Alcohol use disorder (AUD) is characterized by drinking of alcohol despite negative or harmful consequences, a loss of control over drinking behavior, the development of craving for alcohol, and the occurrence of withdrawal symptoms. AUD puts a severe burden on individuals, families, and societies [1] and is highly prevalent in western countries with a twelve-month prevalence of 13.9% in the United States [2] and around 7% in Europe, where prevalence differs over a wide range between countries [3].

More than 80% of people suffering from AUD do not receive formal treatment [2], and of those who do, only 25–43% remain abstinent [4, 5]. To increase the efficacy of AUD treatment, novel treatment approaches are needed which would optimally be designed as modules that can flexibly be administered in personalized multimodal treatment programs.

The progress of basic neuroscientific knowledge about alcohol addiction and advances in methodology now make neuroscience-based treatment approaches like transcranial magnetic stimulation (TMS) [6–9], transcranial direct current stimulation (tDCS) [10–13], deep brain stimulation (DBS) [14, 15], and real-time functional magnetic resonance neurofeedback (rtfMRI NFB) [16–18] available, which are able to target disease-related brain regions and brain processes with more or less precision.

One process that has prominently been linked to AUD and is a promising aim for targeted neuroscience-based interventions is the cue-reactivity response [19–22], the reaction to the presentation of alcohol-related stimuli, e.g. pictures, odors, or taste of alcoholic beverages. In the brain, core regions associated with cue-reactivity are the ventral striatum (VS), the anterior cingulate cortex (ACC), and the ventromedial prefrontal cortex (vmPFC) [23, 24].

Reviewing the cognitive neuroscience literature on behavioral change in AUD, Naqvi & Morgenstern [25] concluded that enhanced reactivity of the ventral striatum to alcohol and alcohol-related cues together with impairments in prefrontal control regions are the major factors responsible for maintaining AUD. Along these lines, Becker et al. [26] have recently demonstrated enhanced striatal reward sensitivity and impairments in prefrontal-striatal connectivity in patients with AUD in a monetary reward task, hinting at a more generalized impairment in reward processing in AUD, which has also been found being predictive for neural changes during psychotherapeutic treatment [27].

To target the cortical and subcortical processes involved in cue-reactivity and cognitive control, real-time fMRI Neurofeedback is a promising approach. In rtfMRI NFB participants are shown a near real-time feedback signal indicating a specific brain process while they are lying in an MR scanner and are instructed to influence this feedback signal in a desired direction (see Fig. 2 for an illustration). In contrast to other treatment approaches like pharmacological interventions, TMS, or DBS, in rtfRMI NFB patients are actively engaging to influence a brain process, which might have additional psychological therapeutic benefits by promoting self-efficacy and self-regulation [28]. Furthermore, in comparison, rtfMRI NFB offers a relatively high spatial acuity, especially in subcortical brain regions.

While rtfRMI NFB treatment is a relatively new tool in psychiatry, first results are promising. Initial evidence has been presented that rtfRMI NFB has effects for example in major depressive disorder [29–34], schizophrenia [35–37], attention deficit hyperactivity disorder [38, 39], phobia [40], autism [41], posttraumatic stress disorder [42], and Borderline personality disorder [43].

With regard to substance use disorders, rtfMRI NFB has been mainly explored in the treatment of nicotine addiction [44]. Several studies have demonstrated a reduction of nicotine craving through NFB [45–48]. Interestingly, a recent study has found that including functional connectivity information in the feedback signal leads to a higher reduction of nicotine craving [49].

So far, only few studies have been conducted to assess the use of rtfRMI NFB as an interventional tool in AUD. Showing that neurofeedback training with feedback from individually chosen cortical regions was able to reduce craving in patients with AUD directly after neurofeedback training, Karch et al. [50] demonstrated the feasibility of neurofeedback interventions in AUD. In a study with non-treatment seeking heavy social drinkers, Kirsch et al. [18] found that participants who received real feedback in comparison to a yoked feedback group and a sham control group were able to learn to downregulate their ventral striatum activation during presentation of pictures of alcoholic beverages. They could further show that downregulation of the VS was correlated with activation in the right inferior frontal gyrus in the real feedback group, which is in accordance with the model of Naqvi & Morgenstern [25].

In the Systems Biology of Alcohol Addiction (SyBil-AA) rtfRMI NFB study we are now testing whether neurofeedback training will have beneficial effects in heavily impaired AUD patients in a clinical setting. The results of this study will provide insights into the efficacy of NFB training in AUD, the involved brain systems, and might help to identify predictors of treatment success which could potentially be useful in developing personalized treatment approaches.

In the Systems Biology of Alcohol Addiction (SyBil-AA) rtfRMI NFB study we are assessing the efficacy of rtfRMI NFB on ventral striatal cue-reactivity and frontal control processes in the treatment of alcohol use disorder. To the best of our knowledge, the study is one of two ongoing clinical trials of rtfMRI NFB in AUD (see Cox et al. [56] for a description of the other trial), and will contribute to the understanding of the use of rtfMRI NFB as a clinical application in the context of a clinical treatment program for AUD.

Besides the primary goal of testing clinical effects, the study will provide data which will be useful to investigate how rtfMRI NFB is leading to changes in the brain, whether intervention success can be predicted by means of neurobiological signatures and clinical variables, and for which patients rtfMRI might be beneficial. Since we focus on ventral striatal activation to alcohol cues and its prefrontal control, it could be expected that particularly those patients will profit from the intervention that show substantial ventral striatal cue reactivity to alcohol cues prior to the intervention as it has been shown for other treatments before [57]. However, it has been shown that AUD is often associated with a shift of cue reactivity from ventral to dorsal striatum [58]. For those patients, showing a more habit like addiction behavior, instead of the ventral striatum, the dorsal striatum might be an interesting target for NFB interventions in AUD. Nevertheless, there is no doubt, that prefrontal control regions play a central role in craving and relapse [25]. Therefore, it could be speculated that patients receiving a PFC upregulation feedback might benefit from the NF treatment even if they do not show strong ventral but increased dorsal striatal activation to alcohol related cues.

An innovative condition in the present study is the connectivity feedback condition. It has been shown in the context of nicotine addiction that adding connectivity information to the feedback signal increases efficiency [49] and reduces cigarette craving but this has neither been shown for alcohol patients nor with respect a long-term outcome of a treatment.

However, the results from Kirsch et al. [18] could also imply that a critical factor for the treatment effect could be whether patients get the impression that they are able to actively control their own brain response to alcohol cues. Since patients can assume that the intention of the training is the reduction of alcohol cue associated craving, they might apply strategies to reduce this craving and, accordingly, to downregulate the feedback signal. Kirsch et al. [18] showed that successful VS downregulation was accompanied by the application of specific cognitive strategies, which might have let to the concurrent increase of prefrontal activation, although this activation was not fed back to the participants. Thus, the direction of the signal modulation could be a critical point in NFB, and patients receiving the downregulation instruction might show more effects than those receiving the upregulation instruction.

Future studies could further elaborate the effect of instruction by converting the feedback signal in upregulation conditions and instructing all participants to downregulate the signal, which from our experience seems to be the natural direction expected by patients in a cue-reactivity addiction context. This would also allow blinding the patients completely, not only with respect to the group (experimental vs. control) but also the kind of regulation they are expected to apply.

The study could also reveal additional mechanistic information about the mode of action of rtFMRI NF. Importantly, we conduct fMRI scans with whole-brain coverage, which will allow us to investigate the role of other brain regions like amygdala, insula cortex, or ventromedial PFC which were not directly targeted with NFB in our study but play important roles alcohol-related cue-reactivity [23, 24].

Another interesting aspect which might have an influence on the results is the type of control for neurofeedback (see e.g. [59]). We chose an active type of control in the NFB study in which participants from the control group receive a real feedback signal from a control region, which in our case is the auditory cortex that is not involved in cue-reactivity [22]. Other possibilities would be to use a control group without feedback, a group that just views feedback signal of other participants without instructions, or a yoke-control group that is told to regulate the process but receives feedback from another person. For task comparability we decided against the first two options, and for ethical reasons a yoke-control group in which participants would receive fake instructions is questionable. Our active-control groups will lead to results based on specific target brain processes, and not per se on brain regulation through NFB. However, differences between groups might still be related to differences in regulation difficulty.

There are some shortcomings of the protocol. First, the study is only single-blind. The staff members must prepare the individual feedback set up (definition of ROI, definition of direction) and are therefore not blinded. In future studies, a double-blind protocol could be implemented by separating staff that prepares the individual NF setup and staff that instructs the participants. However, we think it is worth to test the general applicability of rtfMRI NF to AUD patients before setting up such an extensive study protocol. Second, the study was retrospectively registered in the German Clinical Trials Register (DRKS00010253; WHO Universal Trial Number (UTN): U1111–1181-4218). Due to fast inclusion of the first patient, registration was conducted shortly after the begin of the study, but before any data was processed or analyzed (Inclusion of first patient: 03/30/2016; application for registration: 04/26/2016; registration date: 05/10/2016). However, the registration complies with the goals of the study described in the grant proposal and the study protocol will not be altered during the study. Finally, we have not defined stopping criteria, because we do not expect side effects besides those that can be expected from fMRI scanning, which is a safe and well-established technique. However, the principle investigators will decide whether to continue or terminate the study or specific parts of it if side effects were reported by the participants.

Overall, the study is well suited to provide valuable data on the efficacy of rtfMRI NFB as an add-on treatment for AUD in a clinical setting. Furthermore, the large amount of imaging data collected on the three scanning days will be helpful to better understand how rtfMRI NFB is working, and to identify factors that predict its clinical effects and could potentially be useful for developing personalized treatment approaches.