Connectivity‐Based Real‐Time Functional Magnetic Resonance Imaging Neurofeedback in Nicotine Users: Mechanistic and Clinical Effects of Regulating a Meta‐Analytically Defined Target Network in a Double‐Blind Controlled Trial Authors: Arezoo Taebi, Klaus Mathiak, Benjamin Becker, Greta Kristin Klug, Jana Zweerings Published: 2024 ABSTRACT ABSTRACTOne of the fundamental questions in real‐time functional magnetic resonance imaging neurofeedback (rt‐fMRI NF) investigations is the definition of a suitable neural target for training. Previously, we applied a meta‐analytical approach to define a network‐level target for connectivity‐based rt‐fMRI NF in substance use disorders. The analysis yielded consistent connectivity alterations between the insula and anterior cingulate cortex (ACC) as well as the dorsal striatum and the ACC. In the current investigation, we addressed the feasibility of regulating this network and its functional relevance using connectivity‐based neurofeedback. In a double‐blind, sham‐controlled design, 60 nicotine users were randomly assigned to the experimental or sham control group for one NF training session. The preregistered primary outcome was defined as improved inhibitory control performance after regulation of the target network compared to sham control. Secondary outcomes were (1) neurofeedback‐specific changes in functional connectivity of the target network; (2) changes in smoking behavior and impulsivity measures; and (3) changes in resting‐state connectivity profiles. Our results indicated no differences in behavioral measures after receiving feedback from the target network compared to the sham feedback. Target network connectivity was increased during regulation blocks compared to rest blocks, however, the experimental and sham groups could regulate to a similar degree. Accordingly, the observed activation patterns may be related to the mental strategies used during regulation attempts irrespective of the group assignment. We discuss several crucial factors regarding the efficacy of a single‐session connectivity‐based neurofeedback for the target network. This includes high fluctuation in the connectivity values of the target network that may impact controllability of the signal. To our knowledge, this investigation is the first randomized, double‐blind controlled real‐time fMRI study in nicotine users. This raises the question of whether previously observed effects in nicotine users are specific to the neurofeedback signal or reflect more general self‐regulation attempts.Keywords: functional connectivity, functional network, inhibitory control, neurofeedback, nicotine dependence, real‐time fMRI Introduction Real‐time functional magnetic resonance imaging neurofeedback (rt‐fMRI NF) aims to train individuals to modulate their brain response. Region‐specific neurofeedback training has been shown to yield effects in neural activation during and after training as well as for behavioral and clinical measures (Keller et al. 2021; Yao et al. 2016; Zweerings et al. 2019). Connectivity‐based real‐time fMRI neurofeedback offers simultaneous modulation of several brain areas by targeting specific brain circuits connecting those regions (Morgenroth et al. 2020; Yamashita et al. 2017; Zhao et al. 2019). This is a promising approach for substance use disorders (SUD), which are characterized by alterations in various brain networks including the executive, salience, and reward networks (Koob and Volkow 2016; Zilverstand et al. 2018). However, widespread network‐level alterations impose challenges in defining a suitable target for self‐regulation training. Meta‐analytic investigations may address this problem by identifying network‐level targets for connectivity‐based rt‐fMRI NF in SUD. Accordingly, such approach can provide targets for self‐regulation training that show consistent connectivity alterations across studies. Ultimately, NF allows us to directly investigate the behavioral relevance of the identified network. Previous research implementing rt‐fMRI NF in SUD mainly focused on the modulation of single brain regions (for review Martz et al. 2020). For instance, studies provided participants with nicotine dependency with feedback from the anterior cingulate cortex (ACC) during cue‐induced craving trials (Canterberry et al. 2013; Hanlon et al. 2013; Hartwell et al. 2016; Li et al. 2013). Results indicated that participants successfully reduced ACC activation. In addition, lower craving scores were reported after training. However, due to the lack of a control group as well as the small sample size (N = 13, 12, 9), these findings must be interpreted with caution. Notably, Kim et al. (2015) investigated the feasibility of modulating several brain areas involved in craving using rt‐fMRI connectivity approach in individuals with nicotine dependency. Participants were assigned to either activation‐based NF from anterior regions or a combination of activation‐based (anterior regions) and connectivity‐based NF between anterior and posterior areas. The authors reported that both groups were able to increase neural activation in anterior and posterior regions in the second session compared to the first one with better performance observed in the connectivity‐added group. While these results are promising, generalizability is limited due to a small number of participants (n = 14, randomly allocated to one of the two conditions) and the lack of control group. In addition, the optimal target for connectivity informed NF remains elusive. To identify a suitable network‐level target for a rt‐fMRI NF study in SUD, we conducted a multi‐level kernel density analysis to identify common alterations in resting‐state functional connectivity patterns across different SUDs (Taebi et al. 2022). Our meta‐analysis consistently indicated decreased connectivity between the ACC and the insula and dorsal striatum (DS). Addiction models suggest the involvement of these regions in craving, compulsive drug‐seeking behavior, and self‐control processes (Everitt and Robbins 2016; Goldstein and Volkow 2011; Noël, Brevers, and Bechara 2013). Alterations in connectivity between these regions have been associated with impaired inhibitory control performance and severity of nicotine dependency (Bi et al. 2017; Feng et al. 2016; Yuan et al. 2016). Moreover, altered connectivity between the insula and cognitive control regions has been related to relapse vulnerability in smokers (Addicott et al. 2015; Janes et al. 2010), highlighting the importance of the identified pathway in nicotine dependency. Given the potential behavioral relevance of connectivity between the ACC and insula as well as ACC and DS for regulatory control, we selected these regions as our target network. To investigate the clinical and behavioral significance of the identified network, we chose inhibitory control performance as our primary behavioral outcome and nicotine dependency as our secondary outcome. This selection was made to gain mechanistic insights into how modulating this network might influence key aspects of smoking behavior. In the current pre‐registered connectivity‐based rt‐fMRI NF study, 60 nicotine users were randomly assigned to the experimental (target network) or control sham group (feedback from the previous subject in the active NF condition) in a double‐blind, controlled design. The meta‐analytically defined target network included the weighted functional connectivity between the ACC and insula or DS, respectively. Our primary registered outcome was an enhancement in inhibitory control as assessed by performance in a Stop Signal Task (SST) following the regulation of the target network in the experimental group compared to the sham group. As registered secondary outcomes, we expected (1) neurofeedback‐specific changes in functional connectivity of the target network; (2) changes in nicotine consumption and impulsivity measures between baseline and follow‐up assessments; and (3) group‐specific transient changes in the connectivity strength of our target network during rest before and after the NF training session. Materials and Methods Design In a randomized, double‐blind controlled design, we assigned participants to either the experimental (exp) or sham control group. The study comprised one training day and a follow‐up assessment approximately 1 month later. The training day included behavioral and clinical assessments as well as an anatomical scan, a neurofeedback session with three runs of training, and resting‐state measurements before and after the NF training (Figure 1A). SST was conducted before and after NF training. Prior to the NF training, participants were asked to smoke a cigarette to reduce craving effects. In the follow‐up interview, changes in behavioral and clinical domains were assessed. The data was first unblinded after the collection of half of the sample, however, the experimenters were not aware of the preliminary results. The study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of the University Hospital RWTH Aachen (EK 319/19). All subjects provided written consent. The study was preregistered in the German Clinical Trials Register (DRKS00027056). (A) Study design. (B) The neurofeedback paradigm comprised five cycles of rest blocks cued by a cloud image (18 s) and regulation blocks cued by a brain (92.4 s). The regulation phase comprised 22.8 s of regulation without feedback followed by the presentation of 8 feedback values. The first two windows at the beginning of each run were used for baseline calculation of connectivity within the target network. (C) The target network seeds comprised anterior cingulate cortex [(4, 32, 23), 10 mm], insula [(38, 14, −8), 10 mm], and dorsal striatum [(21, 5, 4), 5 mm]. Participants Sixty subjects (30 exp., 30 sham) were included in our study. All participants met the following inclusion criteria: (1) smoking at least 10 cigarettes per day for the last year; (2) sufficient knowledge of the German language; (3) right‐handedness, and (4) age between 18 and 75 years. The exclusion criteria were severe psychiatric diseases, MR contraindications, and pregnancy. Behavioral and Clinical Assessments To investigate training‐induced changes in inhibitory control performance, participants were asked to complete the SST (Band, van der Molen, and Logan 2003) before and after neurofeedback training. Participants were instructed to respond as fast and accurately as possible during Go trials (presented by an arrow pointing to the left or right). In 25% of trials (Stop trials), however, they were signaled to inhibit their response which had been already initiated. The Stop trials were indicated by presenting a red circle around the Go stimuli after a variable delay. The variation in the stop‐signal delay (SSD) was determined based on the participant's performance on the last Stop trial. Therefore, it decreased after the participant failed to inhibit the response and increased following a successful response inhibition in the previous Stop trial. Inhibitory control performance has been estimated by the stop signal reaction time (SSRT), that is, the average time of successful response inhibition. The SSRT was calculated by first choosing the response time (RT) corresponding to the nth percentile of the accurate Go trials, in which n equals the percent of unsuccessful stop trials. Subsequently, the average SSD was subtracted from the selected RT. Shorter resulting SSRTs reflect better inhibitory control performance. To ensure the reliability of SSRT estimates, we applied predetermined lenient criteria (Congdon et al. 2012) to determine and subsequently exclude outliers. The outlier criteria were defined as a percentage of successful Stop trials below 25 or above 75%, a response rate in Go trials below 60%, a percentage of inaccurate Go trials above 10%, and an SSRT below 50 ms. Accordingly, only data of participants that were able to accurately follow the task instructions were entered in subsequent analysis. In addition to inhibitory control, we assessed trait impulsivity using the Barratt Impulsiveness Scale (BIS‐11) (Patton, Stanford, and Barratt 1995) and UPPS Impulsive Behavior Scale (S‐UPPS‐P) (Cyders et al. 2014). Nicotine consumption was measured by the Fagerstrom Test for Nicotine Dependence (FTND) (Heatherton et al. 1991). In addition to subjective measures, the breath CO level was assessed using the Micro+ Smokerlyzer by Bedfont to obtain a quantifiable, objective measurement for individual smoking behavior prior to neurofeedback training. Further smoking‐related information such as the average number of cigarettes per day, duration of use, and intention to quit was collected in a customized questionnaire. We applied mixed‐design ANOVA to investigate potential interaction effects between group and time point for the SST, impulsivity measures, and FTND. MRI Data Acquisition Data was acquired via a 3 Tesla scanner (Magnetom TRIO, Siemens Medical Systems, Erlangen, Germany) with a 20‐channel head coil. The task data was obtained using a multiband echo‐planar imaging (EPI) sequence with the following parameters: repetition time (TR): 1200 ms, echo time (TE): 32 ms, flip angle: 67°, slice thickness: 2.2 mm, space between slices: 0.55 mm, voxel size: 2.2 mm, matrix size: 96 × 96, number of interleaved slices: 48, total number of volumes: 500. For the resting‐state scans, an EPI sequence (TR: 2000 ms, TE: 28 ms, flip angle: 77°, slice thickness: 3 mm, space between slices: 0.75 mm, voxel size: 3 mm, matrix size: 64 × 64, number of interleaved slices: 34, total number of volumes: 250) was used. In addition to the functional scans, anatomical images were acquired with a magnetization‐prepared rapid acquisition gradient echo (MPRAGE) sequence (TR: 2000 ms, TE: 3.03 ms, inversion time (TI): 90 ms, flip angle: 9°, voxel size: 1 mm, matrix size: 256 × 256, number of slices: 176). Neurofeedback Training All participants were trained to increase the functional connectivity of the target network. The training consisted of three runs, each including five cycles of rest and regulation blocks (Figure 1B). During rest phases (18 s each), participants were presented with a ‘cloud’ picture and instructed to count backward (100, 99, …). Following each rest block, a regulation phase (92.4 s each) started with the display of a ‘brain’ picture. Participants were instructed to use mental strategies for regulation and received the first feedback value after 22.8 s (regulation without feedback). After the specified duration, numerical feedback ranging from 1 to 100 was provided and updated every eight TRs. Each regulation block comprised 8 FB values resulting in a total of 40 FB values per run. The feedback reflected the scaled weighted average of the absolute value of Pearson's correlation (r) between our network nodes (ACC‐insula, ACC‐DS) (i.e., target connectivity (TC)) with a weighting factor of two for ACC‐insula and one for ACC‐DS connectivity for the contrast regulation minus baseline (Formula 1 and 2). A weighted average approach was chosen to adjust for the smaller volume size of the DS seed, potentially leading to a lower signal‐to‐noise ratio in the DS compared to the insula seed. By assigning a higher weight to the ACC‐insula connectivity in our feedback calculation, we aimed to compensate for the difference in seeds' volume size while still incorporating the broader network including the DS. A sliding window approach with a length of 15 volumes was used for FB calculation. The baseline was estimated at the beginning of each run by averaging connectivity in the target network across two windows. Fifty represented no regulation, while FB values above 50 indicated up‐regulation. All participants were asked to maximize the displayed feedback values. No explicit regulation strategies were provided and participants were encouraged to find their own helpful strategies. However, exemplary strategies prior to the training served as potential starting points, including the following: (1) thinking about the benefits of quitting smoking or the negative consequences of long‐term smoking, (2) imagining performing your favorite hobby, or (3) recalling positive autobiographical memories. This approach was chosen to facilitate initial trials for participants. Additionally, we explicitly instructed all participants to maintain stable breathing patterns throughout the experiment. Participants in the sham group received the feedback values from the last participant of the experimental group. Each training run lasted 10 min. We assessed the perceived feeling of control over regulating brain signal on a scale from 0 to 10 in a structured interview after the NF training. Processing of real‐time data was performed using Turbo‐Brain Voyager 4.2 (Brain Innovation, Maastricht, The Netherlands). Real‐time preprocessing steps included skull stripping, spatial normalization to the MNI space, Gaussian spatial smoothing with 4 mm FWHM kernel, and motion correction. Prior to the neurofeedback training, an anatomical scan was performed to co‐register the predefined anatomical masks of the target network seeds. The target network nodes were defined as spheres with the following center coordinates (x, y, z) in MNI space and radius: ACC [(4, 32, 23), 10 mm]; insula [(38, 14, −8), 10 mm]; DS [(21, 5, 4), 5 mm] (Figure 1C). We defined the coordinates based on the regions with altered functional connectivity profiles in our meta‐analysis and their assumed relevance for smoking behavior (Taebi et al. 2022). In parallel to our procedure in the meta‐analysis, the coordinates of the insula and DS were defined as the center of mass of the right insula and right DS based on the Harvard atlas masks. The ACC coordinate was defined as the center of mass of the overlapping areas showing hypoconnectivity with the DS and insula regions. A custom script running under Matlab R2018a (The MathWorks Inc. Natick, MA) was used to calculate and display feedback values. Blinding was achieved by a custom script randomly generating an equal number of group assignments in blocks of six. All real‐time data including the real feedback scores for the sham group and connectivity values for all windows were automatically saved. Offline Preprocessing The images were preprocessed with fMRIPrep (Esteban et al. 2019). We discarded the first six volumes of the task data to account for the T1‐saturation effect. The pipeline included skull stripping, brain tissue segmentation, brain surface reconstruction, boundary‐based registration to anatomical space, non‐linear spatial normalization to the MNI space, and motion correction. For resting state data only, the first five volumes were skipped and slice‐timing correction was applied. Additionally, we applied temporal filtering (high‐pass filter: 0.01 Hz for the task and a band‐pass filter: 0.01–0.08 Hz for the resting‐state data) and a spatial smoothing of 6 mm FWHM Gaussian kernel. The frame‐wise displacement (FD) parameter (Power et al. 2012) was extracted as the sum of absolute relative motions. Volumes with FD values above 0.8 mm were annotated as motion outliers. Data of participants with runs exceeding a threshold of 20% of volumes defined as motion outliers were excluded from the analysis. Functional Connectivity During Neurofeedback Training To evaluate participants' performance during NF training, we used the feedback scores obtained from the real‐time analysis (see section 2.5). For the sham group, the actual scaled weighted connectivity values of the participants were used for further analysis. To examine learning, we calculated the average score for each run (comprising 40 FB values per run) and implemented a mixed‐design ANOVA with the within‐group factor run (1–3) and the between‐group factor group (sham versus experimental). Next, to assess the stability of the connectivity values, we examined the fluctuation of connectivity values within the target network during the regulation blocks. Connectivity values were calculated in windows consisting of 15 volumes in the real‐time analysis. Accordingly, we extracted 40 connectivity values for ACC‐insula and ACC‐DS per run in accordance with real‐time feedback score calculation. Fluctuation was defined as dividing the average of connectivity values by their respective standard deviation for each subject. A Psychophysiological Interaction (PPI) analysis (Friston et al. 1997) investigated how the functional connectivity in the target network changed in response to NF task demand to provide insights into network dynamics. To this end, we first extracted time‐series from the ACC seed for all subjects and runs. Next, we built a General Linear Model (GLM) at the subject level with the task regressors (rest, regulation without feedback, regulation with feedback), the ACC time‐series regressor, and the interaction term between the seed time‐series and task regressors using FSL (FMRIB, Oxford, UK). We regressed out 24 head‐motion estimates as six realignment parameters calculated in the motion correction preprocessing step and their temporal derivatives and quadratics terms (Friston et al. 1996), mean signal in white matter (WM) and cerebrospinal fluid (CSF) (Fox et al. 2005) as well as motion outliers calculated by FD (Power et al. 2012). Additionally, we conducted a supplementary PPI analysis incorporating global signal regression as a control measure, aiming to remove potential links between physiological artifacts and connectivity measures (Weiss et al. 2020). The contrast of interest was based on the interaction terms for the contrasts (1) regulation without feedback versus rest and (2) regulation with feedback versus rest. We then extracted z‐transformed t‐test values for the insula and DS seeds for each run and subject. A mixed‐design ANOVA investigated the interaction effect between group and time to shed light on differences in learning performance. Furthermore, as a supplementary analysis, z‐transformed t‐test values for insula seed were tested for each run at the group level using two‐sided t‐tests. We used the false discovery rate (FDR) to correct for multiple comparisons (Benjamini and Hochberg 1995). Resting‐State Functional Connectivity Resting‐state fMRI was conducted before and after the NF training to investigate transient connectivity changes in the target network. Participants were instructed to minimize movement and look at a fixation cross presented at the center of the screen, and let their minds wander. For the analysis, we extracted time‐series from the target network seeds and calculated Fisher's z‐transformed Pearson's correlation between ACC‐insula and ACC‐DS. Parameters of no interest were included in parallel to the analysis of functional connectivity during neurofeedback training (see 2.7). We used a mixed‐design ANOVA to identify group‐specific changes in connectivity patterns after the NF training. Results Demographic Characteristics Groups did not differ in age (exp: 33 ± 14.603, ranged from 20 to 75; sham: 28.866 ± 8.976, ranged from 20 to 54; t(58) = 1.298, p = 0.199) and gender (exp: 11 female/19 male; sham: 10 female/20 male). Behavioral and Clinical Assessment We excluded six subjects from the experimental and one subject from the sham group based on the outlier criteria defined for the SST. Results showed, that the experimental and sham group did not differ in SSRT prior to NF training (Table 1). No significant main and interaction effects emerged in the ANOVA between group (exp vs. sham) and time point (pre‐NF vs. post‐NF) in SST performance (F Group(51,1) = 0.21, p = 0.644, ηp2 = 0.004; F Time(51,1) = 0.11, p = 0.74, ηp2 = 0.002; F GxT(51,1) = 1.96, p = 0.167, ηp2 = 0.037; Figure 2A). The results did not change after including the whole sample (F Group(58,1) = 1.39, p = 0.242, ηp2 = 0.023; F Time(58,1) = 0.47, p = 0.492, ηp2 = 0.008; F GxT(58,1) = 1.51, p = 0.223, ηp2 = 0.025). TABLE 1Behavioral and clinical assessments for the experimental and sham group.Baseline characteristicsExperimental groupSham groupGroup differenceSSRT prior to NF226.953 (58.616)211.301 (52.428) t(51) = 1.005, p = 0.319FTND score3.30 (2.051)3.50 (1.668) t(58) = 0.407, p = 0.685CO measure in ppm13.17 (7.587)14.72 (9.691) t(58) = 0.813, p = 0.42Number of cigarettes per day14.464 (5.375)13.90 (3.228) t(58) = 0.655, p = 0.514Duration of use12.384 (10.944)10.862 (10.016) t(58) = 0.528, p = 0.599Intention to quite in percentage62.1%70% χ2 = 0.135, p = 0.712Abbreviations: FTND, fagerstrom test for nicotine dependence; NF, neurofeedback; SSRT, stop signal reaction time. Behavioral and clinical assessments for the experimental and sham group. Abbreviations: FTND, fagerstrom test for nicotine dependence; NF, neurofeedback; SSRT, stop signal reaction time. (A) Stop signal reaction time for pre‐ and post‐neurofeedback (NF) measurements in the experimental and sham groups. (B) Fagerstrom Test for Nicotine Dependence (FTND) scores at baseline and follow‐up for the experimental and sham groups. Error bars represent 95% confidence interval. Additionally, no significant group differences in smoking‐related measures at baseline emerged, including FTND score, CO measure, average number of cigarettes per day, duration of use, and percentage of individuals with the intention to quit (Table 1). Five participants were unavailable for the follow‐up interview. Accordingly, 55 participants (N exp = 27 and N sham = 28) were included in the respective analysis. On average, the follow‐up interview was conducted 37 days (±12) after the NF training. Both groups completed the follow‐up interview at comparable time points after training (exp 36 days (±11); sham 38 days (±13); t(55) = 0.594, p = 0.554). No significant main and interaction effects of group and time point (baseline vs. follow‐up) emerged for FTND score (F Group (53,1) = 0.003, p = 0.953, ηp2 = 0.00006; F Time(53,1) = 2.128, p = 0.15, ηp2 = 0.038; F GxT(53,1) = 0.318, p = 0.575, ηp2 = 0.005; Figure 2B) and BIS scores (F Group(53,1) = 0.4, p = 0.529, ηp2 = 0.007; F Time(53,1) = 0.91, p = 0.34, ηp2 = 0.016; F GxT(53,1) = 2.745, p = 0.103, ηp2 = 0.049). UPPS scores revealed significant effect of group (F(52,1) = 4.099, p = 0.048, ηp2 = 0.073). However, the main effect of time and the interaction effect of group and time were not significant (F Time(52,1) = 0.194, p = 0.66, ηp2 = 0.003; F GxT(52,1) = 1.187, p = 0.28, ηp2 = 0.022). Functional Connectivity During Neurofeedback Training Due to a technical issue, data from the last training run was not saved for one participant in the sham group. Moreover, we excluded one subject from the experimental group based on excessive head movement (see section 2.6 for motion outlier's criteria). Accordingly, the final sample size for the real‐time NF data was 58 (N exp = 29 and N sham = 29). No group difference in the learning curve across runs emerged (F(112,2) = 1.263, p = 0.286, ηp2 = 0.022; Figure 3A). Fluctuation in the connectivity values was significantly higher for the ACC‐DS connectivity compared to ACC‐insula connectivity across all participants (ACC‐insula fluctuation: 2.223 ± 0.632; ACC‐DS fluctuation: 1.224 ± 0.5; paired t‐test(57) = 12.964, p < 0.001; Figure 3B). The groups did not differ in the perceived feeling of control over the feedback score (exp: 6.066 ± 2.174; sham: 5.733 ± 2.174; t(58) = 0.583, p = 0.561). (A) Feedback scores across training runs for the experimental and sham groups. (B) Histogram of fluctuation in connectivity between anterior cingulate cortex (ACC) and insula as well as connectivity between ACC and dorsal striatum (DS) across all participants. For the offline PPI analysis, we excluded two subjects from the experimental group, one based on motion outliers and the other due to a signal drop in the structural scan (N exp = 28 and N sham = 30). In line with our real‐time analysis, the PPI analysis revealed no significant interaction effect for the regulation without feedback versus rest contrast (for insula: F(112,2) = 0.013, p = 0.986, ηp2 = 0.0002, for DS: F(112,2) = 0.428, p = 0.652, ηp2 = 0.007) as well as for the regulation with feedback versus rest contrast (for insula: F(112,2) = 0.094, p = 0.909, ηp2 = 0.001, for DS: F(112,2) = 0.141, p = 0.868, ηp2 = 0.002). After implementing global signal regression as a control measure, our results remained consistent with the original findings (regulation without feedback vs. rest contrast: for insula: F(112,2) = 0.404, p = 0.668, ηp2 = 0.007, for DS: F(112,2) = 0.593, p = 0.554, ηp2 = 0.0104; regulation with feedback vs. rest contrast: for insula: F(112,2) = 0.097, p = 0.906, ηp2 = 0.001, for DS: F(112,2) = 0.527, p = 0.591, ηp2 = 0.009). Supplementary analysis showed that participants from both groups increased connectivity between the ACC and insula seed during regulation without feedback blocks (use of mental strategies for regulation only) compared to rest blocks (run1: t 58 = 2.122, p FDR = 0.045, Cohen's d = 0.278; run2: t 58 = 3.935, p FDR = 0.001, Cohen's d = 0.516; run3: t 58 = 3.463, p FDR = 0.003, Cohen's d = 0.454). We also observed increased connectivity between the ACC and insula seed during regulation with feedback blocks compared to rest blocks for run 2 and run 3 (run1: t 58 = 0.415, p FDR = 0.679, Cohen's d = 0.054; run2: t 58 = 2.613, p FDR = 0.022, Cohen's d = 0.343; run3: t 58 = 2.151, p FDR = 0.045, Cohen's d = 0.282) (Figure 4). Psychophysiological interaction (PPI) analysis results with extracted z‐values for insula seed for the contrasts (A) regulation without feedback versus rest and (B) regulation with feedback vs. rest for each run for all subjects as well as both groups separately. The z‐values were extracted from z‐transferred t‐test values for the contrasts of interest. *p < 0.05 after false discovery rate (FDR) correction. Resting‐State Functional Connectivity The final data set for the resting‐state included 59 participants (29 exp., 30 sham). We did not find any significant interaction effect of group and time in our target network (for connectivity between ACC and insula: F(57,1) = 0.030, p = 0.861, ηp2 = 0.0005, ACC and DS: F(57,1) = 0.331, p = 0.566, ηp2 = 0.005) (Figure 5). Resting‐state functional connectivity between (A) anterior cingulate cortex (ACC) and insula, and (B) ACC and dorsal striatum (DS), measured at pre‐neurofeedback (NF) training and post‐NF training in the experimental and sham groups. Error bars represent 95% confidence interval. Discussion In this study, we used a pre‐registered, double‐blind, sham‐controlled design to examine the feasibility of a single session connectivity‐based rt‐fMRI NF targeting a meta‐analytically defined network in nicotine users. The preregistered primary outcome was enhanced inhibitory control performance after the NF training in the experimental compared to the sham control group. In addition, secondary preregistered outcomes were changes in connectivity in the target network during NF and rest as well as changes in nicotine consumption and impulsivity at follow‐up. No post‐training improvement in the inhibitory control task was found irrespective of group. We observed decreased nicotine dependency scores in the follow‐up assessment independent of group assignment, however, the changes were not significant. PPI results showed that across groups, participants increased functional connectivity between ACC and insula. An explorative data analysis indicated significantly higher fluctuations in ACC‐DS connectivity compared to the connectivity between ACC and insula. Taken together, our results raise some critical aspects about the effectiveness of a single session connectivity‐based NF for the targeted network that we will address in the following sections. Importantly, the current investigation was a pre‐registered randomized, controlled, double‐blind trial. Previous studies in the context of nicotine dependency mostly lacked control groups (Martz et al. 2020). This renders a distinction between NF‐specific and non‐specific effects such as beneficiary effects of using cognitive strategies as well as repetition‐related improvements impossible (Ros et al. 2020). Real‐time fMRI NF is perceived as a high‐tech intervention that is likely to induce placebo effects in participants. Therefore, the implementation of adequate control conditions is of considerable importance (Sorger et al. 2019). Our results showed an increase in functional connectivity in both the experimental and the sham control group. Accordingly, in the current investigation, results did not indicate neurofeedback‐specific effects on connectivity in the target network that exceed effects due to non‐specific factors. Our results are in line with a previous connectivity‐based NF investigation targeting the dorsolateral prefrontal‐striatal network in healthy participants (Weiss et al. 2022), indicating no significant difference between the experimental and sham group. Moreover, Mehler et al. (2018) found comparable effects of NF on depression scores after upregulation of process‐specific (i.e., emotion regulation) areas and control regions. The authors concluded that these results may indicate beneficial effects of engaging in mental imagery and experiencing successful self‐regulation per se—independent of condition. However, they did not include a sham control condition to further support this notion. Here, we suggest that cognitive strategies used by our participants during regulation blocks to achieve regulatory control compared to rest blocks more generally may play a role in the observed effects. In addition, the observed increased functional connectivity between ACC and insula might be accounted for by an overlap between our target network and regions involved in cognitive control processes during neurofeedback regulation. Emmert et al. (2016) found that independent of the target region, several brain regions showed increased activation during the regulation phases across rt‐fMRI NF studies. This neurofeedback control network included the anterior insula, striatum, ACC, dorsolateral and ventrolateral prefrontal cortex, temporal–parietal area, and temporal‐occipital junction. Therefore, the involvement of our target network in the cognitive process of self‐regulation might have made it difficult for our participants to utilize the feedback scores to further optimize their strategies and engage the target regions beyond their involvement in general control processes. We also investigated the stability of the connectivity values calculated across windows with a length of 15 volumes. The result revealed a significant difference in fluctuations between the ACC‐insula and the ACC‐DS pathway. Specifically, the connectivity between ACC and DS showed higher changes across windows. The high fluctuation of connectivity values in general, and for ACC‐DS connectivity specifically might significantly impede the controllability of this pathway. Therefore, the optimal window length for connectivity computations for specific target circuits needs to be systematically investigated in future studies. Moreover, despite an interaction effect observed in the SST in our pilot data, the effect did not prevail in the complete trial. This highlights the importance of larger sample sizes in neurofeedback studies. An additional aspect of the connectivity‐based NF that needs systematic investigation is the mental strategies that could be employed for the training. It is necessary to establish links between mental strategies and the functional role of connectivity within circuits. Limitation Our target network was identified based on a binary distinction between SUD and healthy control groups. However, it is an important step in the future to meta‐analytically define target networks with robust links to clinical and behavioral measures. According to FTND scores, our subjects were medium nicotine‐dependent on average. This might limit the generalizability of the findings due to less pronounced dysfunctional brain activation patterns and deviations in behavioral data. Future studies might explore the feasibility of the connectivity approach in a strongly dependent group possibly with greater room for improvement (Haugg et al. 2021). It is also important to consider the potential confound introduced by unequal groups' sample size after excluding more subjects from the experimental group regarding our primary outcome measure, i.e., performance in the SST. In terms of the primary outcome measure, reducing craving through NF training might be more feasible than enhancing inhibitory control. Hanlon et al. (2013) showed that using NF may be more effective in reducing craving than in increasing control over craving. However, the small final sample size (n = 9) limits the generalizability of their findings. While no smoking‐related cues were presented in the present study, it might be more effective to challenge the participants with smoking‐related content during NF training. These aspects have to be investigated systematically in future investigations. Furthermore, fluctuations observed in connectivity values could yield unstable baseline estimations. Future studies might consider alternative approaches to baseline calculation. It is also important to mention that we did not control for the effect of respiratory artifacts in the real‐time feedback calculation. Weiss et al. (2020) found an association between connectivity estimates and respiratory artifacts. Accordingly, we cannot exclude the possibility that online connectivity‐based feedback signals were affected by changing breathing patterns. Last, our study comprised one training session. Repeated training across multiple days might be necessary to yield efficacy for the defined target network. Conclusion This work extends previous NF studies in substance use disorder by integrating a novel approach for the definition of target networks and directly probing the immediate and delayed behavioral relevance for disorder‐relevant constructs (inhibitory control and smoking behavior). Our study did not confirm the preregistered primary outcome of improved inhibitory control performance after one session of NF training. Interestingly, both the experimental and the sham control group showed an increase in functional connectivity in the target network during training. Given the well‐controlled design of our study, comparatively large sample size, and adherence to the preregistered hypotheses, we suggest that single‐session connectivity‐based NF does not yield efficacy for the described target network. Author Contributions The contributions of the authors are as follows: A.T. contributed to the study conceptualization and design, data collection, project administration, statistical analysis, interpretation of the results, writing and revising of the manuscript. K.M. and B.B. were involved in the conception and design, and revision of the manuscript. G.K.K. was involved in data collection, project administration and revision of the manuscript. J.Z. contributed to the research conceptualization and design, project administration, funding acquisition, final approval of statistical analysis, interpretation of the findings, and revision of the manuscript. Conflicts of Interest The authors declare no conflicts of interest. Acknowledgments This work was supported by the START‐Program of the Faculty of Medicine of the RWTH Aachen University; the International Research Training Group (IRTG 2150), The Neuroscience of Modulating Aggression and Impulsivity in Psychopathology of the German Research Foundation (DFG)—269953372/GRK2150; and the Brain Imaging Facility of the Interdisciplinary Center for Clinical Research (IZKF) within the Faculty of Medicine at RWTH Aachen University. The study sponsors were not involved in the study design and in the collection, analysis, and interpretation of data; in the writing of the manuscript; or in the decision to submit the manuscript for publication. Funding: This work was supported by the START‐Program of the Faculty of Medicine of the RWTH Aachen University and the International Research Training Group (IRTG 2150), The Neuroscience of Modulating Aggression and Impulsivity in Psychopathology of the German Research Foundation (DFG) ‐ 269953372/GRK2150. Data Availability Statement The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions. References Associated Data Data Availability Statement The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.