrete motor task requiring highly accurate and rapid movements and with high demands for visual processing and movement prediction. Progressive long-term training not only enhanced learning, but also prolonged increases in corticospinal plasticity without increasing the volume of training. In conclusion, the results suggest that motor learning paradigms should be structured with the aim of continuously challenging the individual. This is critical in order to promote the neuroplasticity that underlies learning and therefore to improve the behavioural outcome of training for neurorehabilitation following acquired brain injury and for athletes who engage in training.

## Materials and methods

We obtained measures of motor performance and applied transcranial magnetic stimulation (TMS) and peripheral nerve stimulation to assess the effects of two different training protocols on motor skill learning and changes in corticospinal excitability. Two groups of participants engaged in 6 weeks of visuomotor training with either maintained (NPT) or progressively increased task difficulty (PT). Training consisted of a visuomotor task developed for this study. The effect of training on corticospinal excitability was assessed using TMS by comparing recruitment curve parameters before training and after 2, 4 and 6 weeks of motor practice. Delayed retention of motor performance and CSE were assessed at Retention test 1 and 2, 8 days and 14 months after the 6-week training intervention respectively, to evaluate the long-term effects of motor practice.

### Participants

Twenty-four adult men aged 21–29 years at the time of enrolment (24 ± 4, mean ± s.d.) were randomly allocated into two different training groups. All participants had a moderate to high level of daily physical activity and had no known medical condition that could interfere with motor skill learning of hand movements. Participants were matched in pairs based on initial visuomotor performance i.e. during the first training session. Each member of the pair was randomly assigned to one of the two groups to ensure comparable ‘Day 1’ performance in the two groups. For details of the performance test, see below. Participants were instructed not to engage in physical training of any kind prior to testing sessions and to eat, sleep and drink similarly on all days of testing. For each subject, all tests were conducted at the same time of the day to minimize intra-individual day-to-day differences in motor cortical excitability52. Twenty-three participants were right handed according to the Edinburgh Handedness Inventory93 and one had no hand preference.

Written informed consent was obtained from all participants prior to their participation in the study. The experiments were approved by the local ethics committee of the capital region of Denmark (KF01-131/03) and all experimental procedures were carried out in accordance with the Helsinki Declaration (1964).

### Design

The two groups of participants engaged in 18 training sessions with their right (dominant) hand over a period of 6 weeks. Training sessions were held three times per week and, when possible, separated by 48 h. Each training session consisted of seven 4-min bouts of activity interspaced with 2-min rest periods. Electrophysiological testing was repeated at 2-week intervals, and behavioural performance was evaluated during the first training session and assessed after 6 weeks of motor training. Eight days after the end of the 6-week training period, all participants were subjected to a retention test i.e. a delayed test of motor skill retention and CSE. Fourteen months after the training period, all available participants (n = 14) were subjected to a second delayed test of motor skill retention.

At least 3 days prior to the ‘Day 1’ session all participants were familiarized to transcranial magnetic stimulation. During the ‘Day 1’ session, TMS recruitment curves were obtained both before training and again immediately after the post training Mmax, which is described below. The electrophysiological measurements obtained pre-training on Day 1 were used for long-term comparisons and are referred to as ‘Baseline’. After 2, 4, 6 weeks of training and in the delayed retention tests after 8 days and the detraining period, only pre-training electrophysiological measurements were obtained. At the 6-week test and in the long-term retention tests 8 days and 14 months after the intervention, electrophysiological measurements were obtained initially and motor performance was tested subsequently. The testing and training were conducted in 2 “batches” due to limited laboratory capacity i.e. 12 participants (six in each group) underwent testing and training procedures in parallel. Testing of a 2nd ‘batch’ also consisting of 12 participants (six in each group) started following the 6 weeks testing of the first “batch”. An overview of the long-term study design is presented in Fig. 3.

It should be noted that the results from the first six participants in the PT group were used to estimate the attained ‘end-level’ motor performance. It was therefore only possible to test the remaining half of the participants from each group at this task level. This was done after 6 weeks of training and at the retention tests.

The first training session took place at the ‘Day 1’ session and was identical for the two groups. After the first training, one group continued training at the ‘Day 1’ level (non-progressive training group, NPT), while the other group trained with task-difficulty that was progressively adjusted to correspond to their skill level in the motor task (progressive training group, PT). No adjustments to the level of difficulty were made for NPT group. For details on the progression, see below. Only the 2nd batch of participants was tested at the ‘End Level’ of task difficulty (i.e. ‘End level’) at the 6 W and retention tests (see Fig. 3).

In each training session the participant started out by training at the level of task difficulty that was reached during the previous training. During performance testing, the total number of residual bricks and lost balls were recorded for 3 (‘Day 1’) or 2 (at 6 weeks and retention tests) of the training blocks for later analysis. Participants received standardized information about the game and the performance score and were asked to do their best at all times. The first 4 min of training on Day 1 served as a familiarization procedure and performance measures from these 4 min were not included in the data analysis.

During training (and testing) the participant was seated in a comfortable chair with both hands on a panel placed on top of a table. The right hand and forearm were secured with VelcroTM straps to maintain the hand position during training. The forearm was kept flat on the panel by two straps; one distal to the elbow joint and the other approximately 2 cm proximal to the wrist. The hand was held in a pronated position by two straps, one distal to the wrist and the other crossing the back of the hand. The thumb and first three fingers were also fixed to the panel by two straps. The trackball was built into the supporting panel and positioned below the fifth digit. The participants manipulated the trackball and thereby the position of the game-paddle by abducting and adducting the finger. The experimental setup is illustrated in Fig. 4. The participant was positioned with the shoulder slightly flexed and abducted and with elbow joint flexed to approximately 90°. During electrophysiological measurements, the left hand was placed in a relaxed position similar to the right arm.

### Recording and stimulation procedures

Electromyographic (EMG) recordings from the ADM muscle were obtained with bipolar surface EMG electrodes (0.5 cm diameter of electrodes; 2 cm distance between electrodes; Blue Sensor, Ambu Inc.,USA) over the belly of the muscles. The EMG signals were amplified (2000×), using NeuroLog EMG amplifiers (Digitimer Ltd., UK), band-pass filtered (5 Hz–1 kHz) sampled at 2 kHz, and stored on a PC for off-line analysis (CED 1401+ with Signal 3.09 software, Cambridge Electronic Design Ltd., UK). EMG activity during training was recorded with Spike 2 (CED, Cambridge UK) and stored for later off-line analysis.

Magnetic stimuli were delivered to the contralateral (left) hemisphere primary motor cortex (M1) by a Magstim Rapid2 stimulator (Magstim Company Ltd., Whitland, UK) via a custom made 90 mm figure-of-eight coil (batwing design, Magstim Company Ltd., Whitland, UK) with the capability to deliver a magnetic field of 2 T. The optimal coil position (hotspot) for eliciting motor evoked potentials (MEPs) in the ADM muscle was established through a standardized stimulation procedure with high spatial resolution covering the primary motor cortex (M1), i.e. a mapping procedure, at each test. During assessment of the resting motor threshold (rMT) and during generation of the recruitment curves the coil was placed with the centre oriented parallel to the scalp over the hot-spot of the ADM representation with the handle of the coil pointing backward at an angle of 45° to the sagittal and horizontal axis. TMS recruitment curves were obtained by delivering 60 single pulse stimuli in a random sequence with an inter-stimulus interval of 3 s and stimulus intensities ranging from 0.8 to 1.8 MT94. The rMT was defined as the minimum intensity required to elicit a peak-to-peak MEP amplitude larger than 2 × s.d. of average background activity (i.e. noise) in three out of five trials (always below 50 μV). All TMS measurements were obtained while the participant was at rest. Trials in which any background activity larger than 2 × s.d. was observed were discarded. A maximum of five trials were discarded from each recruitment curve, which consequently was modelled based on at least 55 MEP amplitudes. During all experiments involving TMS, frameless stereotaxy (Brainsight 2, Rogue Research, Montreal, Canada) was used to identify the coordinates of the M1 hotspot and to monitor the position of the coil relative to the participants’ head.

Before generation of recruitment curves at each test, maximal compound muscle action potentials of ADM (maximal M-waves, Mmax) were elicited by bipolar electrical stimulation of the ulnar nerve (Digitimer 7A constant current stimulator, Digitimer Ltd., UK). The intensity of stimulation was increased from a subliminal level until there was no further increase in the peak-to-peak amplitude of the M-wave with increasing intensity. The purpose of this procedure was to normalize the MEP data obtained on each test day to the corresponding Mmax. This allowed comparison across different test sessions. Assessors were not blinded to the group allocation.

### Data analysis and statistics

Visuomotor performance was computed as the average number of bricks shot down by each ball corrected by a factor 1.n with n being the number of screens/rounds cleared without losing the ball within each 4 min block, i.e. a bonus for completing trials successfully. Each 4-min block of motor training or test of motor performance thereby resulted in a cumulated score. This correction accounts for the fact that the elimination of the last few blocks is associated with greater difficulty and time use. During the first day of practice (‘Day 1’) the performance scores from the first 4, middle 4 and last 4 min were used to depict learning during the first session (see Fig. 1A). The average of these 12 (3 × 4) min of training was used as the ‘Day 1’ score in long-term performance comparisons. For the remaining tests, performance measures were obtained from two blocks of 4 min of testing on the ‘Day 1’ task level. The reduced sampling time was chosen as it adequately represented the performance level of the participants and with the intention of minimizing the influence on later electrophysiological recordings (i.e. on retention tests). For the 12 participants tested in the 2nd test-round (see Fig. 3) the reduced test time also reduced potential anterograde interference of the testing on the ‘End’-task levels.

MEP amplitudes were normalized to Mmax (recorded just prior to TMS testing on that day) to allow comparison between test days, and stimulation intensity was normalized to ‘Baseline’ rMT. The MEP amplitudes obtained after the first training session were normalized to Mmax likewise obtained after the training. The recruitment curves were constructed by modelling the relationship between stimulus intensity and MEP peak-to-peak magnitude with a Boltzmann-like sigmoid equation previously described95,96. The equation relating the magnitude of the MEP to the stimulus intensity (I) is:

$${\text{MEP}}\left( {\text{I}} \right) = {\text{MEP}}_{\min } \frac{{{\text{MEP}}_{\max } - {\text{MEP}}_{\min } }}{{1 + {\text{e}}^{{\frac{{{\text{I}}_{50} - {\text{I}}}}{{\text{S}}}}} }}$$

where MEPmin is the amplitude of background noise, MEPmax is the maximum plateau value, I50 is the stimulus intensity at the inflection point where a MEP amplitude of 50% of MEP max is obtained and S is the slope at the inflection point. The inverse of the slope parameter (1/S) is directly proportional to the maximal steepness of the function. Thus, the MEP recruitment curve is described by the motor threshold (rMT), the maximum elicited response (MEPmax), and the transition between them (S, I50) in relation to stimulus intensity (I)94. The parameters were estimated by fitting this equation to the stimulus–response data with a standard Marquardt–Levenberg non-linear least squares algorithm (Matlab curve fitting toolbox). Based on the parameter estimates, the area under recruitment curve (AURC) between 80 and 180% of rMT was calculated for each individual at each test as a global measure of CSE97.

All statistical analyses were carried out in R (version 3.4.1, R Core Team, 2017). A linear mixed effect analyses was computed using lme498 and multiple comparisons computed using multcomp99. The linear mixed models include the dependent variable modelled against GROUP and TIME (with and without interaction term) as fixed effects and PARTICIPANT as a random effect with random intercept. In order to investigate the relative changes in corticospinal excitability and to ensure that each participant would have the same influence on the statistical analysis, the stimulus response curve parameters described above including rMT were furthermore normalized to individual ‘Baseline’ values. All values are reported as mean ± standard error of the mean (SEM) unless stated otherwise. Homoscedasticity and normality were assessed through visual inspection of the residual and quantile–quantile plots and data was log-transformed if any obvious deviations were observed. p-values were obtained by likelihood ratio tests of the full model against the model without the effect in question. In all tests, statistical significance was accepted at the p < 0.05 level and all a priori specified post hoc pairwise comparisons were Bonferroni corrected. Three post hoc comparisons were conducted to investigate the changes in performance for all 24 participants during the first training and six were conducted to assess long-term effects of motor training on ‘Day 1’ task level performance. Seven comparisons were carried out to test for differences at 6 weeks and the two retention tests on the task ‘End-level’. To investigate within-group long-term changes in CSE compared to ‘Day 1’, eight comparisons were conducted and additional four to investigate differences between groups.