Human neuronal excitation/inhibition balance explains and predicts neurostimulation induced learning benefits

Abstract

Previous research has highlighted the role of the excitation/inhibition (E/I) ratio for typical and atypical development, mental health, cognition, and learning. Other research has highlighted the benefits of high-frequency transcranial random noise stimulation (tRNS)-an excitatory form of neurostimulation-on learning. We examined the E/I as a potential mechanism and studied whether tRNS effect on learning depends on E/I as measured by the aperiodic exponent as its putative marker. In addition to manipulating E/I using tRNS, we also manipulated the level of learning (learning/overlearning) that has been shown to influence E/I. Participants (n = 102) received either sham stimulation or 20-minute tRNS over the dorsolateral prefrontal cortex (DLPFC) during a mathematical learning task. We showed that tRNS increased E/I, as reflected by the aperiodic exponent, and that lower E/I predicted greater benefit from tRNS specifically for the learning task. In contrast to previous magnetic resonance spectroscopy (MRS)-based E/I studies, we found no effect of the level of learning on E/I. A further analysis using a different data set suggest that both measures of E/I (EEG versus MRS) may reflect, at least partly, different biological mechanisms. Our results highlight the role of E/I as a marker for neurostimulation efficacy and learning. This mechanistic understanding provides better opportunities for augmented learning and personalized interventions.

Fig 1. E/I and the aperiodic exponent.

(A) A simplified overview of the difference between periodic and aperiodic activity in the EEG power–frequency spectrum. (B) The aperiodic exponent in log–log space as shown in the EEG spectrum. (C) and (D) are adapted with permission from Gao and colleagues [28], which show that high E/I is related to a flatter (closer to zero) aperiodic exponent and low E/I (i.e., high inhibition) to a negatively steep exponent, compared to the LFP. EEG, electroencephalogram; E/I, excitation/inhibition; LFP, local field potential.

Fig 2. A schematic overview of the task structure and experimental protocol.

(A) First, a fixation screen was shown. Subsequently, a multiplication was presented by voice recording through a headphone. Hereafter, participants were shown a microphone symbol to indicate that they could say the answer into the microphone. This was followed by a 200-ms delay period. Lastly, participants indicated by clicking the left or right mousepad on the keyboard whether they retrieved or calculated the answer. (B) First, a pre-rs-EEG was measured of 8 minutes. Subsequently, a training was presented that contained 4 different multiplication problems. Based on baseline performance, participants either completed the learning or the overlearning task. One block in both the learning and overlearning task consisted of 10 multiplications with 18 blocks, and 180 trials in total. Participants received 20 minutes, 1 mA tRNS during either the learning or overlearning task or sham stimulation. Next, the transfer task was presented with new arithmetic problems containing 10 multiplications repeated 3 times. The recall task contained the identical multiplications as either the learning or the overlearning task and was repeated 3 times. Lastly, another post-rs-EEG measurement of 8 minutes was assessed. (C) Placement of the stimulation electrodes over F3 and F4. EEG, electroencephalogram; rs, resting state; tRNS, transcranial random noise stimulation.

Fig 3. Averaged learning curves of the median RTs of the learning or overlearning task of the sham stimulation.

The mean learning curve of the participants (n = 22) during learning shows a linear gradient as shown in purple. The mean learning curve of the participants (n = 21) during overlearning (in green) shows a clear plateau of performance improvement after approximately block 10, and faster RTs overall. Shading indicates 95% confidence intervals.

Fig 4. Changes in aperiodic exponent for active and sham tRNS and their associated topographies.

(A) Participants who received active tRNS showed an increase in E/I as indicated by the mean (±SEM) decreased aperiodic exponent (change: post-baseline exponent in μV² Hz⁻¹). Participants who received sham tRNS showed a mean (±SEM) decrease in E/I as indicated by an increased exponent. *p < .05. (B) Topoplot illustrates the change in the aperiodic exponent for the sham (left) and active tRNS (left and right respectively). For electrode Fz, a slight increase in the aperiodic exponent is observed as indicated with a lighter color for sham tRNS. For active tRNS, there is a clear decrease in the aperiodic exponent after tRNS, as indicated with a darker color for electrode Fz and anterior electrodes. This observation supports the notion of increased E/I following active compared to sham tRNS.

Fig 5. Posterior density of tRNS and the three-way interaction between tRNS, task, and baseline aperiodic exponent from the best fitted model.

(A) The posterior density of stimulation (tRNS) shows that 90% of the posterior distribution is below zero. Indicating that there is a 90% probability that tRNS lowers RTs during the learning and overlearning task. (B) Posterior distribution of the three-way interaction between tRNS, task, and baseline aperiodic exponent that shows that there is a 82% probability of this interaction being present. (C) The left panel indicates the marginal effects of the learning task for low baseline aperiodic exponent values (mean −1 SD) and high baseline aperiodic exponent values (mean +1 SD). Sham stimulation is indicated in red and tRNS in blue. The right panel indicates the same marginal effects of the overlearning task. This plot shows that tRNS improved performance, but this was restricted to participants with a high baseline aperiodic exponent in the learning task. No effect was found for participants with a low baseline aperiodic exponent in the learning task, and neither were there any beneficial effects of tRNS in the overlearning task. 95% CrI are indicated. RT, response time; SD, standard deviation; tRNS, transcranial random noise stimulation.