Lateral prefrontal stimulation of active cortex with theta burst transcranial magnetic stimulation affects subsequent engagement of the frontoparietal network

Abstract

Background:

A critical unanswered question for therapeutic TMS is what patients should be doing during treatment to optimize effectiveness. Here, we address this lack of knowledge in healthy subjects, testing the hypotheses that stimulating left dorsolateral prefrontal cortex (dlPFC) while participants perform a working memory task will provide stronger effects on subsequent activation, perfusion, connectivity and performance than stimulating resting dlPFC.

Methods:

After a baseline fMRI session to localize dlPFC activation and the associated frontoparietal network (FPN) engaged by an n-back task, healthy participants (n=40, 67.5% female) underwent three, counter-balanced sessions, separated by several weeks, in which they received intermittent theta burst stimulation (iTBS), followed by MRI scans as follows: 1) iTBS to the dlPFC, while resting passively (PASSIVE); 2) iTBS to the dlPFC while performing the n-back task (ACTIVE); and 3) iTBS to a vertex site, not engaged by the n-back task, while resting passively (CONTROL).

Results:

We found no difference in n-back performance between the three conditions. However, FPN activation was reduced while performing the n-back task in the ACTIVE condition relative to the PASSIVE and CONTROL conditions. There was no differential activity in the FPN comparing PASSIVE to CONTROL conditions, i.e. no effect of the site of stimulation. We found no effects of state or site of stimulation on perfusion or connectivity with the dlPFC.

Conclusions:

In this study, the state of the brain while receiving iTBS affected FPN activation, possibly reflecting more efficiency of FPN network activation when subjects were stimulated while engaging the FPN.

Introduction

What should patients with depression be doing while they receive repetitive transcranial magnetic stimulation (rTMS) to treat their disorder? Very little data exists to guide practitioners as to the appropriate mental state to optimize the therapeutic response. Psychotherapy has been delivered during rTMS for depression (1) and post-traumatic stress disorder (2), but it is not clear how, or if, the administration of therapy is modulated by stimulation, and stimulating the brain in the context of therapy can have negative effects (3). Stimulating the brain during cue exposure has been studied in substance use disorders by targeting regions activated by craving cues (4, 5), and rTMS during exposure to smoking cues has been cleared for the treatment of nicotine addiction (6). Based on studies showing abnormal activity in the medial prefrontal cortex in obsessive-compulsive disorder, stimulating this region with high-frequency deep rTMS (dTMS) during exposure to ritual-inducing stimuli reduced symptoms (7) and this pairing of dTMS with exposure therapy now has FDA clearance. However, many questions exist about the optimal mental state for patients receiving rTMS for depression.

Most therapeutic uses of rTMS do not systematically control the mental state of subjects during stimulation, despite the evidence that the state of neurons before and during stimulation affects how those neurons respond to stimulation. The phenomenon known as ‘metaplasticity’ refers to neural plasticity modulated by prior activity in a neuron (8), and an analogous process may occur with rTMS (9, 10). Extracellular recordings in animals have shown that increased visual cortical activity during excitatory rTMS leads to greater post-rTMS activity (11). From the earliest days of TMS research in humans, it was noted that stimulation of motor cortex during active muscle contraction increased the size and number of descending volleys compared to stimulation when the hand was at rest (12). Numerous examples of the state-dependency of stimulation have been described. For example, activating the ipsilateral hand during stimulation alters the response and coupling in the contralateral cortex during TMS applied to premotor cortex (13). Directing attention to the contralateral hand during 5 Hz rTMS leads to larger motor-evoked potential (MEP) increases than when attention is directed to the ipsilateral hand (14). Experimentally manipulating behavior prior to rTMS has been used in brain mapping studies to systematically alter the effect of stimulation on subsequent functions of perception and cognition (15, 16). Global changes in brain state, such as sleep, have demonstrated large effects, reducing the propagation of a single TMS pulse across the cortex when measured by surface electroencephalography (17). In the prefrontal cortex, where therapeutic rTMS is applied for the treatment of Major Depressive Disorder (MDD), there are multiple studies examining the effects of ‘offline’ (rTMS applied before brain activity and/or behavior is measured) and ‘online’ (rTMS applied while brain activity and/or behavior is measured) (18-22). What has been relatively unexplored is how the state of the prefrontal cortex – whether active and engaged in a task or resting idly – interacts with the effect of rTMS on subsequent behavior and brain activity -- a critical question for optimizing brain state for therapeutic rTMS.

The present study sought to address this question in healthy subjects after a single administration of intermittent theta burst stimulation (iTBS) TMS. iTBS increases the excitability of motor cortex for up to 90 minutes following stimulation (23, 24), and it is also an effective antidepressant treatment when delivered to the dorsolateral prefrontal cortex (dlPFC) over multiple sessions (25, 26), We asked whether iTBS to the dlPFC while subjects were engaging that region with a task, compared to idly resting, would differentially affect subsequent activity in the associated frontoparietal network (FPN). To engage the dlPFC, subjects performed an n-back working memory task, a standard task requiring cognitive control (27). It is impaired in multiple neuropsychiatric conditions, including depression (28), and it is carried out in the FPN (27, 29, 30). As iTBS to the motor cortex causes increased excitation in the motor cortex (23, 24), and studies have shown increased cerebral blood flow in the motor cortex when excitatory stimulation was applied during motor activity (31), we predicted increased FPN activation to the 2-back task after receiving iTBS to dlPFC. Because increased neural activation in dlPFC with brain imaging has been associated with better working memory performance (32, 33), we also predicted improved n-back performance with stimulation of activated dlPFC. This prediction was based on a meta-analysis of offline, excitatory rTMS to the dlPFC, demonstrating improvements in n-back performance (34). Furthermore, we sought to demonstrate the effects of dlPFC stimulation, compared to stimulation of a control region, not engaged by the task, on the subsequent activity and performance of the task.

Hypotheses for this study were pre-registered in clinicaltrials.gov (NCT#: NCT04010461) and the protocol was previously published (35).

Methods

Participants

Healthy participants were recruited from community advertisements and an online research registry. Eligibility was confirmed by screening assessment (in person or virtual) after an explanation of study procedures and informed consent was obtained, verifying the absence of any lifetime history of mental illness and contra-indications to study procedures. Of 59 subjects consented and screened, 53 were enrolled, of whom 40 participants completed all study procedures (see CONSORT Figure S1). The mean age was 25.5 years (SD: 7.4), and 27 were female. Participants self-identified as White (n=24), Black (n=5), Asian (n=8) and multi-racial (n=3); 3 participants identified as Hispanic. All study procedures were approved by the University of Michigan Institutional Review Board.

Study procedures -- Overview

After consent and screening, participants were introduced to the n-back task and given practice trials, including practice in a mock MRI scanner to accommodate them to the scanning environment. In the first of four, separate MRI sessions (BASELINE), participants performed the n-back task during fMRI Blood Oxygenation Level Dependent (BOLD) acquisition (Figure 1). Following this initial MRI session, active motor threshold (AMT) was determined. Offline examination of activity during the n-back task allowed for the selection of personalized dlPFC and vertex targets. The subsequent three MRI sessions, each on separate days, began with iTBS delivered in counter-balanced order across the three sessions: PASSIVE (iTBS to dlPFC while participant rested), ACTIVE (iTBS to dlPFC while participant performed the n-back task), and CONTROL (iTBS to vertex while participant rested). Each of these three iTBS sessions was immediately followed by identical MRI sessions which included, in the following order: resting arterial spin labeling perfusion scan, BOLD scan during n-back task, resting state BOLD scans (2 runs). Additional details for methods are in the Supplementary Materials.

Figure 1.

Overview of experimental design. Because of COVID protocols, most of the consent and screening occurred virtually, before the in-person visit to the clinic for n-back practice, which included a mock MRI scanner session. ASL = arterial spin labeling, BOLD = blood oxygenation level dependent, iTBS = intermittent theta burst stimulation, dlPFC = the left dorsolateral prefrontal cortex.

n-Back task

To engage the dlPFC and identify stimulation targets, we used a 2-back version of the n-back task, which robustly activates the FPN, including the dlPFC (36-39). The primary contrast to activate the FPN was between the 2-back and 1-back conditions.

iTBS and neuronavigation procedures

TMS was delivered through a MagPro X100 with a MagOption magnetic stimulator and a 90mm figure-8 coil (MC-B70, MagVenture Inc.). The iTBS sessions used 3 pulses of stimulation at 50 Hz, repeated every 200 ms, for 2 sec trains, repeated every 10 sec, for a total of 600 pulses in 190 sec (24). Stimulation was delivered at 80% of AMT, within consensus recommendations for safety (40) and sufficient to provide excitatory effects on the MEP after motor cortex stimulation for up to 60 minutes (23).

To determine the site for stimulation, we used neuronavigation with structural and functional information to accommodate individual variability. From the BASELINE MRI session, a difference map of the 2-back minus 1-back conditions in native space revealed activation and deactivation for each subject, enabling identification of the dlPFC target and the ‘Vertex’ control region that was not associated with this contrast. Participants received iTBS to the target appropriate for the condition. During the PASSIVE and CONTROL sessions, subjects viewed a fixation cross on a screen approximately 50 cm away. During the ACTIVE condition, they performed the 2-back condition of the n-back task, while they received iTBS to the dlPFC. Immediately after completing the iTBS session, participants moved across the hall to the MRI suite..

T2* BOLD processing and analysis

MRI data analysis and quality control used established, open-source routines. First-level analysis used the general framework of the modified General Linear Model (41), implemented in SPM12. For analysis of group effects, second-level, between-subject analyses on normalized images of the 2-back minus 1-back beta estimate from the first level were entered into regression models, with mean framewise displacement (meanFD)(42) as a co-variate of no-interest, to predict BOLD activation for each day and for the within-subject contrasts between CONTROL, PASSIVE and ACTIVE days. Statistical thresholding used an initial height threshold set at p < 0.001 and peak and cluster corrected significance levels were obtained (false discovery rate [FDR] and family-wise error [FWE] corrected). A small-volume correction was obtained from a mask of the 2-back minus 1-back contrast during the BASELINE session. Set-level significance was also calculated with the SPM12 package.

Analysis of resting-state connectivity was performed using the CONN toolbox (43). Multiple regression was applied to normalized, realigned, 4-D image sets, de-noised, band pass filtered at 0.008-0.09 Hz, and motion scrubbed (frame displacement > 0.5 mm (42)). Two methods for seed placement (6-mm radius sphere) were used: 1) seed placed on each participant’s locus of dlPFC stimulation; 2) seed centered on MNI coordinates (−31, 6, 63), approximating the center of the stimulation foci across individuals. Statistical inference was controlled as above, with the FPN mask as a small-volume correction.

Perfusion analysis

Measurement of cerebral perfusion was done with pseudo-continuous arterial spin labeling (pCASL). The resulting images were preprocessed using a combined preprocessing pipeline of scripts calling functions from FSL 6.0 (44) and SPM12. The steps included realignment, spatial normalization, and smoothing. Using the transformed coordinates of the site of stimulation for dlPFC and vertex, spherical regions of interest (8 mm radius) were placed on the perfusion images and values extracted for analysis. For a secondary analysis, the FPN mask was also used to extract average perfusion.

Behavioral data analysis

For the analysis of n-back behavioral data, the primary performance measure was d-prime, which captures sensitivity to the target while controlling for response biases (45). A secondary measure was median reaction time (RT), which was most sensitive to TMS stimulation in a published study (46). Accuracy (fraction of hits and correct rejections over total trials) was also calculated. Because of a programming error, only 16 participants had usable data for the ACTIVE condition and 39 had usable data for the CONTROL condition. To maximize power with the reduced sample size, separate two (load) x two (session) repeated measures ANOVA, with Greenhouse-Geisser correction, were computed for the ACTIVE vs PASSIVE and PASSIVE vs CONTROL conditions. Statistical analyses were done in SPSS version 28.

Results

Behavioral Results

All participants performed the task at relatively high accuracy, and there were no significant differences between PASSIVE and CONTROL and between PASSIVE and ACTIVE sessions on d-prime, standard accuracy or reaction time (Table 1). Strong effects of load (1-back compared to 2-back) were demonstrated for all measures, and there were no significant interactions between load and session (p’s > 0.25).

Table 1:

n-back performance data

	Session				F-tests
					Passive vs Active		Passive vs Control
	Baseline (n = 40)	Passive (n = 40)	Active (n =16)	Control (n =39)	F	P	F	P
1-back					Effect of load
RT msec (SD)	523.8(111.4)	508.0(115.3)	528.0(110.6)	499.9 (107.4)	6.54	0.02	17.6	<0.001
Accuracy (SD)	97.7%(2.1%)	98.6%(3.0%)	98.1%(2.8%)	98.3% (3.0%)	17.4	<0.001	32.9	<0.001
d-prime (SD)	4.0(0.4)	4.2(0.5)	4.1(0.5)	4.1(0.6)	16.7	<0.001	57.1	<0.001
2-back					Effect of day
RT msec (SD)	580.8(185.1)	573.2(175.8)	572.1(135.4)	556.4(164.2)	3.96	0.06	0.17	0.68
Accuracy (SD)	96.8%(3.3%)	96.7%(3.9%)	96.2%(4.5%)	96.7%(3.6%)	0.38	0.55	0.21	0.65
d-prime (SD)	3.8(0.7)	3.8(0.7)	3.7(0.8)	3.8(0.6)	0.72	0.41	0.46	0.50

RT: reaction time

n-back activation results

For the n-back analysis, 3 subjects had poor image quality, preventing inclusion in the group analysis. For the 37 remaining participants, robust activation for the 2-back minus 1-back contrast was demonstrated across all sessions (Figure 2A). For planned contrasts between sessions, there were no significant differences between the CONTROL and PASSIVE conditions, which is to say no effect of the site of stimulation on n-back activation, either in the small volume correction for the FPN (Figures 2B and 2C) or in the corrected, whole brain voxel-wise analysis (Table 2). However, in the PASSIVE>ACTIVE contrast, assessing the effect of state during iTBS on subsequent activation, there were significantly more clusters of activation within the FPN network (set level: c = 4, p <0.001; Table 2) during the PASSIVE condition in the left hemisphere, but no significant differences for PASSIVE<ACTIVE, either in the FPN or in the whole brain analysis. Although not a primary planned comparison, we examined differences between the ACTIVE and CONTROL conditions, and found a similar pattern to the contrast of ACTIVE versus PASSIVE. More clusters appeared in the CONTROL condition in the FPN network (set level: c = 4, p <0.001; Table 2), also in the left hemisphere, but we found no significant differences for CONTROL<ACTIVE, either in the FPN or in the whole brain analysis. The eigenvariates for the 2-back minus 1-back contrast were extracted from the FPN region of interest mask, and planned contrasts for PASSIVE versus CONTROL and PASSIVE versus ACTIVE were not significant (respectively, t[35] = −0.08, p = 0.94; t[35] = 1.65, p = 0.11). See Figure S2.

Figure 2.

Activation maps for the n-back task with the 2-back minus 1-back contrast and contrasts between sessions. A) Orthogonal, ‘look-through’ representation of n-back activation from the BASELINE session, which was used to derive B) the frontoparietal network (FPN) region-of-interest mask used for the small volume correction. Contrasts between each session are depicted in C, D and E, in look-through views from lateral and anterior-posterior perspectives. Of note, the clusters within the FPN appear on the left side in D and E, which is the side of stimulation. *Significant (p < 0.001) at the set-level for number of clusters in the FPN.

Table 2:

n-back activation contrasts for FPN

Set-Level			Peak			Cluster
				FDR- corr	FWE- corr		FDR- corr	FWE- corr
c	p	Cluster coordinates: x, y, z	Z	p	p	Size	p	p
CONTROL > PASSIVE
		No suprathreshold clusters
PASSIVE > CONTROL
		No suprathreshold clusters
PASSIVE > ACTIVE
4	0.001	-−42, −47, 41	3.69	0.831	0.097	21	0.700	0.106
		−23, 8, 60	3.57	0.831	0.141	17	0.700	0.130
		−1, −66, 55	3.45	0.831	0.198	7	0.705	0.235
		−8, −61, 50	3.16	0.831	0.392	2	0.705	0.344
ACTIVE > PASSIVE
		No suprathreshold clusters
ACTIVE > CONTROL
		No suprathreshold clusters
CONTROL > ACTIVE
4	0.001	−32, 1, 48	4.03	0.376	0.035	48	0.250	0.027
		−32, −68, 41	3.67	0.425	0.113	12	0.786	0.170
		−47, 23, 36	3.49	0.568	0.192	4	0.786	0.301
		−18, −64, 53	3.19	0.893	0.395	1	0.786	0.403

Cluster coordinates from MNI atlas, defined by voxel threshold of p < 0.001 (uncorrected). Peak magnitude Z-score and cluster size with false discovery rate (FDR) and family wise error (FWE) corrected (-corr) probabilities (p)

Approximately half-way through data collection, it was discovered that the dlPFC activation focus was actually a deactivation focus for the first 18 subjects, which was corrected for the subsequent 19 participants (of the 37 analyzable participants). Each group (dlPFC iTBS applied to activation focus and iTBS applied to deactivation focus) was analyzed separately, and there were still no significant differences for the CONTROL-PASSIVE contrast (effect of site), in either group. The effect of PASSIVE > ACTIVE in the FPN was present for the activation focus, but not the deactivation focus, and only when the height threshold for cluster definition was lowered to p < 0.005. See Supplementary Materials for details of this analysis (Figures S4 and S5).

Resting state connectivity

For this analysis, 3 participants failed quality control, leaving 37 participants for the analysis. There were no significant differences in dlPFC connectivity between the CONTROL, PASSIVE and ACTIVE conditions for either the individualized seed placement or the seed placement in a common location across participants.

Perfusion (pCASL) results

Examination of cerebral perfusion from regions of interest placed at the individualized sites of stimulation showed no difference in perfusion between the PASSIVE and CONTROL conditions (t[35] = −0.36, p = 0.72) and between the PASSIVE and ACTIVE conditions (t[37] = −0.61, p = 0.54). Omnibus tests were also not significant for the CONTROL, PASSIVE, and ACTIVE conditions for the dlPFC (F[1.83, 64.2] = 0.10, p = 0.91) and vertex (F[1.52, 53.2] = 0.05, p = 0.95) sites. Similarly, average perfusion in the FPN mask region showed no difference between conditions (F[1.92, 67.1] = 0.22, p = 0.79). There was a nominal drop in perfusion from the BASELINE scan to subsequent scan sessions, for both the dlPFC and vertex sites, but these changes were not significant in repeated measures ANOVAs (see Figure 3A-C). For this analysis, 4 participants were had 1 unanalyzable session, due to quality control failures, leading to different numbers of participants for each comparison.

Figure 3.

Perfusion measured in A) the dlPFC and B) vertex regions showed no effect of stimulation on subsequent cerebral blood flow. There was a nominal drop in perfusion from BASELINE for both regions, although the ANOVAs missed significance in the dlPFC (F[2.49, 84.8] = 1.99, p = 0.13) and vertex (F[2.38, 80.91] = 1.43, p = 0.24). Perfusion averaged over the FPN region did not differ across sessions (F[2.75, 93.47] = 1.85, p = 0.14. Bl = BASELINE session; Ctrl = CONTROL session; Pas = PASSIVE session; Act = ACTIVE session.

Correlations between behavior and BOLD activation

We conducted post-hoc analyses between FPN activity during the n-back task (2-back minus 1-back), examining d-prime and reaction time during the high load, 2-back condition. For d-prime, we found significant, or near-significant, correlations with FPN activation (r’s = 0.27 – 0.58; see Figure 4) for all conditions, such that better performance was associated with stronger activation in the FPN. For reaction time, we found a significant, inverse correlation between reaction time and FPN activation during the BASELINE condition (r = −0.40. p = 0.02), i.e. faster performance associated with greater activation,

Figure 4.

Correlation of d-prime performance on the n-back task for each session with FPN averaged activation for 2-back minus 1-back contrast.

Discussion

Our results supported the prediction that iTBS applied to an active, engaged brain would affect subsequent brain activation, relative to iTBS applied to a resting, unengaged brain. However, the pattern of activation in the FPN was opposite what we had predicted, as stimulation during FPN engagement led to less BOLD activation during the n-back task performed ‘offline,’ after the iTBS session, compared to receiving stimulation at rest. While we did not find any differential effects of brain state during stimulation on subsequent, offline task performance, similar performance with lower activity could be explained as greater efficiency, as we discuss below. Surprisingly, we did not find that the site of stimulation (dlPFC versus vertex), when applied to the brain in a resting state, had any effect on subsequent activation. We also found no effects of brain state or site of stimulation on connectivity or perfusion, measured after the iTBS session. Despite these negative findings, our results suggest the importance of brain state on subsequent activation when stimulating the prefrontal cortex with iTBS.

We interpret the reduced activation during the ACTIVE condition, relative to PASSIVE, as evidence of greater efficiency of activation. We had predicted that n-back activation would increase after excitatory stimulation, as has been observed in the motor cortex (47, 48), although we also acknowledged that less activation and greater efficiency might be seen, instead (35). It is possible that the ACTIVE condition increased activity in the 1-back condition, leading to an apparent reduction in 2- minus 1-back activation, although because iTBS stimulation was applied only during the 2-back condition, we find this explanation less likely. We also predicted that offline n-back performance would improve, as others have found (34, 46), although some groups have not found improved n-back performance after multiple sessions of TMS (49, 50). Interestingly, Gaudeau-Bosma and colleagues reported reduced n-back activation and no change in behavior after 10 sessions of high-frequency rTMS. Reduced activation reflecting greater efficiency in executive function tasks has been put forth in the literature to explain decreased performance and increased frontoparietal activation in aging (51) and schizophrenia (52, 53), as well as decreased activation observed with task practice (54, 55). In our data, increased n-back activation was associated with better performance, as the scatterplots in Figure 4 show, consistent with prior reports (32, 33). This positive correlation between FPN activation and performance, along with the reduced activation observed after iTBS stimulation of active, engaged frontal cortex, suggests that the performance-activation relationship may have ‘shifted to the left,’ which is to say, the same level of FPN activation was associated with better performance. Our failure to find changes in performance with iTBS may have reflected a ceiling effect, as 2-back accuracy was quite high, and subjects had been extensively practiced in the task prior to the 3 experimental conditions, in order to stabilize performance for the experimental conditions. If participants had not been performing at their apparent maximum accuracy, our results may have turned out differently. In spite of this limitation, our results provide support for the conclusion that brain activity in the FPN interacted with iTBS stimulation to improve the efficiency of activation.

More generally, our results align with other work showing that effects of TMS stimulation are enhanced when applied to cortex engaged in a task. It is well known that stimulating finger regions of motor cortex when a person is contracting their fingers yields a lower motor threshold compared to stimulating motor cortex at rest (12, 56, 57). In motor cortex, activity of cortico-spinal cells has been associated with a reduction in cortical inhibition, making those cells more responsive to exogenous stimulation (57-59). A comparable process could be occurring in the frontal cortex, whereby iTBS stimulation might have induced neuroplastic changes in inhibitory interneurons, analogous to long-term potentiation (10, 24). However, this explanation does not fully account for improved efficiency and reduced BOLD signal, necessitating additional study. Importantly, differences in brain regions and activity, as well as the TMS protocol used, likely affect results on subsequent activity. For example, Silvanto and colleagues (60) showed that when subjects viewed visual motion while receiving inhibitory cTBS to direction-sensitive neurons in V1/V2, performance after stimulation was impaired for the direction-sensitive neurons that were not activated during the passive viewing task. In other words, neurons engaged by the task appeared to be ‘protected’ from inhibitory cTBS.

Our study was motivated by the question of the optimal brain state to receive therapeutic rTMS for depression. Our results suggest an answer, that an active cortex might be more receptive to TMS-induced changes. Not regularly engaging the dlPFC for therapeutic rTMS in depression may explain why beneficial effects on cognition have not been found (61), but important caveats should be noted. We do not know how long the effects of stimulation on efficiency would last, or the effects of repeated sessions of rTMS, as is the practice for therapeutic TMS. The 80% AMT intensity used for stimulation is below the threshold used for most therapeutic iTBS (26). Another important limitation of our study, besides the single session used to demonstrate effects, is the lack of a fourth condition (vertex stimulation during task) for a complete factorial design. Our use of personalized targeting based on functional activation patterns is not standard practice for the delivery of TMS for depression, although a recently FDA-approved accelerated rTMS protocol does incorporate personalized targeting with fMRI (25). Furthermore, results obtained in healthy individuals also might not translate to individuals with depression, where alterations in cortical excitability have been noted (62), and the results obtained with a cognitive task, rather than one targeting emotion regulation, might not translate well to depression treatment . In spite of these limitations, our findings support the idea that an active cortex might be a more receptive cortex to iTBS.

We were unable to demonstrate that the site of stimulation – dlPFC versus ‘vertex’ – affected subsequent activation, an unexpected null finding, for which there could be several explanations. It was notable that the magnitude of the activated BOLD signal in the FPN was similar for both the CONTROL and PASSIVE conditions, whereas the ACTIVE condition showed reduced activation in the FPN relative to both conditions. For this experiment, the effect of brain state when stimulated appeared stronger than where the TMS coil was placed. However, as we had a technical error leading to approximately half the subjects receiving stimulation over a deactivation region, and the other half over an activation region, we had less statistical power (only 19 subjects) to detect an effect of site. The effect of reduced power on sensitivity can be seen in our Supplementary Materials, where we saw no significant effects of state or site at the appropriate threshold for good Type 1 error control.

Another negative finding occurred with perfusion and connectivity. Resting state scans came at the end of the scan protocol, with the last of two, 8-minute scans finishing around ~50 minutes after the iTBS session. While data show that motor-evoked potential changes after iTBS persist for up to 1 hour (23), the duration of changes in frontal cortex changes might not persist as long. On the other hand, the perfusion scans occurred within 10 minutes after the participants completed iTBS. Prior work of acute TMS effects on perfusion have found acute effects of increased perfusion at the site of stimulation (47, 48), but the single iTBS session was apparently not enough to change measurable dlPFC perfusion ~11 minutes later in our study (see Supplementary Material). In general, the most consistent neuroimaging result observed in offline studies is that changes occur in regions beyond that stimulated, but anatomically and functionally connected (48, 63-66). A recent meta-analysis of TBS effects on connectivity found inconsistent effects when applied to the prefrontal cortex, although methods (sham control versus control region, sample size, analytic methods) differ widely (67), making comparisons difficult. We cannot rule out a type 2 error here, from undiagnosed technical issues, but it may also be the case that this single iTBS session to the prefrontal cortex does not have effects on perfusion or connectivity assessed at the site of stimulation.

Conclusions

The findings presented here add to the literature demonstrating that the state of the brain when receiving TMS affects how the brain responds to stimulation, focusing on prefrontal cortical activity, a site relevant for treating persons with depression. We have shown that iTBS stimulation of an active cortex appears to increase the efficiency of activation. While alternative explanations need to be considered, and the results cannot be directly extrapolated to persons with depression, they nevertheless provide some nominal support for the idea that an engaged cortex might be a better target than a passive cortex.