Enhanced attention-related alertness following right anterior insular cortex neurofeedback training

Summary The anterior insular cortex, a central node of the salience network, plays a critical role in cognitive control and attention. Here, we investigated the feasibility of enhancing attention using real-time fMRI neurofeedback training that targets the right anterior insular cortex (rAIC). 56 healthy adults underwent two neurofeedback training sessions. The experimental group received feedback from neural responses in the rAIC, while control groups received sham feedback from the primary visual cortex or no feedback. Cognitive functioning was evaluated before, immediately after, and three months post-training. Our results showed that only the rAIC neurofeedback group successfully increased activity in the rAIC. Furthermore, this group showed enhanced attention-related alertness up to three months after the training. Our findings provide evidence for the potential of rAIC neurofeedback as a viable approach for enhancing attention-related alertness, which could pave the way for non-invasive therapeutic strategies to address conditions characterized by attention deficits.


INTRODUCTION
Cognitive control and attention facilitate flexible and goal-directed behaviors in an ever-changing environment.Deficits in these cognitive functions are debilitating for patients suffering from attention-deficit hyperactivity disorder (ADHD), 1 schizophrenia, 2 traumatic brain injury, 3 Parkinson's disease, 4 and other psychiatric and neurological disorders.][8][9][10] Despite this proposed role of the rAIC, little is known about the feasibility of improving cognition and attention by modulation of neural activity in the rAIC.Here, we used a longitudinal real-time fMRI (rt-fMRI) neurofeedback training design to modulate rAIC activity and investigate the impact on attention and cognition.
The salience network encompasses the bilateral anterior insula and anterior cingulate cortex (ACC) and facilitates the detection and filtering of salient events. 5,7,10An influential neurocognitive model has emphasized the critical role of the anterior insula by extending its function to the triggering of other large-scale brain networks and thereby influencing higher cognitive processes such as attention and cognitive control. 7For example, using analysis of dynamic interactions, it has been shown that the rAIC drives the switching from the deactivation of the default mode network (DMN) to activation of the central executive network during rest and attention tasks. 102][13][14] Using multivariate dynamical systems state-space models, Cai et al. 12 found that the causal influence from the rAIC to the ACC increased with higher demand for cognitive control and correlated with general cognitive control abilities across three different cognitive tasks.A study employing graph theory analysis of fMRI data found that the rAIC has strong connections to other important ''inhibition nodes'' during response inhibition, suggesting
To test whether participants in the V1 group were able to upregulate activity in their target ROI, we computed two-way ANOVA and linear regression for extracted V1 activity.For both, right and left V1 activity the two-way mixed ANOVA did not show any significant effect indicating no significant change from the first baseline run to the last transfer run independently of the feedback group.
Additionally In the rAIC and the mental-rehearsal group, the activity for both left (rAIC group p = 0.14, mental-rehearsal group p = 0.08) and right V1 (rAIC group p = 0.82, mental-rehearsal group p = 0.31) did not change.In addition, we analyzed activation in the whole V1 (right and left V1 mask merged).The two-way repeated measure ANOVA did not show any significant main effect or interaction effect.The linear regression revealed only for the right V1 (b = 0.03, R 2 = 0.03, F (1, 118) = 3.9, p = 0.05) and the left V1 (b = 0.05, R 2 = 0.06, F (1, 118) = 7.9, p = 0.005) group a significant increase of overall V1 activity across neurofeedback runs.
Participants reported trying a wide range of mental strategies to increase brain activity, common strategies included mental arithmetic (e.g., division, multiplication, or Fibonacci sequence) or emotional memories (e.g., thinking of sad, successful, or fearful situations).Importantly, strategies reported during transfer runs were similar to strategies used during the baseline runs.A list of strategies used during baseline and transfer runs is shown in supplemental information (Tables S1 and S2).
Together, these results demonstrate that participants in the rAIC group learned to upregulate activity in their rAIC over the course of neurofeedback runs and maintained self-regulation in the absence of feedback.Additionally, they increased activity in other brain regions associated with attention and cognitive control.None of these effects were observed in the other groups.

Short and long-term attentional effects of rt-fMRI neurofeedback
We went on to test whether voluntary increase of rAIC activity resulted in improved performance in behavioral tests.Participants performed before (pre rt-fMRI), just after (post-training), and three months after the rt-fMRI neurofeedback training (3-month FU), a comprehensive test battery implemented in the Psychology Experiment Building Language (PEBL) 34 containing the Attentional Network Task (ANT), a decision rule switching task (Switcher) and Corsi block task measuring visuo-spatial short-term working memory.Furthermore, a Go/NoGo and Gabor patch task were employed.
We used two-way mixed ANOVA with the factors group (rAIC, V1, mental-rehearsal) and session (pre rt-fMRI, post-training, 3-month FU) for each task to probe for effects of rt-fMRI neurofeedback training on behavior.Significant main effects and interactions were only found in the ANT task, which is a combination of a flanker and cueing task and simultaneously assesses alerting, orienting, and executive attention (see STAR Methods section for details).For the alerting component (reaction time (RT) no cue -double cue), a two-way mixed ANOVA with session as within-subject factor (pre rt-fMRI, post-training, 3-month FU) and group as between-subject factor (rAIC, V1, mentalrehearsal) revealed a main effect of session (F (2,88) = 3.66, p = 0.03) and a group 3 session interaction (F (2,88) = 3.92, p = 0.02).The post-hoc comparison showed a significant increase in alerting from pre rt-fMRI to post-training (t( 21 For orienting attention, a three-way mixed ANOVA with cue type (spatial cue, central cue) and session as within-subject factors, plus group as a between-subject factor, was performed.The results revealed a main effect of session (F (2, 88) = 13.13,p = 0.001) indicating faster RTs across sessions.Furthermore, the main effect of cue-type (F (1, 44) = 265.607,p = 0.01) was significant which shows overall faster RT for spatial cues (428 ms G 47.8 ms) compared to center cue (458 ms G 49.3 ms).
For executive control, a three-way mixed ANOVA with flanker type (incongruent, congruent) and session as within-subject factors, plus group as a between-subject factor, revealed the main effect for session (F Overall, these results show that participants increased alertness during an attention task, both short-and long-term, after rAIC rt-fMRI neurofeedback training.However, no rAIC specific neurofeedback training effect could be observed for the response inhibition, flanker, and spatial cuing task.

Correlation analysis self-regulation and behavioral changes
We tested whether improvements in behavior were correlated with activity modulation in the rAIC following rt-fMRI neurofeedback training.Since only the alerting effect and especially the RT for double cue showed rt-fMRI group associated changes, exploratory correlations were performed with the regulation success for the rAIC activity (transfer effect, linear regression over rt-fMRI runs and last minus first rt-fMRI run).Correlating the transfer effect in rAIC activity with the change in RT for double cue from pre rt-fMRI neurofeedback training to the time point ''post-training'' did not show significant results in any of the examined groups.In addition, change in the overall alerting effect was not correlated to the extent of rAIC upregulation.

Self-report and questionnaires
There was no significant difference between groups in the Hospital Anxiety and Depression Scale (HADS) 35 , Barratt Impulsiveness Scale (BIS-11) 36 scores and motivation or attention during rt-fMRI neurofeedback sessions (Mann-Whitney U-tests p > 0.05).Furthermore, a oneway ANOVA revealed no significant group differences regarding age (F (2,53) = 0.501, p = 0.61) or head motion (FD) (F (2,53) = 0.055, p = 0.946).Nine out of ten participants in the mental-rehearsal group reported that they believe they performed real rt-fMRI neurofeedback training.The duration between rt-fMRI neurofeedback sessions was on average 2.75 days (SD 2.24 days).The duration for participants in the mental-rehearsal group (2.9 G 2.

DISCUSSION
We used a longitudinal rt-fMRI design to target the rAIC with the goal of increasing cognitive control and attention based on neurofeedback.We found that participants were able to gain voluntary control over rAIC activity after two rt-fMRI neurofeedback training sessions in which they were provided with feedback about their ongoing rAIC activity.Furthermore, participants showed increased attention-related alertness during a speeded attention task, which persisted three months after the training.Our findings extend evidence for the critical role of the rAIC in attention-related alertness and provide initial insights into the potential of rt-fMRI neurofeedback training to enhance long-term specific components of attention in a variety of clinical conditions associated with rAIC dysfunction and attentional impairments.
The first goal of our study was to determine whether healthy participants can gain voluntary control over their rAIC activity without being instructed to use rAIC related strategies.To address this question, we split healthy participants in an experimental group receiving feedback about their rAIC activity and two control groups, one training to increase V1 activity and another not receiving any feedback during the neurofeedback sessions.Participants in the experimental group were able to modulate rAIC activity and maintain continuous increases of rAIC activity across runs spread over two separate days (within one week) during intermittent neurofeedback.Moreover, participants maintained their self-regulation after the second neurofeedback session even in the absence of neural feedback.
A meta-analysis by Emmert et al. 37 identified the anterior insula as a key component of a regulatory network activated during neurofeedback training, regardless of the targeted area.Consistent with this, our study observed a tendency for heightened rAIC activity over runs in the V1 group.However, the absence of a transfer effect in both the V1 and mental-rehearsal groups suggests that the observed modulation of activity in the rAIC group was specifically tied to the feedback received during neurofeedback sessions, rather than the general process of regulation.This distinction highlights the unique impact of targeted feedback in the rAIC group.
Our implicit study design including two control groups allowed to exclude most of the known confounding factors in rt-fMRI neurofeedback studies. 31,38Specifically, with our mental-rehearsal group we tried to account for a possible effect of mental strategy training on cognitive control and attention, however, due to the fact that we did not provide a specific strategy to participants in the rAIC group (implicit rt-fMRI neurofeedback), and they applied a variety of mental strategies, it was not possible to provide control participants with a specific strategy.Nevertheless, control participants got the same instruction as the other participants and importantly, after the third session, nine out of ten participants believed that they did real rt-fMRI neurofeedback training even though they never saw a feedback signal.Furthermore, reported strategies during baseline and transfer runs were similar across the groups.
Participants in the V1 group were able to increase activity in their target ROI across rt-fMRI neurofeedback runs, yet they failed to transfer their self-regulation skills to the transfer run without feedback, which may explain the absence of changes in perceptual sensitivity for the Gabor patch task.0][41] However, in all these studies participants learned to regulate activity in a specific subpart of their V1 that was sensitive to visual stimulation during functional localizer runs.Since we did not use functional localizer in our study but rather presented participants with a feedback signal computed as average over a whole V1 in one hemisphere, the lack of self-regulation might be explained by low specificity of the feedback signal.Another reason could result from the small sample size due to the split in right and left V1 group (N = 12 for the right and left V1 groups, respectively).Our rt-fMRI neurofeedback findings are in line with previous studies showing modulation of rAIC activity as a consequence of neurofeedback training combined with explicit rAIC related emotion regulation-related mental strategies. 25,30Critically, we extended these findings by showing that participants can increase rAIC activity without knowing what brain region and behavior is intended to be changed.
The second goal of the present study was to assess the long-term impact of rAIC activity modulation on attention.We used the ANT task which is a combination of spatial cueing task 42 and flanker task. 43The combination of cue and flanker conditions allow assessing three attentional subnetworks within a single task.Alerting attention is defined as the achieving and maintaining of intrinsic alertness (''readiness'') to respond to stimuli, while orienting refers to the selection/shifting of attention toward information among various sensory input and executive control is the process which allows resolving of conflict in mental operations and responses. 44,45Contrary to our initial hypothesis, our findings revealed significant effects solely on the alerting component of attention.Notably, improvements in alertness were sustained for at least three months following the modulation of rAIC activity, underscoring the robust nature of these specific cognitive enhancements (Figure 4).Our analysis revealed that the change in alertness was driven by faster RTs for double cue rather than longer RT for the no cue condition.This is a key finding, since larger alerting scores due to longer RT in the no cue condition would reflect difficulty in maintaining alertness. 46hus, our findings suggest that participants in the rAIC group showed increased degrees of alertness after the neurofeedback training.
Notably, only the rAIC group showed a significant improvement in alertness right after neurofeedback training, which was maintained at three-month follow-up (Figure 4A).Likewise, even though RTs for double cue trials were faster in most groups (Figure 4B), this improved performance was strongest and only significant in the rAIC group.We did not observe a significant correlation between rAIC upregulation and individual improvements in alertness.This is consistent with reports that among the few clinical fMRI neurofeedback studies (28% of the studies analyzed) that report correlations, 35% found no significant correlation between regulatory success and behavioral changes. 47uture studies could benefit from incorporating network analysis across multiple brain regions, which may offer a more complete understanding of sources of individual differences underlying neurofeedback effects.In line with our behavioral findings, prior research has highlighted the important role of the rAIC and adjoining frontal operculum in the maintenance of alertness, [48][49][50][51] and a recent lesion-symptommapping study in stroke patients reported that alerting effects, including RT for auditory warning cues, were significantly affected by lesions in the rAIC. 6Together, these results provide convergent evidence for the causal role of the rAIC in alertness and attention modulation.
Several behavioral intervention studies have attempted to modulate attention networks.However, meditation training 52 and working memory training in ADHD 53 did not reveal expected improvements in the alerting effect.Combining behavioral interventions with rAIC rt-fMRI neurofeedback may provide a more powerful approach for improving alertness and remediating diminished attentional abilities in a variety of disorders including schizophrenia, 54 mild cognitive impairment, 55 following chemotherapy, 56 and ADHD. 57Thus, our result is of particular importance for advancing knowledge about the neurophysiological underpinning of attention and suggests an alternate way to improve alertness.

Limitations of the study
One limitation of this study was the relative timing of the behavioral sessions.While the pre rt-fMRI behavioral tests were done at the beginning of the first session, the post-training behavioral tests were performed just after the second rt-fMRI neurofeedback session, meaning after having been $1.5 h in the MRI scanner.Since many participants reported that rt-fMRI neurofeedback training was tiring, our post-training behavioral effects might have been affected by participants' fatigue.Also, due to the single-blind design we cannot fully rule out a possible experimenter bias.Furthermore, although the sample size in our experimental rAIC and control V1 groups were higher than in other recent fMRI neurofeedback studies, findings from the smaller second mental-rehearsal control group should be interpreted with caution.Because this group did not show any effects right after training, we did not test them for long-term effects during follow-up.Finally, although the effects of alertness on the ANT task were specific to the rAIC neurofeedback group, we observed a general improvement for all groups in the two other components (orienting and executive control) of this task over sessions.In addition, rAIC neurofeedback did not lead to higher performance in Go/NoGo task performance compared to the V1 and mental-rehearsal groups.This observation was surprising, given the documented role of the rAIC in cognitive control processes associated with response inhibition.Further studies are required to address these limitations and clarify the behavioral implications of rAIC neurofeedback regulation.

Conclusion
Our study provides evidence that healthy individuals can effectively self-regulate their rAIC activity using real-time fMRI neurofeedback training, even without explicit guidance, such as instructions on emotion regulation.Distinctly, participants receiving rAIC-targeted feedback exhibited significant and sustained increases in rAIC activity, surpassing the outcomes observed in two control groups.After neurofeedback training, participants showed long-lasting changes in attention-related alertness indicating a critical role of the rAIC in attentional processes.Our findings indicate that neurofeedback training focusing on rAIC upregulation can present a promising, non-invasive, and drug-free method for augmenting certain aspects of attentional capacities.ethnicity were not assessed.All participants gave written informed consent before taking part in the study.This study is part of a large singleblind, placebo-controlled study which was approved by the local ethics committee of the Canton of Zurich in Switzerland (2017-00483) and registered at clinicalTrials.gov(https://clinicaltrials.gov, NCT04643340).Another part of this registered study was to examine the neuronal sources of both supraliminal and subliminal perception during a Go/NoGo task (on which we based our power analysis on) and to examine the impact of neurofeedback on supraliminal and subliminal perception and brain function (results will be reported elsewhere).

METHOD DETAILS Experimental overview
Recordings were performed from March 2020 until November 2021.The experimental protocol consisted of four sessions for the rAIC and V1 groups and three sessions for the mental-rehearsal control group (see Figure 1 for a depiction of the experimental protocol).The sessions were conducted on different days, the first three sessions within one week and the fourth session three months after the third session.All sessions for all participants were conducted at the MRI center of the psychiatric University Hospital in Zurich.During the first session, participants were asked to fill out the Hospital Anxiety and Depression (HADS) 35 and the Baratt impulsiveness scale 11 (BIS-11) 36 questionnaires.Furthermore, they performed a cognitive test battery containing tests on sustained attention, cognitive control, visuo-spatial short-term working memory, and visual perception tests (see paragraph attention test battery).After this, a resting-state sequence was recorded and participants performed four runs of the Go/NoGo task in the MRI scanner with simultaneous EEG-fMRI recording.The data presented here does not include any EEG-fMRI data and the protocol and analysis of this part of the study will therefore be presented elsewhere.During the second and third session of the experiment, rt-fMRI neurofeedback training was conducted.In order to evaluate the immediate effects of the neurofeedback training, participants were asked to perform the same test battery as during the first session, just after the rt-fMRI neurofeedback on the third session (post-training).Participants of the rAIC and V1 group took part in a three-month follow-up session where they underwent the same cognitive test battery and EEG-fMRI recordings.

MRI imaging parameters
The MRI images were acquired on a 3 Tesla MRI scanner (Philips Achieva, upgraded to dStream platform), equipped with a 32-channel receive head coil.Functional images for rt-fMRI neurofeedback were acquired with a T2*-weighted gradient-echo-planar sequence with a repetition time

Neurofeedback protocol
Participants in the rAIC and V1 groups underwent two sessions of rt-fMRI neurofeedback training on two different days within a week and during similar daytime (8 am to 1 pm).On each day, participants performed five neurofeedback runs leading to a total number of ten training runs across all sessions.Additionally, in order to assess participants' ability to regulate region-of-interest (ROI) activity in the absence of feedback, so-called baseline and transfer runs were conducted at the beginning and at the end of each session, respectively.Before neurofeedback runs, anatomical images and resting-state fMRI were acquired.During the resting-state sequence, participants were asked to fixate on a central white cross, presented on a grey screen.Neurofeedback runs consisted of five blocks composed of a 20 s (10 TRs) baseline condition, 40 s (20 TRs) regulation condition, and 4 s (2 TRs) feedback presentation (see Figure 2).During the baseline condition, a black downward arrow was displayed, and participants were asked to count backward from 100 in steps of two.This is a common procedure in rt-fMRI neurofeedback studies to assure that participants maintain stable baseline activity. 18,59uring the regulation condition, a black upwards arrow was presented.Here, participants were instructed to increase their brain activity using any mental strategy they thought might work.Participants were told that the scale of the subsequently presented feedback reflected how well they upregulated brain activity and that they should try to make the scale rise as high as possible.Participants were free to choose any mental strategy they want.However, we stressed that they should not use simple counting since the feedback was calculated as the difference in brain activation between regulation and baseline block.In addition, participants were asked to not change their strategy within a run, however, they were free to change or adapt their strategy between runs.After each run, participants were asked to report the strategy they used.Furthermore, we asked them to rate the success of the used strategy to control the feedback on a scale from 1 (very bad) to 5 (very good).Visual feedback was presented in form of a thermometer icon with the temperature scale representing the difference in ROI activity between the previous regulation and baseline block (= average feedback value).The thermometer scale had ten positive and one zero levels and was color-coded in blue (level 0 to 2), violet (3 to 6), and red (7 to 10).In addition, below the thermometer icon, an integer ranging from 0 to 50 (thermometer level multiplied by 5) was presented to indicate the temperature reading numerically.In order to help participants to decide whether they should keep or change the strategy between runs, at the end of each neurofeedback run we presented them with their average feedback value (transposed to average points, see above) for the corresponding run.
During baseline and transfer runs, a white blank was presented instead of a thermometer icon.This difference aside, these runs were identical to the neurofeedback runs.These runs were included to assess whether participants showed a transfer effect meaning if they were able to maintain the ability to regulate ROI activation in the absence of feedback. 19,60During transfer runs, participants were asked to use the strategy that overall worked best for them.The mental-rehearsal control group performed 14 runs identical to the baseline/transfer run.Participants in this group were told that they participated in a rt-fMRI neurofeedback study, and they received the same instructions regarding the baseline and regulation condition, however, we did not inform them about the existence of the other two groups or a feedback condition.For each neurofeedback session, we assessed how motivated and attentive participants were during the training (self-report Likert scale from one to five).Additionally, after the second neurofeedback session, we asked the participants in the mental-rehearsal control group if they believed that they participated in a real rt-fMRI neurofeedback training.Each neurofeedback training lasts approximately 60 minutes.The whole session with instructions, pre-recordings and rt-fMRI neurofeedback training lasts around 90 minutes.

Real-time fMRI setup and feedback calculation
At the beginning of each session, MNI (Montreal Neurological Institute)-based ROI templates (taken from the Willard functional ROIs atlas (http://findlab.stanford.edu/functional_ROIs.html), for details see Figure S1) were transformed into the participant's native space (T1weighted structural scan) using Statistical Parametric Mapping 12 (SPM12; Wellcome Trust Centre for Neuroimaging London, United Kingdom) and custom-made MATLAB scripts.This procedure ensured that the same ROIs were targeted across the two different training days.During neurofeedback and baseline/transfer runs, acquired and reconstructed functional data was transferred from the MRI PC to a separate stimulation PC where the data was preprocessed and analyzed online using OpenNFT. 61Preprocessing included realignment, reslicing, and spatial smoothing with an isotropic Gaussian kernel with a 5-mm full width at half maximum (FWHM).Furthermore, the time course from the ROI was extracted and signal drift, spikes, and high-frequency noise were removed.Finally, the feedback signal was calculated as the difference in percent signal change between regulation and baseline condition for each block separately.The neurofeedback stimuli and feedback were presented to the participants through MR-compatible video goggles (Resonance Technology Inc., USA) using a custommade script in PsychoPy2. 62

Attention test battery
In order to evaluate whether there were effects of the rt-fMRI neurofeedback training on attention, participants performed the ANT (implemented the PEBL 34 ) before and after the rt-fMRI neurofeedback training.The ANT is a cognitive flanker task with different cues and flanker conditions which allows assessing three components of attention: orienting, alerting and executive control.Each trial starts with a fixation cross which is presented for the whole trial.Participants are instructed to report the direction (left or right) of the target.The target is a central arrow pointing leftward or rightward and can be surrounded by two flankers on each side and presented either above or below the fixation cross.The flankers are either congruent (same direction as the target) or incongruent (point in the opposite direction of the target), in the neutral condition there are no flankers.Before the target occurs, there is either no cue or one of three types of warning cues.In the centre cue condition, the cue is presented on the fixation cross, in the double cue condition the warning cues are above and below the central fixation and in the spatial cue condition the cue is either above or below the cross indicating where the target will be presented (100% valid).The no cue and centre cue condition are control conditions, the double cue condition measures alerting, and the spatial cue orienting.Executive control is measured by the incongruent vs. congruent flanker condition.The ANT served as our primary outcome measure because it requires several mental processes ranging from the implementation of attention and cognitive control to the resolution of conflict, which may be all dependent on rAIC activity.

Other cognitive measures
We also examined secondary cognitive measures which assess different aspects of attention and cognitive control such as task-switching (Switcher) and location memory-span (Corsi block-tapping test). 63Furthermore, inspired by the finding that the rAIC seems to be involved in early unconscious cognitive control implementation 16 we also administered a Go/NoGo task with supraliminal and subliminal Go and NoGo cues.We used an adapted version of the Go/NoGo task used in Van Gaal et al. 16 Lastly, to test for possible effects of V1 rt-fMRI neurofeedback training on visual sensitivity we used an orientation discrimination task.During the task, participants had to indicate the orientation of Gabor patches (right or left), which illustrated a circular sinewave raster that had a gradual Gaussian blur edge.

QUANTIFICATION AND STATISTICAL ANALYSIS FMRI analysis
All functional MRI images were analyzed using MATLAB R2020a and SPM12.Preprocessing for each rt-fMRI neurofeedback run, included slice-time correction, realignment to the first scan of the session, co-registration of the functional to the anatomical image, segmentation, normalization into MNI space and spatial smoothing with a Gaussian kernel of 8-mm FWHM.
For the first level GLM analysis, the baseline, regulation, and feedback periods were modelled as boxcar functions and convolved with the canonical hemodynamic function of SPM.To get equal condition length and because OpenNFT is only considering the last six TRs for feedback calculation, 61 we considered the last 10 TRs for the regulation condition and the entire 10 TRs for the baseline condition.Furthermore, six motion parameters were included as regressors of no interest.For each run the contrast 'regulation > baseline' was computed.
For the ROI analysis, we extracted for each participant from each ROI (rAIC, right and left V1) the average values from contrast images (regulation > baseline) of the first-level analysis of each run, using custom made MATLAB scripts.The overall regulation success was defined before data collection as the difference between the last transfer run and the first baseline run. 33We will refer to this effect as the transfer , to examine ROI activity regulation success over the whole rt-fMRI neurofeedback training, linear regression of the extracted ROI activity over training runs was calculated.The rAIC group increased rAIC activity over the course of rt-fMRI neurofeedback runs (b = 0.04, R 2 = 0.03, F (1, 218) = 7.95, p = 0.005).While participants across V1 groups did not show a transfer effect for rAIC activity (F (1,23) = 2.33, p = 0.42), participants in the right V1 group showed a trend for an increase in rAIC activity across rt-fMRI neurofeedback runs (b = 0.03, R 2 = 0.03, F (1,238) = 3.71, p = 0.056).When combining the two V1 groups a significant increase in rAIC activity over the course of rt-fMRI neurofeedback runs was observed (b = 0.03, R 2 = 0.02, F (1,238) = 6.16, p = 0.013).The mental-rehearsal group (b = À0.01,R 2 = 0.003, F (1, 98) = 0.37, p = 0.54) and the left V1 group (b = 0.03, R 2 = 0.022, F (1, 118) = 2.73, p = 0.1) did not show a linear learning effect (see Figures S2 and S3 in the supplemental information).For activity in right V1, the linear regression demonstrated a significant increase over the course of rt-fMRI neurofeedback runs for both the right V1 (b = 0.03, R 2 = 0.04, F (1, 118) = 4.37, p = 0.03) and the left V1 (b = 0.05, R 2 = 0.05, F (1, 118) = 6.94, p = 0.009) groups.Regarding activity extracted from left V1, only participants in the left V1 group increased activity over the course of neurofeedback runs (b = 0.05, R 2 = 0.06, F (1, 118) = 8.45, p = 0.004).Participants in the right V1 group did not significantly increase activity over runs in the left V1 (p = 0.09).

Figure 1 .
Figure 1.Study overview Participants in the rAIC and V1 group took part in four sessions.The first session included questionnaires, behavioral tests and a Go/NoGo task while EEG and fMRI were recorded simultaneously.Session two and three were rt-fMRI neurofeedback training sessions.For details about ROI placement see also Figure S1.After session three, participants again performed behavioral tests.Three months after neurofeedback training, participants repeated the procedure of the first study session.Participants in the mental-rehearsal group followed the same procedure but without the three months follow-up session and we did not record EEG-fMRI during session one.

Figure 2 .
Figure 2. Schematic illustration of one session of rt-fMRI neurofeedback training for each group Participants in rAIC and V1 group performed at the beginning and at the end of each session runs without feedback (baseline/transfer). Between baseline and transfer run, they did five runs where they received neurofeedback training.Participants in the mental-rehearsal group conducted per session seven runs without feedback.For each participant, an anatomical MRI and a resting-state fMRI was recorded before the neurofeedback training.

Figure 3 .
Figure 3. Activation level within rAIC before and after neurofeedback training Activity difference in contrast values (regulation > baseline) between the first baseline run and the last transfer run.Asterisks indicate significant paired t-test results: *p < 0.05, **p < 0.01, ***p < 0.001.For rAIC activity over all runs see also Figures S2 and S3.

Figure 4 .
Figure 4. Alerting effects Participants in the rAIC group showed an increased alerting effect right after rt-fMRI neurofeedback training (post-training) compared to before (**p < 0.01).This effect was also evident three months later (*p < 0.05) (upper panel A).The lower panel B shows that this effect was driven by faster reaction times for the double cue condition for the examined time intervals.Post-hoc test showed that participants in the rAIC group responded significantly faster to double cue trials immediately after training compared to before rt-fMRI neurofeedback (***p < 0.001).RTs remained shorter during the three months FU compared to before neurofeedback training (***p < 0.001).RTs used for this figure were cleaned for outliers (see STAR Methods part).However, additional analysis was calculated without the values marked as outliers in the boxplots and can be found in the SI Figure S5 (results remained unchanged).