Complex motor skill learning over several weeks or years leads to structural changes in the adult human brain. In retrospective cross sectional studies, it has been shown that extensive years of practice in professional musicians was associated with gray matter (GM) volume differences in comparison to non-musicians (Gaser and Schlaug, 2003a,b). Subsequently, it was demonstrated in prospective investigations that just weeks or months of learning a three ball juggling cascade can induce structural brain changes in young normal volunteers (Draganski et al., 2004; Scholz et al., 2009) as well as in elderly subjects (Boyke et al., 2008). Recent findings provided compelling evidence that such use-dependent structural changes can be induced on a much shorter time scale usually within weeks of complex motor skill learning and that structural changes share dynamic properties comparable to those previously reported for functional brain changes (Filippi et al., 2010; Taubert et al., 2010). Additional confirmation about the behavioral relevance of structural changes as a consequence of complex motor skill learning comes from positive correlations between local GM volume changes and performance improvements over time (Taubert et al., 2010).

While these previous studies have qualitatively demonstrated that motor skill learning is accompanied by structural changes in the human brain, several important aspects have yet to be clarified: for example, it is unknown if structural alterations develop during the first week of training exposure and how they are related to individual performance improvements. Furthermore, the directionality of structural brain plasticity and its behavioral consequences has so far not been systematically investigated. While the majority of studies reported significant increases in GM as a consequence of skill learning (for review see Draganski and May, 2008), some studies also identified decreases in specific brain areas that are related to skilled performance (Stein et al., 2012; Taubert et al., 2010; Granert et al., 2011). Although the underlying cellular mechanisms of this learning-related bidirectional GM alterations still remain elusive these findings provide novel evidence that better performance is not exclusively linked with increased GM volume.

Based on the aforementioned studies, our primary hypothesis was that individual performance improvements in a complex motor task can be explained by individual GM alterations in M1 using a correlation analysis. Additionally, since previous studies have consistently shown that in addition to M1 also other brain areas are modulated by complex motor skill learning such as the cerebellum, premotor, and prefrontal areas (for review see Halsband and Lange, 2006) we also investigated structural brain changes using a whole brain exploratory analysis.

Another issue is the pronounced inter subject variability during motor learning. Based on previous findings (Tomassini et al., 2010) we hypothesized that individual (baseline) GM volume in task-related brain areas might predict individual performance improvements across the whole learning period as well as the extent of learning-induced structural brain changes.

We studied 15 healthy subjects between 22 and 32 years of age (five females). All subjects gave written informed consent to participate in the experiment according to the declaration of Helsinki. The study was approved by the ethics committee of the University of Leipzig. Prior to participation, all subjects underwent a comprehensive neurological examination. They were not taking any medication. Subjects who did not meet the protocol criteria and/or had contraindications for the magnetic resonance imaging (MRI) measurements were excluded from participation. All subjects were task naïve. We did not include highly skilled subjects such as musicians, typists or sportsmen. According to the Oldfield questionnaire for the assessment of handedness (Oldfield, 1971), all subjects were right-handed [laterality score: +100 ± 12 (median ± stdev.) over a range of −100 (fully left-handed) and +100 (fully right-handed) where a value of 0 was used as a cut-off score between right and left handers].

In the present study we confirmed previous work showing that skill learning is associated with GM volume changes that are directly linked with individual performance improvements. However, we here extended these findings by providing novel evidence that the individual learning rate is associated not only with increased GM volume but also with bidirectional structural changes. More specifically, we showed that the individual learning success in a SPFT after 5 × 20 min of SPFT training (day 1–5) positively correlated with structural GM volume changes in left M1, right ventral PMC and DLPFC. Interestingly, most of the subjects with large behavioral gains were those that showed learning-related GM volume increases while some subjects with only little performance improvements either showed no change or a decreased GM volume in the respective brain areas. Furthermore, the initial (baseline) GM volume of bilateral cerebellum positively correlated with sequence-specific motor learning and structural changes across days. Hence, initial differences in cerebellar GM volume seems to “predict” whether subjects will show improved performance with training and whether there will be training-related changes in cerebral cortical GM. It seems quite likely that initial differences in cerebellar GM volume reflect important differences in the subjects' motor learning histories prior to their involvement in this experiment. In this view, subjects might have started the experiment at different locations along some hypothetical motor learning continuum, with the initial cerebellar measurements serving as the index of their pre-training potential for behavioral improvement and correlated brain changes. However, it is important to keep in mind that in our study, subjects with a history of enhanced (hand) use (such as musicians and sportsmen) were excluded from participation. Thus, it is likely that other factors such as genetic predisposition, synaptic history before training or other unknown determinants might as well account for individual differences in motor learning across subjects.

To our knowledge, this is the first study demonstrating fast evolving structural brain changes as early as after 5 consecutive days of SPFT learning that correlated with the individual learning rate. Similar fast evolving structural alterations were previously observed after daily application of low-frequency repetitive transcranial magnetic stimulation (May et al., 2007) or complex motor skill adaptation (Landi et al., 2011). In a similar vein, using a neuropharmacological approach, it has been shown recently that a single application of haloperidol, a D₂ receptor blocker, is capable of evoking significant structural alterations in bilateral putamen within hours after intake (Tost et al., 2010). Likewise, a recent study provided evidence that learning to juggle (a complex multimodal task) seems to alter GM volume in the occipito-temporal cortex as early as after 7 days of training even though this was a non-significant trend only. Structural changes in the above mentioned study were not related to the individual performance improvements (Driemeyer et al., 2008). Progressive (dynamic) morphological alterations in gray and white matter were observed during learning a whole body balancing task for several weeks and in that study, individual performance improvements directly correlated with structural gray and white matter changes after six consecutive weeks of learning (Taubert et al., 2010).

We found a positive linear relationship in several motor-related brain regions including the left M1 (BA 4), the right ventral premotor cortex (ventral PMC, BA6) as well as the right dorsolateral prefrontal cortex (DLPFC, BA 9). This finding is in line with recent functional MRI studies showing that the individual positive BOLD signal change was associated with improved task performance (Floyer-Lea and Matthews, 2004; Tegenthoff et al., 2005). For example, in a similar pinch force task, a significant correlation between BOLD signal changes in M1 and DLPFC including middle frontal gyrus and improvements in task performance were observed during short-term motor skill learning (Floyer-Lea and Matthews, 2004). Activation of the (right) DLPFC is commonly reported during complex visuo-motor learning tasks (Jenkins et al., 1994; Fink et al., 1999; Toni and Passingham, 1999) and might serve as a temporal storage region of modality specific sensorimotor associations (for review see Halsband and Lange, 2006).

Several previous studies suggested that PMC has a major role in motor learning and control of sequential movements (Schubotz et al, 2001a,b). Activation in PMC during motor skill learning has been usually observed bilaterally. While activation of the right PMC reflects spatial processing (e.g., transformation of visual signals into an appropriate motor output), which has been hypothesized to be important during the initial phase of motor learning, activation of the left PMC seems to be involved in later phases of motor learning (Ghilardi et al., 2000). Furthermore, it has been shown that the right (ventral) PMC is involved in processes linked to sequential motor actions and might, therefore, have an important role in sequence generation from memory (Tanaka et al., 2009). This is in good accordance with a previous study showing that the right PMC has a role in working memory (Sadato et al., 1996).

Taken together our findings support the hypothesis that even subtle and fast evolving bidirectional structural brain alterations in left M1, right ventral PMC and right DLPFC appear to have a crucial role in specific aspects of complex motor skill learning. Similar fast evolving structural changes have also been observed in animals performing a complex motor learning task. More specifically, motor practice over 2 days resulted in significant spine remodeling that correlated with behavioral improvements after learning (Yang et al., 2009).

We did not detect any learning-related structural changes in the cerebellum which was shown to be involved in functional MRI studies (Debaere et al., 2004; Floyer-Lea and Matthews, 2004, 2005) and lesion studies (Werner et al., 2010) and has been suggested to be a crucial mediator for fine motor control and motor learning (Jueptner et al., 1997; Jueptner and Weiller, 1998; Hikosaka et al., 2002; Halsband and Lange, 2006; Ramnani, 2006). In the present study, the lack of learning-related structural alterations in the cerebellum was unexpected and certainly requires further investigation. However, we believe the learning-related functional changes in the cerebellum as previously described must not immediately translate into structural cerebellar alteration. Instead, our data suggest, that the temporal onset of the emergence of cerebellar structural changes could be potentially different from other cortical areas such as M1, PMC, and DLPFC. Maybe extensive training over several weeks or years is required to induce such structural changes in the cerebellum as previously shown in a cross-sectional study between musicians and non-musicians (Gaser and Schlaug, 2003a).

Consistent with the role of the cerebellum in procedural learning, we here demonstrated that the initial GM volume in the cerebellum (left crus 2 and right cerebellar lobule VI) correlated with the individual amount of motor skill learning in a SPFT and with learning associated structural changes. In other words, subjects who showed the largest gains in motor performance and largest learning-related structural changes were those with the largest initial GM volume in the cerebellum (left crus 2 and right cerebellar lobule VI). This in turn is in line with a recent study showing that individual variations in the function and structure of specific brain regions such as the cerebellum explain some of the differences in an isometric visuomotor tracking task (Tomassini et al., 2010). The finding is corroborated by the additional result that the initial (baseline) GM volume in the cerebellum correlated with the learning-related structural changes in M1, ventral PMC, and DLPFC. Indeed, functional and structural connectivity studies revealed that cerebellar lobule VI has strong connectivity with M1, whereas DLPFC has strong connections with cerebellar crus 2 that sub-serves motor function (Ramnani, 2006; Krienen and Buckner, 2009). Therefore, our results highlight the importance of selective cerebellar-cortical connections for successful motor learning and structural GM changes.

A limitation of the present study is that structural non-invasive brain imaging methods cannot identify the underlying neurophysiological mechanisms of such bidirectional brain alterations. Thus, animal models have to be conducted to understand how and why inter-individual differences in behavior and learning are related to brain anatomy. Since structural GM alterations were detected already after 5 days of short-term motor skill practice, neurogenesis as the sole underlying mechanisms seems rather unlikely (Draganski and May, 2008). Instead, changes as well as differences in cell size, intracortical myelin, intracortical remodeling of dendritic spines, formation and/or elimination of synapses and/or axonal sprouting seem more likely candidate mechanisms (Trachtenberg et al., 2002; Xu et al., 2009). Furthermore, individual differences in synaptic and/or neuronal pruning processes during development might also play a crucial role for the differential outcome of learning-related GM volume alterations (Kanai and Rees, 2011). Alternatively, our findings might be related to differences in metabolic demands during motor skill learning which in turn might cause bidirectional structural brain alterations (Black et al., 1990; Isaacs et al., 1992). Apart from these limitations in the interpretation of the underlying learning-related neural mechanisms, the present study highlights the importance of intra-individual learning success for structural brain alterations rather than the “mere practice.” Hence, we propose that the general view “to perform a learning task results in an increase in GM volume” cannot be generalized, but rather should be modified to consider task-specificity, learning success, duration of learning, and other, until now, unknown determinants.

Another potential limitation of the present study is the relatively small sample size (n = 13). Hence, it is reasonable to assume that the changes in GM structure are spuriously correlated with our behavioral measures. This, however, seems to be rather unlikely due to several reasons. First, our correlation coefficients were quite strong (r = 0.83 for all brain areas detected) even with the relatively small number of subjects tested. The brain-behavior correlations (see Figure 3B) were still significant after removing the subject with the highest change in GM volume and SPFT improvement [left M1: r = 0.691; p = 0.013; right ventral PMC: r = 0.780; p = 0.003; right DLPFC: r = 0.763; p = 0.004, (n = 12)]. Second, we could show that initial GM volume differences in the cerebellum before SPFT training did not only “predict” performance improvements with training but also seem to “predict” the extend of structural plasticity in brain areas where we found correlations with SPFT improvements. Taken together, this line of evidence renders spurious correlations rather unlikely.

Last but not least, we have to acknowledge the fact that we did not introduce a sedative control group in the present study with no training or a comparison group that performed the SPFT with only random sequences for the same amount of time. Hence, we cannot make definite inferences about the effects of the intervention (SPFT training) on structural brain plasticity. Even though a correlation in general cannot provide a causal relationship, the strong correlation between structural brain alterations and behavioral improvements in our study suggest an intuitive brain-behavior interaction. Future studies should be performed to further support our study claims.