Avalilação PREreview de Prefrontal stimulation prior to motor sequence learning alters multivoxel patterns in the striatum and the hippocampus

This preprint study published on Biorxiv sought to determine the influence of  individually tailored theta burst stimulation (TBS) of the dorsolateral prefrontal cortex on multi-voxel response patterns in several cortical and subcortical regions  relevant for motor sequence learning (MSL). In a repeated measures design, participants attended four sessions, where participants had a pre-stimulation resting scan, followed by either a neuronavigated tailored continuous or intermittent theta burst stimulation of an individually-defined DLPFC target outside the scanner. Following stimulation, participants trained on the motor task (sequential or random) whilst undergoing fMRI. Following training there was a second resting scan. Neither TBS protocols (continuous or intermittent TBS) had any detectable influence on MSL performance, where participants were required to perform a random or sequenced series of key presses. However the task did influence patterns of activity in 3 pre-defined cortical and subcortical regions of interest. Changes to activity patterns were tested using MVCS; a technique that measures the correlation of time series between each voxel in the selected ROI.   Inhibitory TBS modified the way that the task influenced patterns of activity in the putamen, and also influenced patterns of activity in the hippocampus during resting state scans. MVCS metrics increased during MSL in the DLPFC irrespective of stimulation type. A similar result was found in the sensorimotor putamen during the task, as well as for the hippocampus in a post-intervention rest scan, both under inhibitory, compared to facilitatory stimulation. 

We would firstly like to commend the interesting question and the combined use of TMS and brain imaging to approach this question. In particular, a priori definition of an anatomical circuit of interest and individually tailored TMS are good practices for combined brain stimulation and neuroimaging studies. As individuals have been shown to have unique functional architecture, tailoring designs that tap relevant cortical circuits seems to be pertinent in modulating specific networks across participants. We would also like to applaud the well controlled behavioural manipulation used to assess motor sequence learning after TBS. While these positives position this study well to help advance knowledge of how TMS modulation of neural activity patterns may influence motor learning processes, we have some concerns and questions that could be addressed, that we think would aid understanding for the readers of future versions of this manuscript. 

The authors report no influence of TBS on RT in both the random and sequential conditions. It is therefore difficult to interpret the findings that TBS modulated behaviourally-relevant activity in the regions of interest. We appreciate the discussion about this point, however the manuscript may benefit from an additional discussion on the relevance and importance of illustrating neurophysiological changes from brain stimulation without associated behavioural changes. Answering questions like “how the TMS associated activity pattern changes demonstrated in this study may guide future research in understanding the neural processes involved in motor sequence learning” could facilitate understanding on this point.

The authors only stimulated DLPFC and no other regions, i.e. there was no control site. We appreciate the time-costly and large endeavour that the current study reflects. However, without control site stimulation we are unable to conclude whether it is DLPFC stimulation, or general stimulation per se that drives the observed effects. We wonder if this work is based on previous studies that use  similar procedures and also include control sites? If so, a discussion of that data in the current manuscript would strengthen the current conclusions. Otherwise, the claims in this manuscript should include the caveat that we are unable to infer whether it is DLPFC specific stimulation that is driving the observed changes. 

As the manuscript is presented, it is unknown whether the task induced any meaningful signal changes in the time course data that was subject to the MVCS analysis. Interpretation of the data would be much aided if the authors could provide a figure of the time course data, averaged across the voxels within each ROI, for the random and sequential conditions. Ideally, a ROI of no interest to the current study would also be included. This would allow the reader to identify the extent to which the task induced systematic changes in the functional activity in the key ROIs, and would therefore improve confidence that the findings pertaining to the MVCS analysis are indeed meaningful.

We have one final question that if addressed may improve the overall comprehension of the study and its aims. Regarding the use of the multivoxel correlation structure analysis (MVCS), we think the manuscript would benefit from an additional discussion of the motivations for using this analysis, including the hypothesised metric changes, given assumptions regarding the anticipated changes in underlying neural activity.

Following these major points, we have listed a few minor points that will also help the reader better understand the study. 

1. Why was TBS order included as a regressor in the models when the current data did not suggest an effect? Was the parameter significant? Are the results the same if you don’t include it?
2. Is there any relevant literature on the duration of TBS effects? This will complement interpretation of the task and resting state fMRI scans.  
3. Why was no investigation of cross-region MVCS conducted? The introduction of the ROIs as being a task relevant network would suggest such an inter-ROI analysis could be both appropriate and informative. 
4. A discussion on the laterality of (or lack of) effects observed in the putamen would be appropriate considering the left hemisphere specific DLPFC stimulation. 
5. Another potential improvement in the statistical approach would be to establish a null distribution against which to compare MVCS metrics, by computing a null distribution based on shuffled data.
6. A final modification would be to consider reformatting the legend for Figure 3. Several readers made the same comprehension error in first pass reads of this figure so making it slightly clearer may help readers more easily understand the figure. One way to achieve this could be to change the legend so that the two sequence conditions are first then the two random conditions, so grouping them by task rather than stimulation. 

We look forward to hearing your thoughts and seeing subsequent versions of this paper.