Insular Cortex Modulation by Repetitive Transcranial Magnetic Stimulation with Concurrent Functional Magnetic Resonance Imaging: Preliminary Findings

1. Introduction

The insula is a deep structure located behind the frontal, parietal, and temporal opercula. Anatomically, it is divided into two portions by the central insular sulcus: the anterior insula (aI) and the posterior insula (pI) [1,2]. Functionally, the insula is divided into three subregions: the mid-posterior insula, and the ventral and dorsal aI. The insula has been implicated in a variety of functions, including autonomic and vestibular functions, interoception, sensory, cognitive, and affective processes [3,4,5,6,7,8]. The mid-posterior insula is connected to primary and secondary somatosensory cortices, whereas the dorsal aI is connected to the dorsal anterior cingulate cortex (dACC) and prefrontal cortex (PFC). The ventral aI shows functional connectivity with the inferior frontal gyrus and the temporal lobe including the amygdala [9].

Research suggests that the insula plays a key role in the regulation of appetite and eating behavior [3,10], as well as interoceptive awareness, including the perception of both hunger and satiety [11]. Functional magnetic resonance imaging (fMRI) studies have shown altered activity in the insula in obese patients [12,13,14]. Studies using electrocortical stimulation in epileptic patients undergoing neurosurgery showed that stimulation of the mid-insula induced gustatory hallucinations, such as acid and metallic tastes [15,16,17]. Furthermore, the insula is involved in the modulation of pain [18,19,20,21,22,23,24,25,26,27]. In individuals with chronic pain conditions, fMRI studies have shown altered activity in the insular cortex [28,29,30,31]. These changes may reflect maladaptive neuroplasticity in which the brain's response to pain is exaggerated or misregulated, contributing to persistent pain [32]. Two studies using intracranial neuronal recordings showed that the earliest response to painful laser stimulation occurs in the pI [33,34]. Low-frequency electrocortical stimulation of the pI elicited pain symptoms [21], whereas subthreshold high-frequency electrocortical stimulation transiently elevated temperature pain thresholds [35,36].

Repetitive transcranial magnetic stimulation (rTMS) is a non-invasive neurostimulation technique that appears to be promising for the treatment of several medical conditions. Technically, rTMS modulates neuronal activity by inducing electrical currents through time-shifting magnetic field pulses [37]. High-frequency rTMS has been widely used in the treatment of depression, mainly targeting the left dorsolateral PFC (DLPFC) [38,39,40,41]. It has also been shown to be safe and potentially effective in the modulation of craving, alcohol and cigarette consumption, and decision-making in addiction when applied to the DLPFC, medial PFC, and dACC [42,43,44,45,46,47,48,49,50,51,52,53]. A recent review reported significant analgesic effects of motor cortex rTMS in neuropathic pain, supported by 14 randomized placebo-controlled trials involving approximately 750 patients [54]. The pI is a promising target, but current evidence remains limited. Finally, a systematic review found that high-frequency rTMS targeting the bilateral DLPFC and the insula was associated with the greatest reduction in body mass index [55].

To date, only a few studies have examined the insula as a therapeutic target, but results are promising. Spagnolo and colleagues [56] reported that low-frequency deep rTMS at an intensity of 120% of the individual’s motor threshold (MT) over the right aI, using an H8 coil, was safe but did not affect performance on behavioral tasks, highlighting that high-frequency stimulation may be required to effectively activate the insula. In 2017, a clinical trial explored the potential of targeting the insula in the treatment of addiction. The study aimed to measure changes in dopamine levels using Positron Emission Tomography (PET) with [11C]-(+)-propyl-hexahydro-naphtho-oxazin. Participants underwent three PET scans after different rTMS sessions (sham, 1 Hz, or 10 Hz). The results showed that low-frequency rTMS targeting the insula significantly decreased dopamine levels in the substantia nigra, sensorimotor striatum, and associative striatum [57]. Dinur-Klein and colleagues [58] showed that high-frequency deep rTMS of the PFC and insula bilaterally reduced cigarette consumption and nicotine dependence, in contrast to low-frequency or sham treatments, and appears to be a promising treatment strategy for addiction. These findings are consistent with Ibrahim and colleagues, who reported the usefulness of combining deep insula rTMS with medication to improve smoking abstinence rates [59]. In addition, deep continuous theta burst stimulation of the right operculo-insular cortex selectively impaired the perception of thermonociceptive input from Aδ-fibre thermonociceptors, without affecting the perception of innocuous warm, cold, or vibrotactile sensations [60]. Another study reported subjective changes in cold perception following rTMS over the posterior-superior insula at a frequency of 10 Hz with an intensity at 80% of MT [61]. In addition, deep posterior-superior insula rTMS was associated with a significant reduction in pain intensity in refractory peripheral neuropathic pain [62]. Finally, a pilot study suggested that deep rTMS over the insula was safe, effective, and well-tolerated in patients with anorexia nervosa [63]. While these study results are promising, a common limitation is the uncertainty regarding whether the stimulation effectively reaches the insula considering its deep location in the Sylvian fissure.

Our study aimed to assess the feasibility of modulating insular activity using high-frequency rTMS and to determine the optimal rTMS parameters required to effectively modulate insular activity. rTMS was applied directly within the MRI scanner, providing a direct measure of neuronal modulation. This simultaneous rTMS-fMRI integration allowed characterization of changes in local cerebral blood flow and oxygenation, which in turn modulated the blood-oxygenation-level-dependent (BOLD) imaging signal [64].

2. Materials and Methods

Participants: Ten healthy participants (five women), all right-handed, were recruited in this study. Inclusion criteria required that participants had no significant health problems, no diagnosis of neurological and/or psychiatric disorders, and no condition that would make MRI and rTMS unsafe (e.g., metallic implants, pacemakers). Prior to rTMS modulation, each participant underwent an individual T1-weighted MRI scan on a separate day to rule out any medical conditions that might affect participation. One participant was excluded due to structural brain abnormalities (SBJ8). Another participant initially agreed to participate in the rTMS but later declined (SBJ7). A technical issue with the coil prevented us from performing stimulation in one participant (SBJ10). We had to interrupt the stimulation in another participant because of severe pain in the right temple caused by the pulses administration from the TMS coil (SBJ5). Thus, a total of six participants received rTMS modulation over the right insular cortex with simultaneous fMRI recording. The study was approved by the CHUM ethics committee (2019-7917, 18.122). All participants gave informed consent prior to all study procedures.

Identification of the target: To target the insula using frameless stereotactic neuronavigation, the previously acquired anatomical MRI was integrated into the Brainsight system, which generates a 3D model of the brain (Brainsight system®, Polaris system, Rogue Research). Four parts of the insula were identified (i.e., the limen insula, the superior border/temporal operculum, the anterior border/frontal operculum, and the posterior border/parietal) (inspired by Ciampi de Andrade and colleagues [61]). Real-time tracking was used to monitor the position of the TMS coil over the participant's head while they lay on the MRI table. At this stage, the MRI table was not yet positioned inside the scanner. The coil was attached to the TMS support, which can be inserted into the MRI scanner during rTMS modulation (see Figure 1). We used three tracking tools: the subject tracker, positioned on the participant's forehead; the coil tracker, attached directly to the TMS coil to guide the coil handle relative to the magnetic field hotspot; and the pointer tool. The TMS coil was positioned on the right side for all participants. However, the exact placement of the coil was individualized for each participant, considering the unique anatomical landmarks of each subject and the limited angulation of the coil in the MRI. Once the coil was positioned over the insula, the subject was moved into the MRI scanner with the TMS coil attached to the support. Table 1 and Table 2 show the individual motor threshold (MT) and the targeted insular subregions.

rTMS procedure: We used the MRI-B91 coil with compressed air cooling (Magventure®, MagProX100, Denmark). This coil generates a magnetic field of 3 to 6 kT/s for deep stimulation (3-5 cm) at 100% stimulator intensity and is compatible with MRI. Individual MT was determined by stimulating the left primary motor cortex and identifying the minimum intensity required to activate the right abductor pollicis brevis muscle in at least 6 out of 10 trials. High-frequency stimulation was performed with 10 trains of 10 seconds each at 5 Hz, with a 50-second interval between trains. Each stimulation train consisted of 50 pulses, for a total of 500 pulses. Stimulation intensity was delivered between 60% and 75% of the individual MT, depending on the participant's tolerance for discomfort during repetitive pulses.

fMRI image acquisition: All participants underwent fMRI acquisition simultaneously with rTMS. Prior to this, a resting-state functional MRI (RS-fMRI) was performed without rTMS. A 3 Tesla whole-body scanner (Skyra, Siemens) with a 16-channel head antenna was used to acquire fMRI data. We used a T2*-weighted gradient echo planar imaging sequence (i.e., slices = 49, TR = 3000 ms, TE = 30 ms, flip angle = 90°; FOV phase = 100%, FOV read = 230 mm, matrix = 64 x 64 matrix, voxel resolution = 2.4 x 2.4 x 3 mm, slice thickness = 3 mm). Participants were instructed to keep their eyes open during the rTMS-fMRI acquisition.

fMRI data processing: The alignment between fMRI slices and rTMS was ensured by synchronizing the onset of rTMS with the fMRI acquisition. We precisely calculated the timing of each stimulation event relative to the fMRI recording. Several parameters were used to determine the onset of rTMS relative to the MRI scans, including the total number of slices, the repetition time (i.e., TR), the echo time (i.e., TE), and the duration of the stimulation trains. This approach allowed us to align rTMS events with specific fMRI slices. We took great care to minimize artifacts from the simultaneous use of TMS and MRI. However, our fMRI data were noisy due to artifacts when TMS pulses occurred. Therefore, we decided to restrict analysis to MRI slices obtained in between stimulations and exclude the MRI slices acquired during stimulations from further analysis. We used Statistical Parametric Mapping software (SPM12) on MATLAB (MATLAB software R2022b) for fMRI data processing. Motion correction was performed using SPM's Realign function, which registered all functional images to a reference image (i.e., the average of all images). The corrected images were visually inspected to confirm the effectiveness of the motion correction. Slice timing correction was then performed. The anatomical MRI was then normalized to the Montreal Neurological Institute (MNI) brain template. Coregistration was performed to align the functional images with the high-resolution anatomical MRI images of each individual subject. Finally, the coregistered functional images were normalized to the MNI standard brain template.

Statistical analysis: Statistical analyses were performed using SPM12 software and xjView toolbox (https://www.alivelearn.net/xjview/). Analyses focused on the fMRI images obtained between rTMS trains. Specifically, paired t-tests were performed to compare the BOLD signal between the average of all fMRI images obtained between stimulation trains to the RS-fMRI images (i.e., RS-fMRI images acquired without the rTMS procedure) using a p-value of 0.01. The interval between stimulation trains was defined as the period from 20 to 50 seconds after the onset of the rTMS trains. This was considered the optimal choice based on increased statistical power. We then used the xjView toolbox to show insular activation and deactivation patterns. The initial voxel-wise threshold was set at p < 0.001 uncorrected. Significant clusters were then identified using the False Discovery Rate (FDR) correction at q < 0.01, with a minimum cluster size threshold of 10 voxels. Due to the insufficient number of fMRI images per rTMS train, it was not possible to make comparisons between different stimulation trains or to visualize the time course of BOLD changes after rTMS. For better understanding, the results are presented on the MNI brain template.

3. Results

One subject's data was excluded from analysis due to significant noise contamination that compromised data quality (SBJ9). For the remaining five participants, we identified between one and two significant activation clusters per subject that survived FDR correction. These activations were characterized by both increases and decreases in BOLD signal, indicating neural responses to stimulation. Specifically, we observed activation in the left (L) middle/aI (SBJ1), L middle insula (SBJ1, SBJ3), L mid-inferior insula (SBJ4), L pI (SBJ6), and right (R) mid-inferior insula (SBJ3). Deactivation was observed in the L ventral aI/aI (SBJ2, SBJ4), and the R and L pI (SBJ4) (see Table 2, Figure 2). In addition, two participants experienced a metallic taste immediately after rTMS. Significant activations were also observed in other brain regions, including the frontal (L and R inferior frontal gyrus, L middle frontal gyrus), temporal (L superior temporal gyrus, L temporal pole), parietal (R supramarginal gyrus, R post-central gyrus), R posterior cingulate, L and R putamen, L parahippocampal gyrus, and L and R globus pallidus regions (see Supplementary Table A1).

Due to the variability in stimulation targets across participants, group analyses were not performed. Heterogeneity in target locations would have introduced significant variability, complicating group-level inferences and potentially masking meaningful patterns of activation.

4. Discussion

In this study, we investigated the feasibility of modulating insular activity using rTMS with concurrent fMRI recording. A single high-frequency 5 Hz rTMS session was administered to six healthy participants over the R insula. Concurrent fMRI revealed significant BOLD signal modulations characterized by increased activity (i.e., activation, positive BOLD signal) in the L middle insula, L and R mid-inferior insula, and L pI subregions. Significant decreased activity (i.e., deactivation, negative BOLD signal) was observed in the L ventral aI, and L and R pI. Two participants reported dysgeusia (i.e., metallic taste). High-frequency stimulation with the MRI-B91 coil over the insular subregions was generally well tolerated despite the limited sample size of this preliminary study.

Our findings provide neuroimaging evidence that rTMS can modulate the insula despite its relatively deep-seated location. Of course, we cannot exclude indirect neural activation and/or deactivation of the insula by rTMS given the dense connectivity of the insula with numerous cerebral regions, including the overlying opercula. For instance, intermittent theta-burst stimulation over frontal regions has been shown to indirectly suppress insula activation by modulating fronto-insular connectivity [65]. Addicott and al. [66] also reported an increase in resting-state functional connectivity between the right postcentral gyrus (PCG) and the left insula following both 1 Hz and 10 Hz rTMS over the right PCG. However, the fact that dysgeusia was reported by two participants suggests that the insula was directly stimulated [67]. The symptom of dysgeusia is consistent with previous electrocortical stimulation studies that have reported gustatory hallucinations, such as metallic taste, following middle insula stimulation [15,16,17]. The two participants who experienced gustatory hallucinations showed an increased BOLD signal in the middle insula (see Table 2). This finding supports the role of this insular subregion in the integration of taste-related sensory inputs [67,68,69,70]. The concordance between increased BOLD activity in the mid-insula and the occurrence of gustatory sensations suggests that the activation threshold of this region may be modulated by rTMS, potentially leading to transient changes in sensory perception, as observed in our study.

In our study, both BOLD signal activations and/or deactivations were observed among our participants. This may possibly be explained by differences in the targeted insular subregions. Furthermore, while low-frequency rTMS is widely recognized for its deactivation/inhibitory effects on neural excitation, the mechanism of action of high-frequency stimulation (i.e., ≥ 5 Hz) remains debated [57]. Indeed, this paradoxical inhibition may result from the activation of inhibitory interneurons, particularly fast-spiking GABAergic circuits, as well as homeostatic plasticity mechanisms that regulate cortical excitability. In addition, stimulation-induced modulation of functional connectivity may lead to downstream inhibitory effects in interconnected neural networks. Furthermore, these activations and/or deactivations in neural responses appear to be influenced by the length and intensity of the stimulation train, with longer and more intense trains potentially enhancing inhibitory effects through sustained interneuron activation and synaptic plasticity [71,72]. In addition, insular response could differ between individuals depending on baseline excitability [73]. Ko and colleagues [74] suggested that the inconsistent findings in the literature of activation or deactivation across studies may be due to the study populations (i.e., healthy vs. clinical), which may reflect differences in neurochemistry, structural integrity, and connectivity. Finally, it may be possible that 5 Hz rTMS falls into a "grey zone" between excitatory and inhibitory effects. Studies suggests that rTMS responses are frequency dependent, and certain frequencies, such as 5 Hz, may produce both excitatory and inhibitory results depending on the context, stimulation parameters, and neural circuits [72,75].

The fact that rTMS over the right insula resulted in the modulation of BOLD activity in both hemispheres is not necessarily surprising considering the connectivity between both insulae and the widespread connectivity of the insula to surrounding lobes [9,76]. In our study, the activated areas are known to be structurally and functionally connected to the insula (see Supplementary Table A1). It is well established that rTMS has widespread effects beyond the target region. Previous neuroimaging studies have reported that rTMS induces cortical activation in both the stimulated and non-stimulated hemispheres [77,78,79]. Negative BOLD responses in the contralateral hemisphere have been observed after both high-frequency [77] and low-frequency rTMS [80]. Future research should explore the neurotransmitter systems underlying these effects. In addition, it would be valuable to investigate the variability of these responses across different rTMS frequencies and intensities, as well as in different populations.

Choosing the optimal coil is critical for effective deep structure stimulation. A study conducted by Lu and Ueno [81] demonstrated that double-cone, H-coil, and HCA coils show significantly deeper field penetration compared to the conventional “figure of eight” (Fo8) coil, albeit at the expense of inducing higher and more widespread magnetic fields in superficial cortical regions. A double-cone and an HCA coil show a superior ability to stimulate deep brain regions compared to an H-coil. At a depth of 40-60 mm, the volume of brain stimulated above threshold by the H-coil is greater than that stimulated by the double-cone coil, suggesting that the stimulation focus of the double-cone coil is superior at this depth. Conversely, at a depth of 60-80 mm, the volume of brain stimulated above threshold by the H-coil decreases rapidly, while that stimulated by the double-cone coil increases. Currently, there is no consensus on the optimal coil for targeting the insular cortex, as some studies have used a double-cone coil (e.g., D-B80 butterfly coil) [61], an H-coil [56,57,58,59,63,82,83], or a Fo8 coil [66,84,85]. The stimulation intensity varied widely across these studies. Spagnolo et al. [56] reported that a superficial cortical intensity of 145% of MT was required to reach a depth of 4 cm below the scalp. However, such high intensity exceeds safety guidelines for insular stimulation [37,86]. Although some studies suggest that higher stimulation thresholds yield better results [57,85], we were unable to apply higher intensities due to the moderate pain induced by pulses at the temple in three participants. We had no choice but to use an MRI-compatible coil for the purpose of these experiments, and thus the intensity of the magnetic field was reduced compared to standard coils due to the thicker casing [87,88].

Several limitations need to be taken into consideration. First, the pulses induced by the coil caused a slight movement of the head during stimulation, making it impossible to obtain clear fMRI images simultaneously with rTMS. Consequently, our results are based on fMRI data acquired immediately post- and between trains of stimulation. High-frequency rTMS has been shown to modulate neuronal activity for up to one hour after stimulation. However, it remains unclear whether this post-stimulation activity mirrors the immediate effects observed during stimulation [89]. Second, we performed only a single high-frequency rTMS session and did not include low-frequency stimulation or a sham condition. Thus, we report the acute effects of rTMS over the insular cortex but cannot infer long-term impacts. Third, the simultaneous rTMS-fMRI recording had inherent challenges, including noise, limited coil positioning on the scalp, restricted movement, and limitations in the choice of coil and rTMS parameters. Even with these limitations, we are confident that our target was effectively stimulated, as evidenced by fMRI images and dysgeusia symptoms specifically associated with the insular cortex.

5. Conclusions

In conclusion, this preliminary work demonstrates that high-frequency rTMS can be applied to the insular cortex without significant adverse effects, and that it is possible to reach the insula with TMS using an MRI-B91 coil guided by frameless stereotactic neuronavigation. Our findings are highly relevant for future clinical applications, where insular modulation could potentially be used for therapeutic avenues such as obesity and chronic pain. Despite the small sample size, this study contributes to the growing body of literature on the effects of neuromodulation with rTMS. Further research needs to validate and extend upon these preliminary observations, including a larger sample size and objective assessments of changes in somatosensory, emotional, and cognitive functions.

Appendix A.1

Table A1. Regions showing significant changes in BOLD signal after rTMS trains over the insula.

Subjects	Side	Anatomic regions	MNI Coordinates			T-score	Cluster size (# of voxel)
			x	y	z
SBJ1	L	Inferior frontal gyrus	-30	28	-7	3.84	170
	L	Middle frontal gyrus	-38	40	-9	4.89	154
	L	BA 47	-33	31	-6	3.28	53
	L	Insula	-34	11	-3	3.17	23
	L	Putamen	-23	-3	-1	3.93	22
	L	Lateral globus pallidus	-19	-3	-6	3.52	13
	L	Insula	-41	-1	1	3.35	10
SBJ2	L	Insula/BA 13	-34	13	-9	3.27	67
	L	Temporal pole	-46	8	-8	3.32	50
	L	Superior temporal gyrus	-43	13	-9	3.32	45
SBJ3	L	Midbrain	-1	-31	-15	4.44	340
	L	Insula (middle)	-39	-1	-2	3.32	34
	R	Insula (middle-inferior)	39	-12	-3	2.73	47
	L	Parahippocampal gyrus	-15	-40	-5	3.06	212
	R	Putamen	27	-17	1	2.72	56
	L	Putamen	-28	-15	-2	2.76	57
	R	Globus pallidus	22	-17	1	2.7	67
	R	Supramarginal gyrus	37	-48	29	3.31	33
	R	Post-central gyrus	53	-17	28	2.92	26
SBJ4	L	Superior temporal gyrus *	-47	-11	-1	3.47	27
	L	Insula (middle-inferior) *	-35	-4	-7	2.36	20
	R	Inferior frontal gyrus	-43	20	-15	2.90	534
	R	BA 41	47	-36	13	2.95	86
	L	Superior temporal gyrus	-38	-40	5	3.02	358
	R	Insula (pI)	39	-25	4	2.18	131
	L	Insula (pI)	-36	-18	6	1.88	134
	L	Insula (aI)	-34	11	-14	2.56	189
	R	Globus pallidus	25	-12	1	1.91	81
SBJ6	L	Insula (pI)	-43	-25	16	3.22	54
	R	Posterior cingulate	5	-51	23	4.57	173
	L	Superior temporal gyrus	-37	-45	15	2.81	24

Clusters ≥ 10 voxels, FDR-corrected at q < 0.01 / q < 0.001 / q = 0.05 * (trend toward significance). pI: posterior insula; aI: anterior insula; L: left; R: right.

Table A1. Regions showing significant changes in BOLD signal after rTMS trains over the insula.

Subjects	Side	Anatomic regions	MNI Coordinates			T-score	Cluster size (# of voxel)
			x	y	z
SBJ1	L	Inferior frontal gyrus	-30	28	-7	3.84	170
	L	Middle frontal gyrus	-38	40	-9	4.89	154
	L	BA 47	-33	31	-6	3.28	53
	L	Insula	-34	11	-3	3.17	23
	L	Putamen	-23	-3	-1	3.93	22
	L	Lateral globus pallidus	-19	-3	-6	3.52	13
	L	Insula	-41	-1	1	3.35	10
SBJ2	L	Insula/BA 13	-34	13	-9	3.27	67
	L	Temporal pole	-46	8	-8	3.32	50
	L	Superior temporal gyrus	-43	13	-9	3.32	45
SBJ3	L	Midbrain	-1	-31	-15	4.44	340
	L	Insula (middle)	-39	-1	-2	3.32	34
	R	Insula (middle-inferior)	39	-12	-3	2.73	47
	L	Parahippocampal gyrus	-15	-40	-5	3.06	212
	R	Putamen	27	-17	1	2.72	56
	L	Putamen	-28	-15	-2	2.76	57
	R	Globus pallidus	22	-17	1	2.7	67
	R	Supramarginal gyrus	37	-48	29	3.31	33
	R	Post-central gyrus	53	-17	28	2.92	26
SBJ4	L	Superior temporal gyrus *	-47	-11	-1	3.47	27
	L	Insula (middle-inferior) *	-35	-4	-7	2.36	20
	R	Inferior frontal gyrus	-43	20	-15	2.90	534
	R	BA 41	47	-36	13	2.95	86
	L	Superior temporal gyrus	-38	-40	5	3.02	358
	R	Insula (pI)	39	-25	4	2.18	131
	L	Insula (pI)	-36	-18	6	1.88	134
	L	Insula (aI)	-34	11	-14	2.56	189
	R	Globus pallidus	25	-12	1	1.91	81
SBJ6	L	Insula (pI)	-43	-25	16	3.22	54
	R	Posterior cingulate	5	-51	23	4.57	173
	L	Superior temporal gyrus	-37	-45	15	2.81	24

Clusters ≥ 10 voxels, FDR-corrected at q < 0.01 / q < 0.001 / q = 0.05 * (trend toward significance). pI: posterior insula; aI: anterior insula; L: left; R: right.