Real-time fMRI neurofeedback is an emerging technique that allows to learn voluntary control over spatially localized brain activity (Weiskopf et al., 2004; deCharms, 2007; Sulzer et al., 2013). It can be used to study causal brain-function relationships by investigating how learned self-regulation of brain activity affects perception or behavior. For example, several studies have shown that self-regulation leads to behavioral effects that are specific to the functional role of the targeted cortical area (e.g., Weiskopf et al., 2003, 2004; deCharms et al., 2005; Bray et al., 2007; Caria et al., 2007; Rota et al., 2009; Shibata et al., 2011; Scharnowski et al., 2012). Real-time fMRI neurofeedback also holds great promises for clinical applications. For example, recent studies demonstrated therapeutic effects of neurofeedback training in chronic pain patients (deCharms et al., 2005), Parkinson’s disease (Subramanian et al., 2011), tinnitus (Haller et al., 2010), and depression (Linden et al., 2012).

However, in order for neurofeedback to be effective as a tool for cognitive enhancements or clinical applications, it needs to be shown that learned self-regulation transfers to situations where neurofeedback is not available anymore, and that learned self-regulation is maintained beyond the initial training period. Previous real-time fMRI neurofeedback studies already demonstrated that once learned, self-regulation can also be performed in transfer runs without feedback information, but this only for transfer runs immediately following the neurofeedback training (deCharms et al., 2004, 2005; Hamilton et al., 2011; Lee et al., 2012; Scharnowski et al., 2012; Sitaram et al., 2012; Ruiz et al., 2013). Unfortunately, published evidence of lasting neurofeedback training effects is very sparse. In healthy participants, two studies have shown that neurofeedback training induces plastic brain changes that last for at least 1 day after the training (Shibata et al., 2011; Harmelech et al., 2013). Another study found that neurofeedback-induced changes in resting state fMRI persisted for at least 2 months (Megumi et al., 2015). While this is by far the longest time period after training that has been investigated so far, the resting state effects demonstrate plastic changes, but they are not associated with applying learned self-regulation by the participants. In patients, only one study showed lasting changes due to neurofeedback training (Scheinost et al., 2013). Scheinost et al. (2013) found that patients with obsessive-compulsive disorder who successfully learnt to increase activity in the orbitofrontal cortex showed persistent changes in resting state connectivity and a significant reduction in contamination anxiety several days following the neurofeedback training. Even in the field of EEG neurofeedback, which has been used in research and in clinics for several decades and therefore has a much longer tradition than real-time fMRI-based neurofeedback, follow-up data has rarely been collected. The few studies which measured the effects of EEG neurofeedback over a period of up to 12 months to determine if self-regulation and its behavioral consequences are maintained, indicate that it remains stable (Tansey, 1993; Kotchoubey et al., 2001; Vernon et al., 2004; Leins et al., 2007; Kouijzer et al., 2008).

However, because EEG- and fMRI-based neurofeedback differ significantly with respect to the target areas and the physiological basis of the feedback signals, the findings obtained with EEG might not be transferable. We therefore re-scanned five participants 6 and 14 months after they successfully learned self-regulation of differential visual cortex activity, i.e., participants had learned to control the interhemispheric balance between their left and right visual cortex activity (Robineau et al., 2014). For the 6-month follow-up, participants were provided with the same neurofeedback information as for the initial training (top up session). For the 14-month follow-up, participants did not receive feedback information (transfer). We hypothesized that learned self-regulation is an acquired skill which can be maintained over longer periods of time. Specifically, we hypothesized that participants can still successfully self-regulate their visual cortex activity during the follow-up sessions, and that performance during the follow-up sessions is similar to that achieved at the end of the successful initial training period.

Details of the neurofeedback training study have been described previously (Robineau et al., 2014). In brief, participants were trained to control the differential feedback between a target region of interest in early visual cortex (ROI_target), and its homolog in the opposite hemisphere (ROI_opposite) (Figure 1). The ROIs were delineated in separate functional localizer scans and represented specific locations in the left and right visual fields. For the localizer, participants had to maintain fixation on a central point while a flickering checkerboard wedge (100% contrast, 8 Hz contrast reversal, 30° eccentricity along the horizontal meridian at a 45° angle) was presented on a gray background. The checkerboard was presented for three blocks of 30 s alternating in the left and in the right visual field, respectively, which were interleaved with baseline blocks during which participants fixated without any visual stimulation. Training participants to control the differential feedback signal was undertaken in three separate scanning sessions spread over the course of 3 days. For neurofeedback training, the activity difference between the two ROIs was fed back to the participant in the form of a visual thermometer display on a projection screen in the scanner bore. No other visual stimuli were presented. Neurofeedback training runs were interleaved with transfer runs during which participants performed learned self-regulation in the absence of feedback. For clarity, the main experimental parameters will be repeated here, but the methods section focuses on the follow-up extension of our previous study.

The subgroup of participants in the follow-up sessions successfully learned to control the differential feedback signal (Figure 2). The beta estimates increased significantly between sessions 1 and 2, and between sessions 2 and 3 [training: F(10,97) = 15.81, all ps < 0.01; transfer: F(10,52) = 12.88, all ps < 0.01]. Also, the t-test showed significantly positive beta values in the last session of the initial training (training: t = 3.60, df = 31, p = < 0.01; transfer: t = 3.46, df = 15, p ≤ 0.01), indicating a reliable modulation of brain activity during regulation compared to baseline blocks. Importantly, performance during the follow-up sessions did not differ from performance in the last training/transfer session of the initial training study [training sessions: F(10,97) = 15.81, p = 0.21; transfer sessions: F(10,52) = 12.88, p = 0.29].

Our study showed for the first time that learned voluntary control over brain activity can be maintained for a substantial period of time. In follow-up sessions we found that participants were able to perform self-regulation with neurofeedback just as well as they did at the end of the initial training period which had taken place 6 months earlier. Moreover, 14 months after the initial training participants were still able to exert learned control, and this during transfer runs without further neurofeedback information. These preliminary results in a small sample indicate that neurofeedback training effects last, and that participants can use their learned skill far beyond the initial training.

Neurofeedback based on real-time fMRI neurofeedback is a rapidly growing field because it allows for learning voluntary control over localized brain activity. To date, several studies show that neurofeedback training causes behavioral effects that are specific to the functional role of the targeted cortical area (Weiskopf et al., 2004; Bray et al., 2007; Caria et al., 2007; Rota et al., 2009; Shibata et al., 2011; Scharnowski et al., 2012, 2015; Robineau et al., 2014; Koush et al., 2015; Scharnowski and Weiskopf, 2015). Even more importantly, real-time fMRI neurofeedback training has also been shown to have therapeutic effects in chronic pain patients (deCharms et al., 2005; Guan et al., 2015), Parkinson’s disease (Subramanian et al., 2011), tinnitus (Haller et al., 2010), depression (Linden et al., 2012; Young et al., 2014), obsessive-compulsive disorder (Scheinost et al., 2013, 2014), spider phobia (Zilverstand et al., 2015), and addiction (Li et al., 2013; Karch et al., 2015; Kirsch et al., 2015; Hartwell et al., 2016). Especially for clinical applications of neurofeedback it is crucial that the learning effects persist beyond the initial training period and that voluntary control transfers to situations without neurofeedback information.

Our study significantly extends previous investigations which showed that neurofeedback training effects are still present 1 day (Shibata et al., 2011; Harmelech et al., 2013), several days (Scheinost et al., 2013), or 2 months after the training (Megumi et al., 2015). Firstly, we demonstrate that learned self-regulation is maintained for at least 14 months, which is by far the longest time period that has been investigated so far (Shibata et al., 2011; Harmelech et al., 2013; Scheinost et al., 2013; Megumi et al., 2015). One of the factors that probably contributed to such long lasting effects is that – like Scheinost et al. (2013) and Megumi et al. (2015) – we trained participants over several days, which gave them the chance to consolidate their learning overnight. Sleep has been shown to be beneficial for memory encoding and learning, which may have enhanced the persistence of learned self-regulation (Walker and Stickgold, 2004; Abel et al., 2013). Our results are in line with previous findings from the field of EEG-based neurofeedback, where follow-up studies revealed that learned self-regulation and its behavioral consequences are maintained for at least 12 months (Tansey, 1993; Kotchoubey et al., 2001; Vernon et al., 2004; Leins et al., 2007; Kouijzer et al., 2008).

Secondly, we show that self-regulation can be applied months after the initial training without neurofeedback information. That learned control over brain activity transfers to situations where neurofeedback is no longer available is another key requirement for clinical uses of neurofeedback (Stoeckel et al., 2014). Transfer effects have been previously demonstrated immediately following neurofeedback training (deCharms et al., 2004, 2005; Hamilton et al., 2011; Lee et al., 2012; Scharnowski et al., 2012; Sitaram et al., 2012; Ruiz et al., 2013), but here we extend these findings by demonstrating that transfer effects last for several months. To promote transfer, we interleaved the neurofeedback training with transfer runs without neurofeedback (Robineau et al., 2014), which might have contributed to the persisting transfer performance. Finally, the lasting changes that we found relate to self-regulation skills that can be applied voluntarily by the trained participants rather than to concomitant training effects such as the previously reported persistent resting state changes (Scheinost et al., 2013; Megumi et al., 2015).

The main limitation of this study is the small sample size of only four participants in the first follow-up session and five participants in the second follow-up session. Thus, our results should be considered as a preliminary demonstration of lasting effects that require further verification in larger samples. The second limitation is that we did not include a control group without neurofeedback that attempts self-regulation based only on cognitive task instructions. Without such a control group, we cannot completely exclude the possibility that mere practice led to the improvement across sessions during the initial training and transfer runs, and that these practice effects could have then carried over to the follow-up study. However, in the original training study, the non-learners also practiced self-regulation for as long as did the learners, but they did not show improvements. Moreover, other real-time fMRI neurofeedback studies that included control groups who received either sham feedback or no feedback have firmly established that neurofeedback is necessary for learning to self-regulate brain activity [e.g., for the anterior cingulate cortex (Hamilton et al., 2011), for the inferior frontal gyrus (Rota et al., 2009), and most importantly for the visual cortex (Shibata et al., 2011; Scharnowski et al., 2012)]. Finally, in this follow-up study, we did not ask participants to perform the visual detection and line bisection tasks that they performed during the original training. Hence, long-lasting behavioral consequences of neurofeedback training need to be addressed in future follow-up studies.

Taken together, we show in a small sample of healthy participants that learned self-regulation of differential visual cortex activity can be maintained over long periods of time and transfers to situations without neurofeedback. These findings are likely not specific for learned control over visual cortex activity, but probably also apply to other neurofeedback target regions. The visual cortex is an area that has been shown to be less responsive to neurofeedback training (Harmelech et al., 2015), which is also reflected by the relatively large number of participants who failed to learn to control visual cortex activity (Scharnowski et al., 2012; Robineau et al., 2014). Moreover, compared to standard ROI-based feedback, the differential feedback signal that was used in the present study likely is more difficult to control because uncorrelated Gaussian noise of the signals from the two ROIs is additive and may thus reduce the signal-to-noise ratio (SNR). Hence, our results imply that neurofeedback training with real-time fMRI causes plastic brain changes rather than just short-lived state changes. Although the neural mechanisms underlying these lasting changes are still unknown (Sitaram et al., 2016), evidence for such lasting changes as well as for transfer to situations without neurofeedback support the potential of neurofeedback as a promising novel experimental therapy for neurological and psychiatric conditions.

FR, PV, and FS conceived and designed the experiments. FR performed the experiments. FR, DM, YK, SR, CM, SM, DV, and FS analyzed the data/contributed analysis tools. FR and FS wrote the paper and all authors critically reviewed and approved the final manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.