Twenty-five participants (Females = 16, Mage = 23.5, SD = 1.37, Males = 9, Mage = 23.67, SD = 1.00 years) volunteered to take part in the experiment. We recruited participants via advertisement posters. All participants reported being free from illness and injury at the time of the experiment. We obtained informed consent from all participants. The experiment was approved by the University research ethics committee.
G*Power 3.1 power calculation software (Faul et al. 2013) indicated that by adopting an alpha of 0.05 and a sample size of 25, the experiment was powered at 0.80 to detect within-participant differences for effect sizes exceeding f = 0.26 (i.e., medium-size effects) by repeated-measures analysis of variance (ANOVA; Cohen 1992). In a previous study of the effects of neurofeedback training on motor performance, Cheng et al. (2015) reported a significant and large within-participant effect (ηp2 = 0.63; performance improvement from pre- to post-training). Accordingly, if similar effects were to emerge, our sample was adequately powered to detect them.
We adopted a within-participant design with three primary factors: Task, Test, and Condition. The Task factor had two levels: single-task and dual-task. All participants completed a timed up-and-go walking assessment on its own (i.e., single-task), and while concurrently performing a serial-sevens cognitive task (i.e., dual-task). The Test factor also had two levels: pre-test and post-test. All participants completed their walking tasks both before (i.e., pre-test) and after (i.e., post-test) receiving neurofeedback. The Condition factor had three levels: decrease alpha power, increase alpha power, and sham. Each condition comprised 30-min of neurofeedback training. In the decrease alpha power neurofeedback condition, participants received neurofeedback training that encouraged them to decrease their central midline alpha power. In the increase alpha power neurofeedback condition participants received neurofeedback training that encouraged them to increase their central midline alpha power. In the sham neurofeedback condition participants received non-contingent (i.e., fake) neurofeedback training. All participants made three laboratory visits on separate days to complete all three neurofeedback conditions. This ensured a fully within-participant 3 Condition (decrease alpha; increase alpha; sham) × 2 Test (pre-test; post-test) × 2 Task (single-task; dual-task) design. A schematic of the design is provided in Fig. 1, and a more detailed description of the Task and Condition factors are respectfully provided in the “Task” and in the “Neurofeedback training protocol” sections below.
Single-task: timed “Up and Go” test
The Timed “Up and Go” test (TUG), is an established clinical tool utilized to assess instability, postural control and lower limb functional mobility (Vance et al. 2015). It requires the execution of everyday movements (i.e., standing, walking, turning) therefore, providing an efficacious means of predicting the risk of falls and identifying individuals with impaired motor function (Brustio et al. 2017; Nocera et al. 2013). The TUG test measures the total amount of time required to rise from a seated chair position (seat height 45 cm), walk a distance of 3 m, turn around a cone, walk back to the chair and sit back down. Healthy adults typically complete the task in 8–10 s; times exceeding 10 s normally indicate reduced physical capacity and/or balance and mobility problems, with a positive correlation between TUG and risk of falls (Bohannon 2006; Kear et al. 2017).
Dual-task: timed “Up and Go” test + serial sevens
In the dual-task condition participants completed the TUG while simultaneously performing a serial sevens cognitive task. Specifically, we instructed participants to serially subtract by sevens from a prescribed three-digit number (ranged from 100–300) and recite out loud the calculations while performing the TUG test. The serial sevens task loads working memory and increases the demand for attentional resources and is regularly used alongside walking to establish a genuine motor and cognitive dual-task (Montero-Odasso et al. 2012).
Neurofeedback training protocol
Participants received a 30-min session of neurofeedback training during each of their three laboratory visits. Cortical activity was recorded during the increase alpha and the decrease alpha neurofeedback conditions from the central midline of the scalp (i.e., Cz electrode site; Jasper 1958) which roughly overlies the supplementary motor area (Gerloff et al. 1997), using an active electrode connected to a wireless 4-channel neurofeedback system (Brainquiry PET-4, Nijmegen, Netherlands). Additionally, an active electrode was placed over the orbicularis oculi muscle of the right eye to remove eyeblink artefacts, with reference and ground electrodes attached to the right and left mastoids (Ring et al. 2015). We focused our feedback at the Cz site because the supplementary motor area supports movement planning and posture preparation to ensure balance regulation during gait (Takakusaki 2013), and because supplementary motor area dysfunction is associated with gait disturbance (Iseki et al. 2010). In tandem with cortical recordings, a computer running Bioexplorer software (Cyberevolution) used a 6th order Butterworth infinite impulse response 8–12 Hz bandpass filter to extract alpha power (8–12 Hz) from the EEG signal and fed this back to the participants in the form of an auditory tone (Ring et al. 2015). Importantly, the tone was programmed to vary in pitch based on the level of alpha power and silence completely when alpha power was decreased (during the decrease alpha condition) or increased (during the increase alpha condition) by 30%, relative to each participant’s individual baseline. As alpha power is inversely related with cortical activity, the decrease alpha condition encouraged increased activation at the Cz site above the supplementary motor area, which is characteristic of relatively autonomous locomotion (Morris et al. 1996), and the increase alpha condition encouraged decreased activation above the supplementary motor area, which is characteristic of de-automized locomotion and gait disturbance (Iseki et al. 2010). In addition to changing alpha power by 30%, the system also required < 10 µV of 50 Hz activity in the signal (i.e., low impedance) and the absence of eye-blinks, as detected by the electrode paced adjacent to the right eye (eye-blinks were detected as > 75 µV of 1–7 Hz activity at the eye-electrode), for the tone to silence. These control features helped ensure the signal was being regulated by cognitive processes and was not contaminated by electrical, muscular or eye-blink artefacts (Ring et al. 2015).
The auditory neurofeedback training was delivered to participants over ten 3-min blocks, each separated by a 1-min break. Participants were seated in the chair used for the TUG. Each time the thresholds described above were met, the auditory tone was set to silence for 1.5 s and participants were instructed to stand up from the chair. This instruction was designed to help participants associate relative increased or decreased central midline activation (in the decrease alpha, and increase alpha conditions, respectively) with the onset of movement.
During the sham feedback condition, the procedure was identical except cortical activity was not recordedFootnote 2 and the tone that participants heard was not based on their brain activity. Instead, participants were played a recording of a tone from either their previous visit or from a matched participant taken during one of the other neurofeedback conditions. Accordingly, unbeknownst to the participant, they received no systematic encouragement to change their central midline activation during the sham condition.
Our primary outcome measure was TUG completion time in the single and dual-task conditions and across the pre-test and post-test phases. This is a well-established performance measure in the gait and movement rehabilitation literature (Zirek et al. 2018). Time to completion scores were obtained by recording each walking trial using a video camera (Apple Ipad) and analysing the time-stamped footage.
A secondary outcome measure was cognitive task performance. We calculated percentage accuracy of serial seven responses provided during each trial by recording the number of responses and the number of errors (Tamura et al. 2018).
EEG activity was recorded during the increase alpha and the decrease alpha neurofeedback sessions from the Cz site on the scalp (Jasper 1958), using an active electrode connected to a DC amplifier (Brainquiry PET-4), with reference and ground electrodes attached to the right and left mastoids, respectively. Recording sites were cleaned, abraded, and conductive gel (Signagel, Parker) was applied to ensure that electrode impedances were below 10 kΩ. The signals were digitized at 24-bit resolution (Brainquiry) and transmitted via Bluetooth at a sampling rate of 200 Hz to a computer running Bioexplorer (Cyberevolution) software. We employed Butterworth infinite impulse response (6th order) bandpass filters at 8–12 Hz (alpha power), 4–8 Hz (theta power) and 13–30 Hz (beta power) to extract EEG data from each three-minute recording. EEG alpha power was extracted to provide an indication of how the neurofeedback interventions modulated EEG alpha power. Power from the neighbouring theta and beta power bands was extracted to examine whether the effects of the neurofeedback manipulation were localised to the alpha band. The EEG recording system was only switched on during the neurofeedback phase of the experiment. Cortical activity was not recorded during the pre-test and post-test phases of the experiment due to short recording epochs, low number of trials, and risk of the wireless signals dropping out as participants moved away from the data-receiving computer during TUG trials.
Participants attended three separate 75-min testing sessions (increase alpha, decrease alpha, sham) with each visit separated by a minimum of 24 h and a maximum of 7 days (M inter-session-interval = 3.61 ± 1.22 days) to reduce the potential for any carry over effects. The order of conditions was counterbalanced across participants. At the onset of the first laboratory visit participants were briefed and provided informed consent. All visits then followed the same identical structure. First, participants were seated and fitted with the neurofeedback system. We prepared the skin by lightly abrading over the mastoids and the right orbicularis oculi muscle with exfoliating paste, and with a blunt needle at the scalp site (Cz). The sites were then cleaned with an alcohol wipe, conductive gel was applied, and disposable spot electrodes (BlueSensor, Ambu) were placed and secured using tape and a lycra cap. The EEG amplifier was attached by an elastic and Velcro strap to the participant’s right arm.
After instrumentation, we provided instructions about the TUG test and serial sevens dual-task, and the TUG was demonstrated by the experimenter. Next, the participant completed one TUG familiarisation trial without the leg brace, and we recorded this baseline TUG score. On completion of the familiarisation trial, participants were fitted with a 60 cm wrap around aluminium splint leg brace (Knee Immobiliser, NeoG, Harrogate, United Kingdom—Fig. 2) to immobilise the knee joint of their dominant leg. This feature was designed to disrupt the autonomous control of gait and thereby induce a more conscious form of movement control (Beilock and Carr 2001). De-automizing regular gait was necessary to establish the efficacy of neurofeedback at re-automizing movements, and thereby examine the primary aim of this experiment.
Participants then began the pre-test phase of the experiment, comprising three trials of the single-task and three trials of the dual-task (see task-section above). The order in which they completed the single-task and the dual-task conditions was counterbalanced across participants. There was a 30 s break in between each trial, and a 1-min break in-between the single-task and dual-task conditions.
After the pre-test, the EEG amplifier was switched on and participants were prepared for the neurofeedback phase. We first assessed their baseline alpha power by asking them to fixate on a cross taped to the wall at eye level, for a period of five seconds. During this time, Cz alpha power was monitored. This process was repeated five times and the average was used as their baseline Cz alpha power. Having established individual baselines, the experimenter manually set the threshold for silencing the neurofeedback tone in the neurofeedback software. In the increase alpha condition, the threshold was set as baseline + 30%. In decrease alpha condition, the threshold was set as baseline -30%. In sham condition, the experimenter pretended to enter a threshold in the computer, but no real threshold was set. Participants were then supplied with the following instructions:
“In the next phase of the experiment the computer will play an auditory tone that is based on your real time brain activity. When you reach the target level of brain activity, the tone will turn off and that is your cue to stand up. The neurofeedback training session will last 30 min, in the form of ten 3-min blocks, interspersed with short breaks. Try to recognize how to control the tone with your thoughts. It may be difficult at times, but it should become easier with practice. The goal during each 3-min block is to silence the tone and stand up out of the chair as many times as you can.”
Participants completed the ten 3-min blocks of neurofeedback training, with a 1-min break between each block. On completion of the neurofeedback training, the EEG amplifier was switched off and participants entered the post-test phase, which comprised three trials of the single-task and three trials of the dual-task, in the absence of the auditory tone. This was identical to the pre-test. At the end of each session the leg-brace and neurofeedback hardware were removed, and the participant was thanked and reminded of the time and date of their next visit. At the end of the third visit participants were thanked for a final time and invited to contact the experimenter for the results of the experiment at the end of the data collection period.
We analysed cortical activity during the decrease alpha power and during the increase alpha power neurofeedback training sessions. Separate 2 Condition (decrease alpha, increase alpha) × 10 Block (each of the 3 min neurofeedback sessions) ANOVAs were performed for the alpha, theta, and beta power bands. We expected a Condition × Block interaction to emerge for the alpha band, characterised by significantly greater alpha power in the increase alpha condition than in the decrease alpha condition during the final block of neurofeedback training. We expected no significant effects to emerge in the theta or beta power bands. Significant ANOVA effects were probed by polynomial trend analyses and paired samples t-tests. EEG data were missing for three participants in the increase alpha and one participant in the decrease alpha conditions due to software error. Missing data are reflected in the reported degrees of freedom.
Our primary hypotheses concern walking performance and cognitive performance (i.e., serial sevens response accuracy), so these variables were the focus of our main analyses. We analysed walking performance using a 3 Condition (decrease alpha, increase alpha, sham) × 2 Test (pre-test, post-test) × 2 Task (single-task, dual-task) repeated measures ANOVA. We analysed cognitive performance using a 3 Condition (decrease alpha, increase alpha, sham) × 2 Test (pre-test, post-test) ANOVA. Significant effects were probed by paired-samples t-tests.
Finally, to test our hypothesis that increased automaticity would be responsible for any pre-test to post-test improvement in dual-task walking performance during the decrease alpha power neurofeedback condition, we performed within-participant mediation analyses (Montoya and Hayes 2017). We set pre-test and post-test dual-task walking performance scores from the decrease alpha power condition as the dependent variables and pre-test and post-test serial-sevens response accuracy scores from the decrease alpha power condition as the potential mediator variables using MEMORE for SPSS (MEdiation and MOderation analysis for REpeated measure designs; Montoya and Hayes 2017). The mediation effect (B), standard error (BootSE) and 95% CI (low and high) were reported (Montoya and Hayes 2017).
In addition to our main analyses, we also conducted some control analyses to test some assumptions of our experiment and provide further insight into the data. We tested our assumption that walking with a leg brace would slow gait akin to what happens when individuals revert from autonomous to conscious motor control (Beilock and Carr 2001). We tested the assumption of neurofeedback experiments that participants progressively gain more control over their brain activity as their training sessions progress by analysing the number of times that participants stood up during neurofeedback training. Finally, we analysed number of serial-sevens responses to screen for possible speed-accuracy trade-offs in cognitive task performance. The results of these analyses are presented in the Supplementary Material. In brief, both our assumptions were confirmed, and there was no evidence of speed accuracy trade-offs.