Closed-loop neurofeedback reshapes preparatory brain states to bias subsequent pain processing

Reviewer #1 (Public review):

Summary:

Zhang et al. investigated EEG neurofeedback as a method to modulate brain activity prior to painful stimulation and its effect on pain perception. Neurofeedback was designed to train participants to upregulate alpha power contralateral to the site of painful stimulation. Real or sham neurofeedback was administered to two independent groups. Each group performed two tasks: one in which participants were asked to modulate their brain signals (training task) and another in which they were asked to passively watch the feedback (non-training task). The authors reported an increase in alpha power during real neurofeedback training compared with sham training and non-training conditions. The authors also reported a decrease in pain perception during the training task, both in the real and sham neurofeedback groups. Additionally, in an offline analysis, the authors investigated brain dynamics with microstate analysis during the neurofeedback training. Also, they implemented a mediation analysis to infer which brain responses to neurofeedback training mediated changes in pain perception.

Strengths:

(1) The research question is licit and sound. EEG neurofeedback is a promising non-invasive technique with the potential to alleviate at least the sensory component of pain. The rationale for applying neurofeedback at the alpha band in the somatosensory cortex is well justified by the alpha-gating theory in pain modulation.

(2) The sample size is adequate to capture neurofeedback effects. The effort to conduct a double-blind study with a complex design paradigm and an adequate sample size is valuable and appreciated.

Weaknesses:

(1) Reported behavioral effects on pain reduction might be due to the placebo effect rather than neurofeedback, as pain ratings were reduced both in the real and sham neurofeedback groups during training. It is important that authors report this effect appropriately and disclose which information was given to the participants when they enrolled in the study, i.e., whether the paradigm was designed to reduce pain perception.

(2) The utility of training effects, especially in the sham group, is unclear. I understand that including the non-training condition allows the distinction between neurofeedback effects and arousal effects. However, interpreting training effects should not be the point of this study. What does it tell us that participants who received sham stimulation increased or decreased alpha power in the training session vs the non-training session?

(3) There might be hidden time effects (habituation/sensitization) on pain responses and/or on brain responses to neurofeedback. A within-session analysis comparing the first half of the training with the second half should be conducted to discard them.

(4) Connectivity analysis reflects spurious effects. In EEG, deriving phase-based functional connectivity at the sensor level is problematic due to volume conduction effects. EEG functional connectivity should be performed after source reconstruction, and measures discarding instantaneous phase lags should be preferred, which is not the case with magnitude-squared coherence. See (Bastos and Schoffelen, 2015).

Although neurofeedback is a promising technique for modulating pain perception, the current study adds limited novelty to the field, as its design could not disentangle whether behavioral effects (reductions in pain intensity and unpleasantness) were specific to neurofeedback training or due to non-specific effects (e.g., placebo). Nevertheless, the authors corroborated that brain states before painful stimuli could be modulated with neurofeedback (enhancement of alpha power).

Reviewer #2 (Public review):

Summary:

This study uses neurofeedback to modulate alpha-band activity and examines how this influences pain-related processing. The question is timely and methodologically elegant, because it addresses whether noninvasive modulation of ongoing oscillatory activity can causally shape pain perception and/or expectation-related processes.

Strengths:

The use of neurofeedback as a tool to modulate alpha activity is a major strength, because it provides a noninvasive and conceptually clean approach to probing the functional role of oscillatory brain activity. The design is also attractive because it links neurophysiological regulation to a psychologically meaningful outcome, namely pain processing. Further, the induced changes were also related to different EEG microstates and ERP components during the processing of the pain stimulus, and therefore the authors demonstrate a clear relation between preparatory prestimulus states and stimulus processing.

The manuscript appears to address an important and clinically relevant question, and the idea of testing whether alpha regulation can alter pain-related responses is of high interest for systems neuroscience and pain research.

Weaknesses:

Methodologically, it is unclear what alpha values were used in the analyses. It is stated that alpha was extracted within 2s windows of the 16s long feedback period. However, the values change across this period. Which value is used for the correlation with the pain ratings and all other analyses? Using the average across the 16s could reflect large values in the first half and low values in the final half, but for the relationship between alpha and pain, the last segments should be more relevant. If the initially elevated alpha activity subsides several seconds before the onset of the pain stimulus, it is difficult to see how it could influence subsequent pain processing.

Related, after the 16s feedback period, a fixation period is used with a 3-5s length. If alpha band activity is relevant for the consecutive pain processing, the amount of alpha in this period should be relevant. The authors should demonstrate that the induced alpha change during the feedback period remains stable during the fixation period and that the activity in this period is related to pain processing.

Further, it should be noted that the alpha band modulations related to alpha band training were accompanied by significant effects in other frequencies. Therefore, a clear relationship between alpha and behavioral pain ratings is not the only interpretation. Correlations with other frequencies or combinations of frequency band modulations should be incorporated to allow a more precise interpretation. Furthermore, in the sham feedback group, an increase in alpha band activity was observed (p=0.06), and the small difference in the pain intensity rating may be related to a clear outlier in the Sham group (Figure 4a).

In both groups, a main effect of training, regardless of sham or real feedback, was reported with a small difference between groups. But the main modulator seems to be related to the instruction to modulate the neural activity, and this large effect should be discussed in more detail regarding, for example, possible attentional processes.

A further central concern is that the visual feedback signal (the ball movement) may generate expectations that are not specific to alpha activity and that these expectation processes modulate the pain processing (ball down may indicate more pain). It is well known that intensity cues can generate expectations about upcoming perceptions, and the used feedback signal with an increasing or decreasing visual curve clearly signals what intensity should be expected. Therefore, it is important to show that the amount of positive (ball up) and negative visual displays is matched between the sham and real feedback group. Further, the authors should report whether the final ball position can predict the latter pain rating in both groups or differentially. Following this interpretation, alpha band activity is not directly related to pain processing but only serves as a signal that is transformed to a visual stimulus that then generates expectations.

Finally, the manuscript would benefit from a more explicit analysis of whether individual alpha changes are related to pain ratings within each subject. If higher alpha is truly linked to reduced pain perception, this should be visible at the participant level during learning of the neurofeedback procedure. Relatedly, there is no learning period incorporated, and usually participants are not able to regulate their alpha activity from the first trial on. The authors should include an analysis of the development of alpha band activity over learning and a relation of these individual alpha values and the corresponding pain ratings.

I cannot find a link to the preregistration in the current manuscript.

In summary, a "causal" relation of alpha activity with pain perception -that is mentioned several times in the manuscript- is not fully supported by the present results