Neurofeedback fMRI in the motor system elicits bi-directional changes in activity and white-matter structure in the healthy adult human brain

Neurofeedback can be used to alter brain activity and is therefore an attractive tool for neuromodulation in clinical contexts. Different contexts might call for different patterns of activity modulation. For example, following stroke, alternative therapeutic strategies could involve up or down-regulation of activity in the ipsilateral motor cortex. However, effects of such strategies on activity and brain structure are unknown. In a proof of concept study in healthy individuals, we showed that fMRI neurofeedback can be used to drive activity up or down in ipsilateral motor cortex during hand movement. Given evidence for activitydependent white matter plasticity, we also tested effects of activity modulation on white matter microstructure using diffusion tensor imaging (DTI). We show rapid opposing changes in corpus callosum microstructure that depend on the direction of activity modulation. Bidirectional modulation of ipsilateral motor cortex activity is therefore possible, and results not only in online changes in activity patterns, but also in changes in microstructure detectable 24 hours later.


31
Many neuropsychiatric and neurological conditions are associated with aberrant patterns of brain 32 activity so neuromodulation approaches that drive activity towards more favourable patterns are 33 of therapeutic interest. Neurofeedback (NF) taps into the brain's intrinsic capacity for activity  arguably more relevant to rehabilitation. Therefore, the current study aimed to address the degree 53 to which activity in ipsilateral motor cortex can be bidirectionally modified with fMRI NF in 54 healthy individuals during executed movements as a proof of concept test of how this approach 55 might be applied and tailored to the patient as an adjunct to neurorehabilitation. of white matter microstructure, provide a powerful approach to test these effects in humans. We 71 therefore employed real-time NF, using fMRI at 7 Tesla, to manipulate the activity of the 72 sensorimotor cortices (S1M1s) in opposite directions in two separate conditions, and tested for 73 effects on white matter structure against a sham group.

76
Each participant was scanned 4 times and experienced two different NF conditions (only one NF 77 condition was experienced in each session), with DTI acquired before each condition and again 78 24 hours later (Fig. 1A). During NF, participants were instructed to modulate the height of two 79 bars (representing activity in ipsilateral and contralateral S1M1) on a visual display, by moving 80 only their left hand during 30s movement blocks, which alternated with 30s rest blocks. In the 81 'Association condition' participants were required to co-activate both S1M1s (Fig. 1B, C), while 82 in the 'Dissociation condition' they were required to maximize contralateral S1M1 activity, 83 while minimizing ipsilateral S1M1 activity (Fig. 1B, D). Participants in the Sham group received 84 the same instructions but were shown the NF videos of a matched participant in the real NF 85 group, and experienced the same two conditions (Association and Dissociation). 80 scans were 86 successfully completed, 28 participants were enrolled and complete data sets were obtained in 20 87 participants. For each feedback condition, participants trained for approximately 20 minutes (in 3 88 or 4 runs of ~6 minutes). EMG was used to monitor hand movements online and as expected the 89 muscle activity of the moving (left) hand was significantly higher than the non-moving (right) 90 hand and was similar between Real NF and Sham groups ( Supplementary Fig. 1   We assessed fMRI activity to test whether participants could modulate S1M1 activity with 102 feedback as instructed. We first analysed signal change within the regions selected during NF 103 using a mixed ANOVA including within-subject factors of condition (Association, Dissociation) 104 and time (Run 1, 2, 3) and between-subject factor of group (Real, Sham). Instructions required 105 participants to increase ipsilateral S1M1 (iS1M1) activity for the Association condition and 106 decrease it for the Dissociation condition. Compared to the Sham group, participants in the NF 107 group were able to modulate activity in iS1M1 as instructed ( Fig. 2A, B; main effect of

(D) (E)
We used a data-driven approach and performed whole-skeleton voxel-wise non-parametric 148 permutation testing of these between-group differences, which revealed a statistically significant 149 cluster in the corpus callosum, no other clusters were identified elsewhere in the brain ( We expected that changes in white matter structure would reflect successful modulation of 171 activity with neurofeedback. For each participant we therefore identified which of the two 172 neurofeedback conditions they performed best (see Supplementary Table 3 for more details). We Our results support the hypothesis that bidirectional activity modulation of ipsilateral 195 sensorimotor activity during executed hand movement can be achieved via neurofeedback and 196 that this results in rapid, directional, and anatomically specific changes in white matter structure.

197
This finding in healthy individuals is relevant to considering application of neurofeedback in 198 therapeutic contexts and in particular as an adjunct to motor neurorehabilitation after stroke.

200
The two conditions here could potentially be used as alternative interventions in stroke patients. The structural findings suggest that alterations in the elicited brain activity is a possible mediator 248 of previously described experience-related white matter changes in the human brain resulting  Siemens scanner 4 times (Fig. 1A).

286
Unbeknownst to the participants they were assigned to two different groups: Neurofeedback or 287 Sham. Total number of analyzed scans was 80 (NF group n=10x4=40; sham group n=10x4=40).

288
One participant in the NF group did not complete the experiment due to back pain and DTI was 289 not acquired in 1 participant due to scanner crashes. Four participants in the Sham group did not 290 complete the full study and data was not fully collected in further 2 Sham participants due to 291 scanner crashes.

293
Participants were blind to group assignment. The experimenter could not be blinded to group due    Positive values were represented above the centre point, and negative values below the centre 357 point (Fig 1C,D). The bar for contralateral (right hemisphere) activation was displayed on the left 358 as this hemisphere should be most active during left hand movement. The bar for ipsilateral (left 359 hemisphere) activation was displayed on the right.

361
Sham Group

362
Participants in the Sham group were matched to a participant in the NF group and received 363 feedback videos from that participant (rather than their feedback from their own brain activity).

364
This allowed the Sham participants to have a similar experience as the NF group. All scans and 365 instructions received by the Sham group were identical to those received by the NF group. In this way, the goal of the Association condition was to maximise activation in both left and 374 right S1M1, whereas the goal of the Dissociation condition was to maximise right (contralateral) 375 S1M1 activity and minimize left (ipsilateral) S1M1 activity.

377
Participants were only told that the bars represented their brain activity. For both conditions other strategy as long as they did not move their right hand: participants were asked to report if they used the strategy and, if so, how successful they thought 409 the strategy was on a scale of 1-5.  Tractography analysis 489 We used tractography to identify the probabilistic connectivity map of the significant corpus