In this experiment we examined the neural correlates of a novel and complex skill acquisition task that combined perceptual, motor and cognitive processes in healthy middle-aged adults (40 - 50 years old). There was a significant improvement in performance following training. This performance improvement could not be attributed to the participant demographics, including their age, sex and education. We specifically tested for potential effects of sex, because of the documented effects of the menopause transition in cognition, e.g. (Greendale et al. 2020). At a neurophysiological level, there was increased engagement of both cortical and subcortical areas within a relatively short time, supporting improved task performance. And finally, we found significant associations between brain microstructure and training outcome.
Behavioural Performance
The number of successful trials increased significantly from the Early to the Late Learning phase of the PCM task, and the effect size was large and positive. The participants completed 160 trials within 31 minutes of training before the Late Learning phase. The fact that a relatively short training duration and a relatively low number of trials resulted in large performance improvement is in line with several studies showing significant task improvements following training, e.g., (Howard and Howard 1992) (Singer et al. 2003) (Rebok et al. 2007) (Spencer et al. 2007) (Basak et al. 2008) (Rieckmann and Bäckman 2009) (Nemeth and Janacsek 2011) (Wilson JK 2012) (Karbach J 2014). This improved performance in middle-aged adults is in agreement with findings in young adults, showing that practice with the PCM task led to better performance compared to controls that received no training (Bennett SJ 2018).
The PCM scores for our participants were low, and there was still much room for improvement. The mean percentage of successful trials significantly increased from the Early Learning (M = 24.71%) to the Late Learning (M = 37.00%). By comparison, in the study by Bennett et al. (2018), performance improved significantly in the post test for the practice group (M = 55%) compared to the control group that received no training (M = 20%) within 31 minutes of practice. One explanation for the lower training gains in our experiment could be that middle-aged participants are less able to learn the task than young adults. However, direct comparisons of the middle-aged adults and young adults of these different studies are difficult because the Bennett et al. (2018) study was behavioural, whereas the present study was carried out in an MRI scanner. Conceivably the task is more difficult in the scanner as participants have to habituate to the scanner environment, in addition to completing the task while lying down, with a head coil mounted around the face, visualising the computer screen through a mirror system, and manipulating an unfamiliar mouse apparatus without being able to see one’s hand.
It is likely that 31 minutes of training was not enough time for the middle-aged participants to achieve expertise at this complex task. Changes in motor skill performance are known to evolve slowly, requiring many repetitions over several training sessions (Karni 1996) (Ungerleider et al. 2002) (Doyon et al. 2003). The acquisition of motor skills follows distinct stages, including an early, fast learning stage, in which considerable improvement in performance can be seen within a single training session, and a later, slow learning stage, in which further gains can be observed across several training sessions (Nudo et al. 1996) (Karni et al. 1998) (Ungerleider et al. 2002) (Doyon et al. 2003). With extended practice, the skilled behaviour becomes resistant to both interference and the passage of time, and can be readily retrieved at reasonable performance levels despite long periods without practice (Penhune and Doyon 2002) (Ungerleider et al. 2002) (Doyon et al. 2003). While it is clear that the middle-aged participants demonstrated significant performance improvements and plasticity in the early fast learning stage, longer term training would be needed to see if further gains could be observed, and whether the perceptual-cognitive-motor skills could be trained to comparable levels seen in young adults.
Moreover, we tested whether impulsivity and computer game experience could predict the performance improvement for each individual. Impulsivity is associated with specific measures of dysfunction of inhibitory control, ranging from disinhibited comments to risk-taking and aggression in clinical and non-clinical populations, e.g. (Lemke MR 2005) (Lage GM 2012) (Enticott PG 2006). Although we expected that participants with lower impulsivity scores might improve their performance more during the learning phase, possibly by exercising more inhibitory control and suppressing inappropriate motor responses or decisions, there was no significant association of the BIS-11 scores, or their components, with task performance. It is likely that sophisticated kinematic measures of movements were needed to reveal differences associated with impulsivity scores, e.g. (Lage GM 2012) (Lemke MR 2005), rather than difference scores that capture overall performance accuracy and improvement. Computer game experience was associated with improved performance at the PCM task, which can be classified as near transfer to another type of computer game within the scanner environment. This result is in agreement with research that shows significant transfer benefits of computer gaming in the training of real world skills, such as surgery and flight performance (Enochsson et al. 2004).
fMRI
Whole brain analyses
We found a significant main effect of testing phase: bilaterally in the cerebellum and pons, in the R thalamus, R subthalamic nucleus (STN), and R lingual gyrus. Specifically, there was greater activity bilaterally in the cerebellum (crus I and II), and in the lingual gyrus during Early Learning. This means there was a decrease in the activity of those areas in Late Learning, which may indicate a fast switch to more efficient processing in early visual areas (Pfeifer et al. 2019), and in areas of the cerebellum that are connected with the dorsolateral prefrontal area 46 (Schmahmann 2019). The activation of the unimodal (visual) lingual gyrus (middle occipitotemporal area) is congruent with findings that show its activation enhanced when visual and tactile information are combined to strengthen the representation of the visual stimulus (Macaluso et al. 2000), suggesting back projections from multimodal convergence areas can feedback and modulate representations in a primary modality (Driver and Spence 2000).
Conversely, we observed increased activity in Late Learning in the cerebellum (vermis X), pons, thalamus, STN, precuneus, midcingulate cortex, SMA and paracentral lobule. Vermis X is part of the flocculonodular lobe, which is involved in visual tracking and oculomotor control (Cacciola et al. 2019). The STN regulates function related to the basal ganglia, which includes motor, as well as cognitive and motivational processes (Temel et al. 2005). Precuneus is central in integrated tasks that include visuo-spatial imagery, episodic memory retrieval and operations such as first-person perspective taking and experience of agency (Cavanna and Trimble 2006), which were important in the PCM task. Midcingulate cortex is crucial in execution of extended behaviours, by encoding distributed, dynamic representations of action sequences (Holroyd et al. 2018), which were a main feature of the PCM task. Finally, the SMA has a role in self-initiated movements (Jenkins et al. 2000). The significant activation and engagement of these areas within a short time corresponds to those proposed by Doyon and Ungerleider (2002) for motor skill learning.
We did not observe a significant main effect of performance or interaction between testing phase and performance. This is likely because this is an early learning stage, and the neural signatures for successful and unsuccessful trials are not differentiated enough to be detected reliably.
ROIs
Successful trial completion required working memory for navigating effectively to the target, anticipation/prediction of obstacle trajectories, and monitoring allocentric spatial relationships between objects, in addition to motor aspects, such as fine motor control, and adaptation to kinematics of self-referent motion and to cursor movement. Furthermore, the task goal could be achieved in multiple ways by executing cursor trajectories from a range of potential options. As expected, there was increased engagement of both cognitive and motor networks with just 31 minutes of training (160 trials).
Using an exploratory uncorrected threshold of p < .005 and k = 5 voxels, we observed increased activity in Late Learning in a number of areas, in line with studies showing increased activation in single-session training, before the task is well-practiced, e.g., (Nyberg et al. 2003) (Kelly and Garavan 2005) (Soldan et al. 2008) (Braver et al. 2009). We expected increased activity in the striatum, cerebellum, SMA, preSMA, M1, premotor cortex, ACC, dPFC, and inferior parietal cortex. Activity in this network has been interpreted as representing the enhanced demand for error correction (cerebellar cortex) and planning (premotor cortex) during early learning (Steele and Penhune 2010). These results for middle-aged adults are in line with findings in young adults showing that the early fast learning phase of motor skill acquisition elicits widespread activation in subcortical (basal ganglia, cerebellum, hippocampus), as well as cortical areas (SMA, preSMA, M1, premotor cortex, ACC, inferior parietal regions, and dPFC), e.g., (Grafton et al. 1995) (Sakai et al. 1998) (Ungerleider et al. 2002) (Doyon et al. 2003) (Floyer-Lea and Matthews 2005) (Albouy et al. 2008) (Steele and Penhune 2010) (Albouy et al. 2013) (King et al. 2013).
The Doyon and Ungerleider (2002) model proposes that there are two loop circuits, a cortico-striatal and a cortico-cerebellar system, which are both recruited during the early learning stage of motor skill training regardless of the type of motor task.
However, in the later stage, after several sessions of training, the cortico-striatal and cortico-cerebellar systems contribute differentially to different types of motor tasks. For example, for motor sequence training the cerebellum becomes no longer essential, and the long-lasting retention of the skill will now involve representational changes (reflected through increased activity) in the striatum and its associated motor cortical regions, including the parietal and motor-related structures (Doyon et al. 2003). By contrast, a reverse pattern of plasticity is proposed for motor adaptation (learning to adapt to environmental perturbations): the striatum is no longer necessary for the execution and retention of the acquired skill; increased activity in regions representing this skill will now be present in the cerebellum, parietal cortex and motor-related cortical regions (Doyon et al. 2003). Thus, both the cortico-striatal and cortico-cerebellar loops are recruited in the early stage of motor skill training, while the later stage of motor sequence learning recruits the cortico-striatal system, whereas motor adaptation skills recruit the cortico-cerebellar system. Both the cortico-striatal and cortico-cerebellar systems were recruited in middle-aged adults. Indeed, our findings corroborate the regions suggested to be recruited in the early learning phase of the model – we found increased activity in the striatum, cerebellum, motor cortical regions (e.g., premotor cortex, SMA, pre-SMA, ACC), as well as prefrontal and parietal areas. However, in the present experiment we did not assess motor skill acquisition over the entire course of learning, and thus cannot assess the late training stage in order to fully examine this model.
In a similar model proposed by Hikosaka et al. (2002), learning spatial coordinates during motor skill training is supported by a frontoparietal-associative striatum-cerebellar circuit, while learning motor coordinates is supported by an M1-sensorimotor striatum-cerebellar circuit. The Hikosaka et al. model postulates that the regions engaged in the early stage of motor skill training are associative and involved in the fast learning of spatial coordinates, whereas in the late stage sensorimotor areas engage in the slower learning of motor coordinates. In line with this model, we found increased activity in frontoparietal-associative striatum-cerebellar regions (i.e., dPFC, inferior parietal cortex, ACC, caudate, rostrodorsal regions of the putamen, and regions in the cerebellum) indicating that this circuit was recruited to learn spatial coordinates in the PCM task.
We also found increased activation in M1-sensorimotor striatum-cerebellar regions suggesting this circuit was being recruited to learn motor coordinates in the PCM task, although further increases in activity would be expected in this circuit with additional training as more expertise is achieved in the task. In addition, we found increased activity in premotor cortex, SMA, and preSMA, supporting the suggestion that transformation from spatial to motor coordinates involves these areas.
The increased activation in the vPFC and precuneus during post-training is surprising because other studies report increased activation during later stages of motor skill learning, e.g., (Doyon 1997) (Sakai et al. 1998) (Doyon 2002) (Ungerleider et al. 2002) (Doyon et al. 2003) (Lehéricy et al. 2005). Early activation in these regions may be related to the specific demands of the PCM task, with precuneus contributing to the need for visual–sensorimotor integration, and visuospatial attention and processing (Bushnell et al. 1981) (Posner and Rothbart 1992) (Petrides 1996) (Clower et al. 2001) (Doyon et al. 2002) (Cavanna and Trimble 2006), and vPFC contributing to response inhibition, goal-appropriate response selection, and abstract decision and action planning processes (Aron et al. 2004) (Aron et al. 2014).
Furthermore, the vPFC and precuneus are association areas of the cerebral cortex and early activation in these regions are in line with the Hikosaka et al. model, which suggests that associative regions are engaged early in the learning process to acquire spatial coordinates.
We found decreased activity post-training in cerebellum, hippocampus, and parahippocampal gyrus. This is contrary to what was expected; we expected PCM learning to be supported by increased activity in these regions as this was short-term training during the early learning stage of motor skill acquisition. The cerebellum is especially critical for early motor skill learning and its activity is not thought to decrease until the later phase of motor skill acquisition after longer-term training, e.g., (Jenkins et al. 1994) (Grafton et al. 1995) (Doyon et al. 1996) (Penhune and Doyon 2002) (Tamás Kincses et al. 2008) (Doyon et al. 2009) (Orban et al. 2010) (Steele and Penhune 2010). While we did indeed find increased activity in the cerebellum in Crus I, lobule IV-V, and cerebellar vermis X is in line with the above studies, we also found decreased activity in cerebellum Crus I, Crus II, and lobule VI. This result seems inconsistent but may reflect anatomical and functional differentiation in the cerebellum between sensorimotor and cognitive regions (Schmahmann 2019).
The sensorimotor cerebellum is mostly in the anterior lobe (lobules I through V), parts of lobule VI, lobule VIII, and the cerebellar vermis is interconnected with the vestibular and other brainstem nuclei which are engaged in midline body control, gait, and equilibrium (Schmahmann 2001) (Kelly and Strick 2003) (Schmahmann et al. 2004) (Habas et al. 2009) (Krienen and Buckner 2009) (O'Reilly et al. 2010) (Buckner et al. 2011) (Guell et al. 2018b; Guell et al. 2018a) (Stoodley and Schmahmann 2009); whereas the cognitive cerebellum is in the posterior lobe (Buckner et al. 2011) (Guell et al. 2018b; Guell et al. 2018a) (Schmahmann 2019). Task-based fMRI using cognitive paradigms has demonstrated that there are functionally distinct regions within the cerebellar posterior lobe with lobule VI engaged in visuospatial tasks; and lobules VI, Crus I, Crus II, and VIIB activated by executive functions such as working memory and planning (Stoodley and Schmahmann 2009) (Stoodley and Schmahmann 2010) (Stoodley et al. 2012; Stoodley et al. 2016). Interestingly, we found increased activity mostly in sensorimotor areas of the cerebellum, i.e., lobule IV-V and cerebellar vermis X; while decreased activation was found in the cognitive regions, i.e., Crus I, Crus II, and lobule VI, which are involved in visuospatial processing and executive functions. Decreases in these regions may indicate more efficient processing, supporting the improved performance after 31 minutes of training.
Our findings are also consistent with the role of the cerebellum in correcting motor errors, with a decrease in cerebellar activity being associated with a reduction in error rate and improved performance (Flament et al. 1996) (van Mier et al. 2004; van Mier et al. 1998) (Imamizu et al. 2000) (Van Mier and Petersen 2002) (van Mier et al. 2004) (Orban et al. 2010). Thus, the decrease in cerebellar activity seen in the present experiment may be related to the significant performance improvement at the PCM task. However, it should be noted that despite the improvement, the error rate was still quite high (M = 53.75%) and there was further room for learning.
Climbing fibres in the cerebellum not only encode sensorimotor error signals, but also a timing error (Kitazawa et al. 1998) (Doya 2000) (Medina et al. 2000) (Sakai et al. 2000) (Hikosaka et al. 2002). In particular lobule VI is a key structure for the timing of movement (Sakai et al. 2000) (Schubotz and von Cramon 2001) (Sakai et al. 2004). Sakai et al. (2000) examined how the brain decides ‘what to do’ (response selection) and ‘when to do it’ (timing adjustment). The preSMA was selectively involved in response selection, whereas the cerebellar posterior lobe was selectively involved in timing adjustment. An essential element for successful completion of the PCM task is the timing of movement, for example, when to move the white cursor in order to avoid the green objects. Thus, the decreased activity that was found in lobule VI may reflect improved timing of movement in the PCM task.
The decreased activity in the hippocampus and parahippocampus were also contrary to what we predicted. The hippocampus has shown increases in activity in both the early and later stages of motor skill learning (Schendan et al. 2003) (Albouy et al. 2008) (Fernández-Seara et al. 2009) (Gheysen et al. 2010) (King et al. 2013). However, decreased hippocampal activity in the early learning stage, or increased activity only in the later phase of motor skill training, has also been reported, e.g. (Jenkins et al. 1994) (Schendan et al. 2003) (Steele and Penhune 2010). For example, Jenkins et al. (1994) found an extensive decrease in the activity of the hippocampus in both new learning during the initial stage of motor skill training, and during the overlearned sequence in the later stage of training. The authors suggest that this is evidence that motor learning need not engage the hippocampal system. In the study by Steele and Penhune (2010), hippocampal regions increased in activity on day 2 of training, but were not part of the early learning network identified on day 1. The relatively few studies investigating the role of the hippocampus in motor skill training have yielded heterogeneous findings that may be the result of different types of tasks tapping into different cognitive processes. Thus, the research on hippocampal activation in motor skill learning remains contradictory and inconclusive.
The parahippocampal cortex is involved in many cognitive processes, including visuospatial processing (van Strien et al. 2009) (Aminoff et al. 2013) (Hohenfeld et al. 2020). This region is engaged by tasks involving scene perception, spatial representation, and navigation, e.g. (Aguirre et al. 1996; Epstein and Kanwisher 1998) (Maguire et al. 1998) (Mellet et al. 2000) (Ekstrom et al. 2003) (Janzen et al. 2007) (Kravitz et al. 2011) (Mullally and Maguire 2011) (Park et al. 2011) (Stevens et al. 2012) (Aminoff et al. 2013). Visuospatial processing is a key aspect of the PCM task. For example, monitoring the location of the green objects with respect to the cursor and to each other is a necessary component of the task. Thus, we predicted increased activation in this region. The observed decreased activation after half an hour may be an indication of early increased efficiency in the region for this task.
Correlation of Learning Index and activity in the putamen and anterior cingulate
Increased activity in the putamen and ACC in Late Learning was correlated with training improvement. This is in line with studies showing that the rostrodorsal (associative) regions of the putamen are involved early in the learning process and are critical for acquiring a new motor skill, by extracting action value representations (Jueptner et al. 1997) (Lehéricy et al. 2005) (Averbeck and Costa 2017). By contrast, activity in the caudoventral (sensorimotor) areas of the putamen increases as a function of practice, suggesting that this region is involved in the execution of well-learned motor skills (Jueptner et al. 1997) (Lehéricy et al. 2005) (King et al. 2013). Lehericy et al. (2005) demonstrated that performance was positively correlated with signal changes in areas activated during early learning, including the associative putamen. Conversely reaction time was negatively correlated with signal changes in areas activated during late learning stages, including the sensorimotor putamen. In addition, Jueptner et al. (1997) showed that the shift of activation from the associative to the sensorimotor territories of the putamen was already completed after 50 min of training. These results indicate that motor representations shift rapidly from the associative to the sensorimotor territories of the putamen during early learning. Notably, we found increased activation in both rostrodorsal and caudoventral areas of the putamen, providing support for the notion that motor representations shift from the associative to the sensorimotor territories of the putamen during learning.
A central role of the putamen in motor skill learning is the processing of reward prediction error signals, the discrepancy between the reward and its prediction. These signals originate from midbrain neurons that provide the basal ganglia with dopaminergic inputs, e.g. (Aosaki et al. 1994) (Jog et al. 1999) (Doya 2000) (Hikosaka et al. 2002) (Schultz et al. 2003) (Orban et al. 2010). Reward error signals attach a positive value to actions and objects accurately predicting positive outcomes in the early stage of learning, which shapes adaptive behaviour, e.g. (Doya 2000) (Hikosaka et al. 2002) (Averbeck and Costa 2017). Reinforcement Learning (RL) provides a useful framework for the learning of action and object values, and their consequences in terms of punishments and rewards in a given environment. A large body of experiments has mapped the RL mechanisms on fronto-striatal brain areas, modulated by midbrain dopaminergic inputs, and in association with the amygdala and the thalamus (Averbeck and Costa 2017). A study that tracked the dynamic representation of action values during learning (Seo et al. 2012), found that both the value of actions chosen based on previous outcomes, as well as those based on immediately available perceptual information, are represented in the dorsal striatum (caudate nucleus and putamen). This would be relevant for the PCM task, as both memorised successful action choices and immediately available visual information for action selection would be crucial for navigating each trial.
Putamen activation also increases in non-motor tasks involving a reward prediction error (Daniel and Pollmann 2012) (Sommer and Pollmann 2016). Sommer & Pollmann (2016) investigated if the occurrence of a target in a visual search display would elicit an increase of activation if the target's location is predicted by a previously learnt spatial context. They observed increased putamen activation when visual search targets were presented at the location predicted by the spatial context and when the prediction was uncertain (50% probability = prediction error), rather than certain (100% probability = no prediction error). Thus, they demonstrated an intrinsic prediction error signal in the putamen in memory-driven visual search. Similarly, in the PCM task, a successful trial would result in a positive intrinsic reward for a particular pattern and for the trajectory taken to the target, whereas an unsuccessful trial would generate a prediction error signal because at the early stage of learning, the possibility of reward is still very uncertain. Overall, the increased activity in the putamen most likely reflected a reward prediction error signal during task learning. As the activity in the putamen increased, indicating greater processing of reward prediction error signals, so the behavioural performance and training outcome improved.
As with the putamen activation, the significant positive correlation of activity in the ACC and task improvement most likely reflects a neural prediction error signal. The ACC plays a central role in error detection and performance monitoring, and several studies have reported activity in response to negative feedback (Carter et al. 1998) (Gehring and Knight 2000) (Luu et al. 2000) (Procyk et al. 2000) (Daniel and Pollmann 2010), for an overview see (Ridderinkhof et al. 2004). The ACC activity contributes to a signal that has been termed feedback-related negativity (FN) or error-related negativity (ERN), indicating violations of expected outcomes. Similarly to midbrain dopamine neurons, the FN differentiates unpredicted rewards from unpredicted non-rewards (Averbeck 2017), and has been reported in human EEG studies, e.g. (Holroyd et al. 2009), as well as in single unit recordings in the rodent ACC (Hyman et al. 2017).
The PCM task required evaluation of the trial outcomes based on whether a choice of actions led to a successful trajectory to the target. The increased activity in both the putamen and ACC was significantly correlated with a better training outcome. Both regions process reward prediction error signals. ACC is primarily responsible for strategy selection, e.g. whether to approach the target from a particular direction, while the prefrontal and dorsal striatum circuits are responsible for the execution of the required actions (Averbeck 2017). Therefore, our results provide evidence that extracting action values at a high, strategic level in the ACC, and a more specific way in terms of action execution in the putamen, are critical steps in the early stage of task learning to optimise action selection and maximise performance.
Diffusion Data
Diffusion indices can be used to indirectly localise microstructural variation that might be indicative of learning outcome. Indeed, we found significant relationships between MD, FA, ODI, and the Learning Index. Our results show that inter-individual variation in brain structure was associated with extent of learning in middle-aged adutls. This is in line with studies using diffusion MRI in young adults to demonstrate relationships between tissue microstructure and performance on cognitive and motor tasks (Klingberg et al. 2000) (Moseley et al. 2002) (Madden et al. 2004) (Tuch et al. 2005) (Johansen-Berg et al. 2007) (Sasson et al. 2010) (Sagi et al. 2012) (Hofstetter et al. 2013). For example, Johansen-Berg et al. (2007) used DTI to show that variation in white matter integrity, indexed by FA, in the corpus callosum is significantly associated with variation in performance of a bimanual coordination task, supporting the idea that variation in brain structure reflects inter-individual differences in skilled performance. Our results are also in line with diffusion imaging studies in older adults investigating associations between brain microstructure and performance in cognitive and motor domains (Bennett et al. 2011) (Nazeri et al. 2015). For example, Bennett et al. (2011) found that caudate–dPFC and hippocampus-dPFC tract integrity were significantly related to motor skill learning in healthy older adults (aged 63–72 years). Specifically, for both tracts, higher integrity, indexed by FA, was associated with greater motor sequence learning. Our results provide strong evidence of a relationship between brain microstructure and learning outcome, such that pre-existing inter-individual differences in brain structure could determine variations in skill learning.
DTI and NODDI indices
We correlated the diffusion indices with the Learning Index, and we hypothesised that parameters indicating grey matter complexity and white matter integrity would be associated with better learning. We found a significant negative correlation between the training outcome and MD in the grey matter of the left middle temporal gyrus (specifically in human mid-temporal area: hMT+/V5) and bilaterally in the cerebellum (left Lobules IV, V, VI, right lobule VI); a significant negative correlation of the training outcome and FA in the grey and white matter of right SMA; and a significant positive correlation of the training outcome and ODI in the white matter of right SMA.
The area hMT+/V5 is especially critical for the perception of visual motion, e.g. (Zeki 2015), and therefore central for improving performance at the PCM task. Lobules IV and V are in the anterior lobe of the cerebellum, and process sensorimotor information, e.g. (Stoodley et al. 2012) . Therefore, our result is consistent with the contralateral connections between the cortical and cerebellar hemispheres for right-handed participants. Lobule VI is in the posterior lobe of the cerebellum and, additionally to sensorimotor processing, it has association area projections, including with temporal, parietal and prefrontal areas. Moreover it processes cognitive information, and shows bilateral activations in working memory tasks (Stoodley et al. 2012). Consequently, the bilateral lobule VI involvement in our experiment is consistent with the cognitive demands of the PCM task. Furthermore, the ipsilateral SMA FA and ODI indices correlation with the training outcome is consistent with the SMA ipsilateral connectivity with the primary motor (M1) hand area. Functionally, stimulation of SMA can lead to LTP-like or LTD-like effects in M1, providing a potential physiological mechanism for neuroplasticity (Nirkko et al. 2001) (Koch 2020).
MD can indicate tissue density, e.g. (Basser 1995) (Pierpaoli and Basser 1996) (Assaf and Pasternak 2008) (Sagi et al. 2012) (Hofstetter et al. 2013). Lower levels of MD correspond to lower water diffusion rates, resulting from greater tissue density, i.e., a greater density of axons or dendrites, which restricts the overall rate of diffusion. Accordingly, we expected lower MD to be associated with greater improvement in PCM performance and indeed, this was the case in the grey matter of left hMT+/V5 and bilaterally in cerebellum. This is in agreement with other DTI studies that demonstrated an association between reduced MD in grey and white matter and greater task improvement (Sagi et al. 2012), (Hofstetter et al. 2013). For example, Sagi et al. (2012) examined grey matter microstructure in participants performing a spatial navigation task. They showed significant negative correlations between improvement rate on task performance and MD reduction in the left hippocampus and right parahippocampus. Using the same task, Hofstetter et al. (2013) investigated white matter microstructure and found that improvement on the task correlated with reductions in MD in the fornix. However, although diffusion metrics are sensitive markers for subtle microstructural tissue organisation, they are not specific and are difficult to attribute to particular biological processes (Dowell et al. 2019). So, although we have established a clear relationship between lower MD and better training outcome on the PCM task, we can only speculate as to the cellular mechanisms underlying the variation in structure that supports better learning on this task.
FA refers to the orientation of water diffusion, independent of rate, and is a measure of fibre organisation and integrity (Basser 1995) (Basser and Pierpaoli 1996) (Pierpaoli and Basser 1996) (Assaf and Pasternak 2008) (Bennett et al. 2011) (Sagi et al. 2012). Higher FA values indicate that the diffusion of water molecules is restricted in the direction along axons, that fibres are more coherent and aligned, reflecting higher tissue integrity. Lower FA values indicate diffusion in the perpendicular direction. Several previous studies have shown that higher FA is associated with improved behavioural performance on visuospatial and cognitive tasks, e.g. (Klingberg et al. 2000), (Madden et al. 2004), (Wolbers et al. 2006), (Johansen-Berg et al. 2007), (Sampaio-Baptista et al. 2013). However, there are also reports of correlations of lower FA values with task performance. For example, Hofstetter and colleagues (2013) showed that reductions in FA values in the fornix were correlated with improvement on a spatial learning and memory task. Tuch et al. (2005) demonstrated that slower reaction times on a visuospatial task, i.e. worse task performance, were significantly correlated with higher FA in white matter of the right optic radiation, right posterior thalamus, right medial precuneus, and left superior temporal sulcus.
FA is a complex measure that is influenced by myelination, axon diameter, axon density (Beaulieu 2002), as well as by path geometry and the presence of crossing fibre pathways (Johansen-Berg et al. 2007), (Jones et al. 2013), (Jeurissen et al. 2013). In areas with intravoxel fibre crossings, higher FA of an individual fibre population can result in a lower overall FA (Wiegell et al. 2000), (Pierpaoli et al. 2001), (Tuch et al. 2003), (Tuch et al. 2005). The SMA contains a large number of connections, including short association or U fibres that connect neighbouring gyri, e.g. from the pre-SMA to the SMA and from the SMA to M1, e.g. (Bozkurt et al. 2017). The strong relationship between lower FA in both grey and white matter of the SMA and better training outcome on the PCM task that we demonstrated may be at least partly explained by this anatomical feature.
Using NODDI, we found a significant positive correlation of training outcome with ODI in the white matter of the right SMA. This was in contrast to our prediction of lower ODI values in the white matter, which would indicate less dispersion of water molecules and thus tracts that are more compact, parallel, directional and aligned (Zhang et al. 2012), resulting in faster signal transmission (Tuch et al. 2005).
The fact that the last two correlations of the FA and ODI indices are found in the ipsilateral SMA can be interpreted as additional recruitment during a time window of intense motor learning, planning and plasticity, potentially sharing similarities with longer reorganisation periods requiring recruitment of the SMA in both hemispheres, e.g. (Zemke et al. 2003), (Sailor et al. 2003), (Pinson et al. 2021). Furthermore, NODDI provides a way of interpreting changes in FA, i.e., to decouple the effects of axonal density (higher density would increase FA) and orientation dispersion (higher dispersion would decrease FA). The correlations with FA and ODI were overlapping in the SMA – within which we saw lower FA and higher ODI correlated with better training outcomes. Therefore, the ODI results confirm that the associations with FA reflected primarily an effect of orientation dispersion. The fact that the correlation is in the opposite direction could be due to the specific structure of SMA, and the fact that it contains many connections with multiple directions.
Limitations
One of the limitations of our study is that it does not provide data on how the activated regions interact with one another and how information is transferred from one circuit to another during the course of motor training, for example, from the associative to the sensorimotor circuit, i.e., transformation from spatial to motor coordinates (Hikosaka et al. 2002). Functional and effective connectivity approaches could be used to assess connections between regions of a network, as well as between networks.
The question regarding the detailed underlying biological mechanisms of the observed relationships between the diffusion indices and training outcome in the PCM task cannot be addressed in this study. Histology offers the possibility to validate diffusion indices and to shed light on the cellular events that underlie the measures obtained in human neuroimaging studies of motor skill training (Sampaio-Baptista et al. 2013). An animal study with a similar PCM protocol that correlates diffusion indices with histological measures such as the number of synaptic vesicles, number of dendritic spines and astrocytic processes would provide further information on the mechanisms underlying better training outcomes. Indeed, evidence suggests that in both grey and white matter, there is a strong link between neurite morphology determined from diffusion MRI and independent measures derived from histology (Sagi et al. 2012) (Zhang et al. 2012) (Hofstetter et al. 2013) (Sampaio-Baptista et al. 2013).
We have demonstrated specific associations between diffusion indices and training outcome in healthy middle-aged adults, suggesting that inter-individual variation in brain structure influences variation in skill learning. However, as this is a correlation study, we cannot confirm a causal role of brain structure on differences in skill learning behaviour.
Finally, our experiment had a short timescale, and future research should use a longitudinal design to investigate the relationship of motor training with functional and structural brain changes and performance. There is emerging evidence that changes in diffusion indices can also occur in response to short-term training (Sagi et al. 2012) (Hofstetter et al. 2013) (Marins et al. 2019). For example, Marins and colleagues (2019) trained healthy individuals to reinforce brain patterns related to motor execution while performing a motor imagery task. After just one hour of training, participants showed increased FA in the sensorimotor segment of corpus callosum. Therefore, it may also be possible to design a pre- and post-training study of structural brain changes with short-term training.