We retained 18 neurologically healthy young adults (8 women; mean age of 22 years, ranging from 18 to 29 years) to participate in our study after screening (see below). All participants were right-handed and reported normal hearing and no history of neurological disease. Nine participants had some musical training, with an average duration of 2.5 years and no participant having more than 4 years of training. They gave their written informed consent and received monetary compensation for their participation. Ethical approval was obtained from Ethics Review Board of the Montreal Neurological Institute (NEU-14-043).

Prior to enrollment, eligible participants completed a 10-minute in-person screening test for their ability to perform an auditory working memory task (shown in Figure 1A and described in section “2.3 Training task”) above chance level. Behavioral data from this screening test were analyzed using the Hits – False Alarm rate. To avoid a potential ceiling effect, an additional exclusion criterion was defined as a Hits – False Alarm rate above 70% on the screening test.

Of 32 participants screened, 19 participants were enrolled for 7 experimental sessions taking place 48–72 h apart. An overview of the training protocol is depicted in Figure 1A. One participant was terminated after the pre-training session due to below chance performance on simple trials of the working memory task (see below and Figure 1A for task description), yielding a final sample of 18 participants for assignment to experimental and sham conditions as described in section “2.5 Experimental protocol.”

The auditory stimuli [taken from Malinovitch et al. (2023)] used in the working memory training task consisted of sequences of three 400 ms pure tones randomly selected from a frequency range of 250 to 1000 Hz, with tones in the same sequence having frequencies at least 20% different from one another. These sequences were delivered binaurally through air-conducting tubes with foam ear tips (70 dB SPL).

The auditory working memory task [from Malinovitch et al. (2023)] administered during the screening test and throughout the 7 experimental sessions involved the mental manipulation of the order of three pure tones and is depicted in Figure 1A. In each 15-s trial, a sequence of three tones was presented, followed by a silent retention period of 3800 ms. A visual cue then appeared on the screen, consisting of the numbers 1, 2, and 3 in variable order. The order of these numbers corresponded to the order of tones in a reordered target sequence. For example, the visual cue “312” instructed participants to reorder the sequence of the original tones (say, A-B-C) such that the third tone now came first, that is, C-A-B. Finally, after 5 s participants were presented with a second tone sequence and indicated by left or right click on a mouse whether it was a ‘match’ or ‘mismatch’ to the target sequence, that is, the correct manipulation of the encoded sequence. At the end of each trial, visual feedback indicated whether the response was correct.

An important aspect of the task design was the intermixing of two different types of trials: simple and manipulation trials. A simple trial is any trial wherein the visual instruction is “123”. In such trials, participants are not required to perform any mental reordering of musical tones; they must simply judge whether the second sequence is the same as or different from the encoded sequence. This melodic comparison does not involve working memory, so it can be classified as a short-term memory task. On manipulation trials, the visual instructions are not in numerical order, such as “213” or “321”, indicating that participants must perform a manipulation on the auditory information stored within short-term memory. The contrast between these two trial types allows us to identify neurophysiological properties that are unique to manipulation abilities.

There were 42 consecutive trials (14 simple and 28 manipulation) within one 10-minute run, and three runs within each 30-min testing session for a total of 126 trials. Within each run, the trials were presented in a fixed randomized order. The runs presented in the pre-training and post-training sessions were identical, while unique runs were generated for each training session (but were similar for all participants). The order of the runs in each session was counterbalanced across participants.

On pre-training and post-training sessions we administered a working memory task with identical design to the training task, except with different auditory stimuli, to assess evidence of near transfer. The stimuli used in this transfer task, which we refer to as the noise task (Supplementary Figure 1A) were comprised of 42 different sequences of three unfamiliar environmental sounds, or “noises”. Each noise sequence was comprised of 500 ms audio excerpts from a variety of categories including weather and machinery [material from Zatorre et al. (2004) and used in Malinovitch et al. (2023)]. The audio clips were time-reversed to render them less likely to elicit a verbal label and bandpass filtered (500 – 8000 Hz; high- and low-pass roll-offs of 6 and 12 dB/octave, respectively) to equate for overall spectral range and equalized for root mean square intensity. Note that there were only manipulation trials in this near transfer task, with 50% match trials and 50% mismatch trials.

An additional visual mental rotation task was administered after the auditory working memory task, exclusively in experimental sessions 1 and 7 to test for far cognitive transfer. Two three-dimensional asymmetrical forms, each comprised of 7–9 cubes as in Shepard and Metzler (1971) were displayed side by side on a computer screen. The 10 visual stimuli used in this task – three 7-cube shapes, six 8-cube shapes, and one 9-cube shape – were composed of white cubes on a black background (see Supplementary Figure 1B). The paired shapes were either identical or the left-hand shape was horizontally mirrored. Relative to the shape on the left, the right-hand shape was rotated around its own axis by either 0, 60, 120, 180, 240, or 300 degrees.

Each trial began with the presentation of a fixation cross for 250 ms, followed by the paired target image and rotated probe. Participants were instructed to indicate, using left or right arrow keys, whether the two forms were identical or mirror images of one another after mental rotation. No feedback was delivered, and subsequent trials occurred immediately after a response was recorded, or after a delay period of 8 s if no response was recorded. The task began with a practice phase comprising 10 trials for which responses were not recorded, followed by the experimental data collection phase: all 10 stimuli pairs were presented once at each of the six relative rotation angles for both mirror and non-mirror trials, for a total of 120 trials in random order. One full run of the task took a maximum 16 minutes to complete.

We used EEG data from the entire sample of 18 participants who completed the pre-training session to validate that parietal theta activity was associated with manipulation abilities in our dataset. Time-frequency maps were generated for correct manipulation trials and then baseline corrected and averaged over trials for each participant (see methods). The same procedure was applied to correct simple trials. We then subtracted the averaged simple trials from the averaged manipulation trials for illustration, to isolate the activity elicited by mentally reordering the auditory stimuli and averaged the difference maps across all participants. In the resulting averaged difference map shown in Figure 1B, a burst of theta band activity coincided with presentation of the visual cue and was sustained until presentation of the test auditory stimuli. The corresponding scalp topography and source reconstruction for the manipulation time period (5–10 s) show prominent theta power (4–8 Hz) over the posterior parietal cortex and frontal cortices. Note that this difference was significant as illustrated in the scalp topography in Figure 1B (cluster corrected, alpha = 0.05, frontal cluster, k = 7, p = 0.01, parietal cluster, p = 0.004, k = 23).

Additionally, individual participants’ theta oscillatory power during this time period of 5–10 s was positively correlated with accuracy on manipulation trials (statistical peak r(17) = 0.85, p < 0.001). As shown in Figure 1B, we generated a time-frequency map of the correlation between oscillatory activity, averaged across pre-training manipulation trials, and d′ on pre-training manipulation trials. Highly behaviorally correlated theta (4–8 Hz) power during the manipulation period (5–10 s) was concentrated in parietal electrodes.

A two-tailed Mann-Whitney U-test indicated that rhTMS and sham control groups did not differ in terms of the discriminability index (d′) on manipulation trials at pre-training (U = 21, p = 0.71). Similarly, there was no group difference when looking at simple trials (U = 11, p = 0.1). We plotted the average d′ of both groups in each experimental session (Figure 2).

To compare the rate of learning between groups, we generated a simple linear fit (linear regression) of each participant’s accuracy (d′) in training sessions 2–6. We then extracted the slope of the linear fit for each participant (Figure 2), reflecting the rate of change in accuracy over sessions and conducted a two-tailed Mann-Whitney U-test that revealed a significant group difference for manipulation trials (U = 9, p = 0.05) but not simple trials (U = 16.5, p = 0.34).

We further investigated this finding with a one sample Wilcoxon signed rank test and found that participants stimulated with rhTMS had on average a manipulation trials learning slope significantly greater than zero (W = 26, p = 0.02), whereas participants in the sham condition had on average a slope of learning on manipulation trials that was not greater than zero (W = 9, p = 0.81). Neither experimental group had a slope greater than zero for the d′ on simple trials (rhTMS W = 14.5, p = 0.23; sham W = 13, p = 0.59).

We then performed a one-tailed Wilcoxon signed rank test of d′ at pre-training vs. post-training, which showed that the rhTMS group exhibited a significant post-training increase in performance on manipulation trials as compared to pre-training (W = 4, p = 0.05), but that the sham control group did not have significantly improved performance on manipulation trials at post-training compared to pre-training (W = 16, p = 0.66). The same analysis applied to performance on simple trials revealed a similar pattern, with the rhTMS group exhibiting an improvement at post-training (W = 0, p = 0.01) but not the sham group (W = 5, p = 0.08). Note however, that the contrast rhTMS group vs. sham group was not significant for post-training manipulation trials (U = 19, p = 0.54) or simple trials (U = 16.5, p = 0.33).

We did not find any group differences in behavioral performance on either the noise or mental rotation tasks. The data and statistics for each task are reported in Supplementary Figure 1.

We next compared task-relevant theta oscillatory activity between groups during the two training days with concurrent EEG recording (training days 2 and 4) because we were interested in evaluating the oscillatory entrainment in participants receiving real and sham rhTMS. Time-frequency maps were generated for correct manipulation trials, baseline corrected, then averaged across both training days for each participant. With these individual maps we constructed averaged time-frequency maps for both the rhTMS group and the sham group (Figure 3A).

To evaluate group differences in theta oscillatory power within the manipulation period of auditory working memory task performance, we conducted a one-tailed t-test with cluster correction for the difference in the theta (defined as the 4–8 Hz frequency band) oscillatory activity, over the time window of 5–10 s in correct manipulation trials from sessions 3 and 5, between rhTMS and sham groups. A significant cluster of theta activity was revealed at the sensor (cluster corrected, alpha = 0.05, k = 8, Figure 3B) and source levels (cluster corrected, alpha = 0.05, p = 0.05, k = 3308 vertices, Figure 3B) overlying the left fronto-parietal network.

Next, we tested our main hypothesis that combined working memory training and rhTMS would induce a lasting enhancement of task-relevant oscillatory dynamics, specifically theta oscillatory power in the fronto-parietal network. For illustration, we performed a simple subtraction of individual pre-training manipulation trial time-frequency maps from the corresponding post-training maps; the resulting contrast maps for each group are displayed in Figure 4A.

We then measured the change in theta oscillatory activity by conducting a paired one-tailed t-test with cluster correction contrasting theta (defined as the 4–8 Hz frequency band) oscillatory activity, over the time period of 5–10 s in correct manipulation trials, between post-training and pre-training sessions. A significant cluster of theta activity was revealed in left frontal cortex at the sensor (cluster corrected, alpha = 0.05, p = 0.01, k = 7) and source level (cluster corrected, alpha = 0.05, p = 0.035, k = 1898) for the rhTMS group (Figure 4A, left panel), whereas no significant clusters of theta activity appeared for the sham group (Figure 4A, right panel).

Finally, we looked for group differences in oscillatory activity during task performance in the post-training session. Figure 4B shows the time-frequency map produced by subtracting the sham group’s post-training manipulation activity from the rhTMS groups. An independent samples one-tailed t-test with cluster correction for the difference in post-training theta (4–8 Hz) activity from 5 to 10 s in correct manipulation trials, between rhTMS and sham groups was significant for clusters overlying distributed frontoparietal regions (cluster corrected, alpha = 0.05, p = 0.01, k = 30, Figure 4B).

We first confirmed (Figure 1) that frontoparietal theta activity was indeed a useful marker of working memory function in our training task and participant pool by showing that it is specific to the manipulation condition and is positively correlated with individual manipulation abilities. These data confirm and replicate previous findings that implicate the IPS in auditory manipulation (Foster and Zatorre, 2010; Foster et al., 2013) and fronto-parietal theta oscillations in working memory (Lisman, 2010; Sauseng et al., 2010; Fell and Axmacher, 2011; Albouy et al., 2017, 2018, 2022). Importantly, in each trial of our working memory task we used a visual instruction at 5 s to mark the precise time-point at which participants began to perform the mental manipulation of auditory stimuli, and applied rhTMS accordingly during the trial time window from 5 to 7 s, wherein the dorsal pathway is selectively engaged by endogenously generated oscillatory activity within the theta frequency band. This element of training task design is critical to our central hypothesis, that repeated sessions of rhythmic stimulation, during specific moments in task performance when the relevant networks are engaged, will produce cumulative and durable improvements in network function.

The majority of studies that have combined non-invasive brain stimulation and cognitive training with a similar gain-of-function aim to improve working memory, have used tDCS or tACS methods that are not tailored to endogenous brain rhythms evoked during the training tasks (see Polania et al., 2018, for review). It is difficult to compare these approaches to our findings given the fundamentally different mechanisms of TMS, but one idea that has emerged from the literature is that online brain stimulation, that is, stimulation applied during task performance, is more effective at enhancing or disrupting brain activity than passive brain stimulation protocols (Luber and Lisanby, 2014). We decided that there was minimal value in testing this idea in the present study with an additional control groups receiving rhTMS prior to task performance, and without cognitive training, given the evidence against efficacy of offline 5 Hz TMS in healthy adults (Beynel et al., 2019).

First, we observed a significant difference in learning slopes between experimental and sham groups during training, indicating that working memory training can be accelerated with information-based rhTMS (Figure 2). Additionally, we observed a significant behavioral improvement on the trained task between post-training and pre-training sessions in the rhTMS group but not in the sham group (Figure 2). These results suggest that the known performance-enhancing benefits of information-based rhTMS on auditory working memory can persist beyond the period of active stimulation when combined with cognitive training (Alekseichuk et al., 2016; Violante et al., 2017; Polania et al., 2018; Hanslmayr et al., 2019). Their specific novelty is in suggesting that those behavioral improvements are cumulative with repeated sessions and can accelerate learning in a complex auditory working memory task.

The slope of learning during cognitive training was different between groups for manipulation trials. Whereas the rhTMS group had a positive slope, the sham group’s slope was not significantly different from zero. This result was expected and can be explained by making a parallel with the results of Malinovitch et al. (2023) with the same task. In a sample of 28 young healthy individuals, the authors administered an adaptive cognitive training protocol consisting of 40 sessions of training on the same auditory working memory task used in the present study, only with the number of tones in each sequence increasing as participants accuracy increased, beginning with three tones and up to a maximum of 8 tones per sequence. They found that no participants exhibited an improvement on reordering three-tone sequences after the first 5 sessions. Additionally, as the number of tones in a sequence increased, the error rate decreased due to alternative strategies being employed that reduced the cognitive demand for true manipulations within auditory working memory. These findings were used to inform the design of our training task and training protocol, such that we chose to use only sequences of three tones and to limit the number of training sessions to 5, in order to test our hypothesis that information-based rhythmic stimulation would increase the rate of learning.

For learning on simple trials, neither group had a slope that differed significantly from zero. This result is also in line with the previous findings of Albouy et al. (2017) insofar as 5 Hz rhTMS specifically enhances performance on manipulation but not simple trials. However, the most likely explanation of the absence of learning on simple trials is related to ceiling effects. It is generally accepted that a range of d′ values between 0.5 and 2.5 is desirable for avoiding obscured results from either floor or ceiling effects, and these sensitivity values roughly correspond to an accuracy rate of 60 and 90%, respectively (MacMillan and Creelman, 2005). The average d′ of sham and rhTMS groups on manipulation trials at baseline was 1.53 and 1.91, respectively, whereas the average d′ on simple trials was higher at 2.61 for sham and 3.3 for rhTMS groups. With such high performance in both groups, there is little room for meaningful improvement.

Unexpectedly, the benefits of rhTMS to behavioral performance did not appear to transfer to two untrained cognitive tasks. At post-training, the rhTMS group did not perform better than the sham group on either the auditory working memory task with noises, or the visual mental rotation task (Supplementary Figure 1). After five sessions of training on the main task, neither experimental group displayed improvement on the untrained noise task. Given that the design of the noise task was identical to the main task apart from the stimuli used, we expected to observe some near transfer of learning, specifically for manipulation of order within the auditory domain. We did however find some evidence of what may be construed as far cognitive transfer to the visual mental rotation task from the auditory task, albeit not as a function of brain stimulation, since both the rhTMS and sham groups displayed significantly improved performance at post-training compared to pre-training, even though they did not practice the visual task. These results could be related to the inherent difficulty of the two tasks; it may require more sessions of cognitive training to see transfer to the noise task, or to see a group difference in transfer to the rotation task. Individual variability in frontoparietal network connectivity could also explain differences in learning generalization to untrained tasks, as in Wards et al. (2023).

By performing simultaneous rhTMS and EEG recording, we were able to characterize the immediate consequences of rhythmic and sham TMS on oscillatory activity. As shown in Figure 3, and as expected, participants receiving real 5 Hz rhTMS had significantly greater entrainment of theta activity in the dorsal pathway than did participants in the sham TMS group. Considering these findings, we believe that our choice to fix the frequency of rhythmic stimulation at 5 Hz, unadjusted to each participant’s individual theta frequency, was justified based on previous literature (Albouy et al., 2017). Indeed, our results demonstrate that it is not required to set the rhTMS rate to each participant’s individual self-generated frequency in order to enable entrainment (Figure 3A) and observe behavioral effects (Figure 2).

Subsequent analyses on larger datasets could investigate potential behavioral correlates of specific theta band frequencies generated during task performance. There is currently a lack of evidence supporting different functional significance of high versus low theta band activity. Overall, these effects observed on theta power are consistent with previous studies showing upregulation of oscillatory activity at the target frequency (Thut et al., 2011b; Weisz et al., 2014; Romei et al., 2016). The basic principles of neural entrainment can explain how online rhTMS increases the oscillatory power of a neuronal population as the phase of its spiking activity becomes increasingly aligned to the stimulation frequency [see Box 1 and Figure 1 in Hanslmayr et al. (2019)].

As the requirement for alignment between the stimulation frequency and each participant’s preferred frequency tends to be reduced with increasing stimulus intensity (Glass, 2001), we purport that our choice to set rhTMS intensity at 60% was justifiable and effective in enabling entrainment despite variation in preferred frequencies. Although TMS intensity is often individualized to participants’ motor thresholds, our unadjusted stimulus intensity was chosen intentionally to ensure that we would be stimulating above threshold for all participants. Our choice to set the output at 60% was not arbitrary but based upon empirical data from the previous study by Albouy et al. (2017), which showed positive behavioral effects with the same stimulation parameters. What’s more, there is no guarantee that the stimulus intensity required for generating evoked potentials in the motor cortex is relevant for rhythmic entrainment of parietal cortex. Therefore, we believe that it would not be necessary to individualize the output based on motor thresholds or e-field simulations. The most interesting contribution of the present study to the oscillatory entrainment literature is the investigation of after effects that are discussed below.

The results displayed in Figure 4A contrasting theta oscillatory activity in the dorsal pathway during manipulation post- vs pre-training are an exciting reflection of the behavioral changes, with significant differences over time observed only in the rhTMS group. The EEG analysis revealed a significant increase in the 4–8 Hz frequency band in the post-training compared to pre-training recordings only in participants who received real 5 Hz rhTMS (Figure 4A). Furthermore, post-training theta-band activity was significantly different between the two groups in brain areas corresponding to the frontoparietal network (Figure 4B). Taken together, we can say that training with rhTMS caused an increase in theta power that outlasted the period of stimulation by 48–72 h. This evidence implicates neuroplastic mechanisms, specifically induced by rhTMS, underlying the prolonged up-regulation of oscillatory activity within the dorsal pathway.

Previous efforts to explore the relationship between non-invasive brain stimulation, neuroplasticity, and the enduring impact on neural circuitry, have largely focused on Hebbian mechanisms like long-term potentiation and long-term depression (Cirillo et al., 2017). There remains a lack of direct verification for the physiological consequences of TMS protocols for enhancing cognitive function in healthy subjects. While the complex cellular and molecular mechanisms of plasticity that are modulated by TMS, among other methods of non-invasive brain stimulation, are presently best elucidated by in vivo and ex vivo animal studies, we can obtain indirect, systems level evidence from human clinical studies, albeit with notable inter and intra individual variability (Jannati et al., 2023). This multidisciplinary research supports our finding that precisely timed pulses of magnetic stimulation and subsequent entrainment of neural rhythms can be combined with cognitive training to facilitate consolidation of learning in specific neural networks.

Drawing upon the literature, one possible explanation for how this might occur is that during training, the ongoing intrinsic theta rhythms become phase-locked to exogenous rhythmic stimulation and that this amplifies signals of cortical synaptic plasticity. By virtue of the strong resonant frequency, theta oscillatory activity is consistently boosted in frontoparietal brain regions during task performance. Over repeated trials and sessions, the enhanced synchronous cortical activations reinforce communication between frontal and parietal areas. Long-lasting changes to tissue microstructure thereby improve the efficiency of information flow along relevant white matter pathways, a process that would organically occur much more slowly in the absence of rhTMS.

We have shown here that cognitive training combined with rhTMS accelerates learning and potentiates endogenous theta oscillatory activity within the frontoparietal network. In healthy individuals tested 48–72 h after receiving a 5-day cognitive training program with information-based rhTMS, there is a sustained benefit to auditory working memory performance and increased frontoparietal theta oscillatory power during manipulation, compared to baseline measures. These changes were not observed in participants who received sham rhTMS. This finding is important in demonstrating both behavioral and neurophysiological after-effects of non-invasive brain stimulation in a small sample of healthy individuals.

Our preliminary data analyses have yielded encouraging results in a sample which, while underpowered, establishes a proof-of-concept that contributes new evidence to the cognitive neuromodulation literature. The main limitation of the current work is due to the premature suspension of data collection in compliance with public health measures at the onset of the SARS-CoV-2 pandemic. With a sample size of 7 participants per experimental group, we were not able to detect a significant between-group difference in working memory abilities between stimulation and sham control participants in the post-training session, despite finding a significant within-group improvement in d′ between pre- and post-training sessions only for the rhTMS group. Moreover this small sample size did not allow us to investigate the relationships between individual changes in behavioral performance and changes in oscillatory activity. Based on our observed effect sizes however, we can estimate the sample size necessary to draw conclusions on the behavioral after-effects of rhTMS with cognitive training, compared with cognitive training alone, which we hope will be useful for future studies.

We can estimate the effect size of our rhTMS intervention by calculating Cohen’s d for the group difference on training day 5, where we observed peak performance for the rhTMS group. Given the stimulation group mean d′ of 2.4 and the control group mean d′ of 1.4, with standard deviations of 0.9 and 1.1, respectively, Cohen’s d is 0.995, which can be interpreted as a large effect size. With this effect size estimate, the sample size would need to be 32 (17 participants per group) in order to detect an effect of stimulation, if present, using a two-tailed t-test with α = 0.05 and power = 0.8 (Faul et al., 2009).

As an additional consequence of the SARS-CoV-2 pandemic, we were unable to invite participants back to the lab for longer-term follow-up of behavioral effects. With post-training data acquired ∼3 days after the last rhTMS session, we have assessed relatively short-term effects on auditory working memory and oscillatory dynamics. However, by demonstrating that oscillatory entrainment can be observed beyond the immediate post-stimulation period, we provide an evidence base for future studies to systematically evaluate the duration of such after-effects.

Future studies could also benefit from performing computational simulations of the rhTMS-induced electrical field (Thielscher et al., 2015; Saturnino et al., 2019) to precisely map which areas of the parietal cortex are most strongly stimulated in each participant during training. Given that the IPS lies adjacent to and deep to the gyral crowns of the superior and inferior parietal lobes, it is possible that these superficial regions receive more intense stimulation than the sulcal cortex (Thielscher et al., 2011; Kuhnke et al., 2020). Modeling the direct effects of rhTMS-induced electrical fields by simulation would be potentially useful as a complement to the mixed effects that we have shown by source reconstruction of simultaneous TMS-EEG recordings. With both simulations and EEG data, future studies could delineate the extent to which theta activity in IPS and surrounding regions is modulated directly by rhTMS versus through communication within the dorsal manipulation network. Moreover, by comparing the strength of association between both simulated and recorded electrical fields and behavioral performance, such studies could investigate the relative contribution of central and widespread network nodes to learning (Hartwigsen, 2018).

We have demonstrated that our combined training and rhTMS protocol is feasible and tolerable for young adults without any neurologic or psychiatric conditions. Larger studies should also be appropriately powered to investigate sex-based differences in behavioral and electrophysiological results. Sex differences in working memory, particularly visuospatial rotation, have been extensively researched, and there is a large body of literature to support the association of male sex with higher proficiency in visuospatial rotation (Gurvich et al., 2020). It is plausible that biological sex is associated with neurodevelopmental differences in the robustness of connectivity within networks relevant to our cognitive training task (see Moore and Johnson, 2020) and therefore that sex may mediate response to neuromodulation, as in Weller et al. (2023). As a potential clinical tool for cognitive rehabilitation, it is imperative to ensure that our brain stimulation and training protocol is optimized to be effective for male and female patients.

In a patient population with deficits in working memory, we might expect to see different, perhaps more pronounced effects from theta rhTMS interventions, reflecting impaired functional or structural connectivity and network dysregulation (Grover et al., 2023). By the same token, we would expect improvement to be confined only to patient populations in whom the relevant white-matter pathways in the auditory dorsal stream are at least partially preserved. Identifying electrophysiological biomarkers for treatment response will also be important for identifying individual patients more likely to benefit from stimulation and training interventions (see Andrade et al., 2023), thereby informing selection of participants for future trials and maybe one day providing personalized stimulation parameters optimized to individual network dysfunction.