Individualized, connectome-based, non-invasive stimulation of OCD deep-brain targets: A proof-of-concept

non-invasively


Introduction
Obsessive-compulsive disorder (OCD) is a severe neuropsychiatric condition (Ruscio et al., 2010).Despite exhaustive use of cognitive-behavioral and pharmaco-therapy, an estimated 10-20 % of OCD patients still suffer from severe refractory symptoms (Denys, 2006).In the last decades, deep brain stimulation (DBS) has surfaced as a promising last-resort option for treatment-resistant patients, offering a significant improvement to their clinical symptoms (amounting to a ~47 % reduction in 50-60 % of patients), general level of functioning and quality of life (Blomstedt et al., 2013;Kohl et al., 2014;Alonso et al., 2015;Denys et al., 2020;Mar-Barrutia et al., 2021).By stereotactic electrodes placement at specific locations deep in the brain, the discharge of constant or intermittent high-frequency electrical pulses modulates pathological neural activity (Lozano and Lipsman, 2013;Ashkan et al., 2017).Although the exact mechanisms of action remain elusive, the distal effects of this highly focal stimulation on connected brain regions are acknowledged (Smith et al., 2020;Figee et al., 2013), and have prompted the idea that a common neuronal circuit might mediate the observed clinical improvement (Baldermann et al., 2021).The last years have thus seen a rise in the number of connectomic studies, showing how different DBS targets promote clinical efficacy via modulation of structurally or functionally connected networks (Baldermann et al., 2021;Li et al., 2020;Fridgeirsson et al., 2020;Barcia et al., 2019).
Notwithstanding its benefits, the risks inherently associated with an invasive procedure, the rare yet existent reported side-effects (Mar-- Barrutia et al., 2021;De Haan et al., 2015), and the still limited knowledge on individual target selection and indicators of clinical response (Tastevin et al., 2019), impede wider applications of DBS (Wu et al., 2021).Within this context, it is thus desirable to perfect less invasive treatment options.Transcranial magnetic stimulation (TMS) is by now an established and valuable alternative in the treatment of depression (Fitzgerald et al., 2016;Brunoni et al., 2017).In OCD, several combinations of stimulation targets (e.g., bilateral or unilateral dorsolateral prefrontal cortex, supplementary/pre-supplementary motor area, orbitofrontal cortex) and protocols (high frequency, low frequency) have been investigated using standard focal TMS coils (Zhou et al., 2017;Rehn et al., 2018;Fitzsimmons et al., 2022).Despite positive results have emerged, there is still limited evidence investigating the superiority of one specific target/protocol combination (Carmi et al., 2018).The use of TMS H-coils, specifically built to reach slightly deeper areas (~3-5 cm below the skull) at the expense of decreased focality, is on the other hand regulated.In 2019, a high-frequency deep TMS protocol stimulating the dorsal anterior cingulate cortex has received FDA clearance as adjunct treatment for refractory OCD (Carmi et al., 2018;Carmi et al., 2019).However, the lack of replication across wider cohorts and tailoring on individual patients is still recognized as a general shortcoming, stimulating further research into finding alternative approaches.
Similar to DBS' mechanisms of action, not only does TMS influence neural activity at the stimulation site, but also its effects extend to remote, connected regions (Bestmann et al., 2004;Denslow et al., 2005;Davis et al., 2017;Tik et al., 2017;Luber et al., 2022).In keeping with a mechanistic understanding increasingly relying on networks rather than regions, the fields of invasive and non-invasive brain stimulation have started to converge and inform each other's experiences.Particularly in the context of depression, a number of connectivity-based TMS approaches have emerged, where (frontal) stimulation targets are chosen based on functional (Fox et al., 2012;Fox et al., 2013;Siddiqi et al., 2020) or structural (Luber et al., 2022) connectivity to deeper structures (subgenual cingulate) key to the disorder pathology and to DBS treatment.Harnessing the established network effects of cortical TMS, such connectivity-based approaches might increase the likelihood of effectively influencing subcortical structures with a key impact on clinical symptoms.Given the accumulating evidence in the context of depression (Cash et al., 2021), the scope of such approach may conceivably widen to other psychiatric disorders.
In a proof-of-concept study with healthy volunteers, we investigate the feasibility and effect of a personalized, connectivity-based paradigm for TMS targeting that relies on the structural connectivity of known OCD DBS targets.In the DBS context, both white matter tracts (e.g., anterior limb of the internal capsule (Lam et al., 2009), inferior thalamic peduncle, median forebrain bundle) and gray matter structures (e.g., nucleus accumbens [NAc], subthalamic nucleus [STN], bed nucleus of the stria terminalis) have been targeted (Denys et al., 2020;Sturm et al., 2003;Mallet et al., 2008;Jiménez et al., 2013;Coenen et al., 2017;Raymaekers et al., 2017) and resulted in comparable response rates (Kohl et al., 2014;Alonso et al., 2015).Independent studies have investigated the anatomical profile associated with successful DBS across stimulation sites, focusing on the ALIC/NAc and STN (see Baldermann et al., 2021 for a review).Results overall suggest the modulation of two pathways as potentially critical in mediating symptoms improvement: an hyperdirect connection of the medial and lateral prefrontal cortices to the STN, and a central subsection of the ALIC connecting the anterior thalamus/STN and the prefrontal cortex (Baldermann et al., 2021).Thus, although overall no superior stimulation site has been identified for DBS in OCD, we focus on the deep targets for which an OCD connectomic model has been proposed (Li et al., 2020;Smith et al., 2021;van der Vlis et al., 2021): the STN and the NAc.In each individual, using high resolution 7 Tesla magnetic resonance imaging (MRI) data and advanced methods for probabilistic tractography, we identified two accessible cortical stimulation sites structurally connected to the STN and the NAc.Ideally, we would have been able to identify cortical sites with (almost) exclusive, direct connectivity to the STN and NAc, respectively.While an (hyper)direct pathway connecting the STN and the cortex can be more reliably identified using a tractographic approach, results are less clearcut for NAc' connections.Therefore, based on the above-mentioned evidence pointing to a common pathway (Baldermann et al., 2021), we selected the cortical site displaying the strongest connectivity to the STN, and the cortical site displaying the strongest connectivity to both STN and NAc.Using a sham-controlled TMS-functional MRI setup, we address the question whether and how these deep-brain nuclei can be modulated non-invasively, as measured by changes in their functional connectivity (FC) with other regions of the brain.We finally discuss TMS-induced FC changes across the identified stimulated sites in light of the networks they critically participate in, and their potential relevance for OCD treatment.

Participants
Nine healthy, right-handed volunteers took part in the current study (age 18-25, 3 females), after one participant dropped out.Participants were screened for TMS and MRI contraindications and excluded accordingly when necessary.The experiment was approved by the Ethics Review Committee Psychology and Neuroscience (ERCPN) of Maastricht University, and all participants provided written informed consent.At the end of the experiment, participants filled a questionnaire about their experience across conditions, asking then to rate on a scale from 0 (not at all) to 100 (extremely) comfort and tolerability of the stimulation session that they judged as the worst.

Procedure
A schematic representation of the study procedure is shown in Fig. 1.First, participants were scanned on a 7 Tesla MRI scanner, where highresolution anatomical and diffusion-weighted imaging (DWI) protocols were acquired.In each individual brain, the anatomical data was used to manually segment bilateral STN and NAc.DWI data was preprocessed and probabilistic tractography was used to reconstruct the fiber tracts seeded from and traversing each segmented deep brain structure and reaching the cortex.For each participant, we selected two cortical regions as targets for the subsequent TMS procedure: the cortical region mostly connected to the STN (target-1), and the cortical region mostly connected to both .In a within-subject, singleblind, cross-over design with separate sessions on different days, we administered TMS over the two individually defined cortical targets.Additionally, placebo stimulation using a sham coil over target-1 was used as control condition.The order was randomized and counterbalanced across participants.During each session, participants were first scanned on a 3 Tesla MRI scanner to acquire baseline anatomical and resting-state functional MRI (rs-fMRI) protocols.Once outside the scanner, we used neuronavigation to correctly place the TMS coil over the stimulation target and we applied continuous theta-burst stimulation (cTBS).Within five minutes from end of stimulation, participants underwent a second MRI scan to acquire two follow-up rs-fMRI protocols, interleaved by an anatomy scan.

Manual segmentation of subcortical nuclei
Bilateral STN and NAc were manually segmented in individual space using Insight Toolkit (ITK-SNAP, v3.4.0 www.itksnap.org).The anatomical landmarks used to identify the boundaries of the structures were based on previous literature (Neto et al., 2008;Massey et al., 2012;Salgado and Kaplitt, 2015;Güngör et al., 2018).Based on the T1w UNIFORM image, the NAc was located anterior to the posterior border of the anterior commissure, lateral to the inferior border of the lateral ventricle, ventral to the caudate nucleus and internal capsule, and dorsal to the external capsule and Broca's diagonal band, appearing in a round, biconvex, dorsally flattened shape (Neto et al., 2008;Salgado and Kaplitt, 2015).
A Quantitative Susceptibility Mapping (QSM) image was reconstructed from the multi-echo T2w image to obtain better contrast for STN delineation, using the Sepia toolbox (Chan and Marques, 2021) within Matlab (Matlab R2019b; Mathworks Inc).The STN was identified medial to the ventral area of the globus pallidus, lateral and anterior to the red nucleus, and dorsomedial to the anterior edge of the substantia nigra, appearing in an oblique position in all three planes (Massey et al., 2012;Güngör et al., 2018).The size of the NAc and STN has been variably reported in the range of 300 to 800mm 3 (Lauer et al., 2001;Seifert et al., 2015;Lee et al., 2020) and approximately 100mm 3 (Massey et al., 2012;Zwirner et al., 2017), respectively.Individual volumes and average sizes were calculated across participants.All segmentations were visually inspected by an experienced stereotactic neurosurgeon.In each individual brain, we used probabilistic tractography to reconstruct fiber tracts traversing manually segmented STN/NAc and reaching the cortex.The cortical region mostly connected to the STN (target-1) and that mostly connected to both STN and NAc (target-2) were selected as TMS targets (PART 1).In a within-subject crossover design with separate sessions, continuous theta burst stimulation was administered over the two individually-defined targets.Additionally, a sham coil over target-1 was used as control condition.The order was randomized and counterbalanced across participants.A 10-minute resting-state fMRI protocol was acquired before and at two time points after stimulation, interleaved by an anatomy scan (PART 2).yo: years old; MRI: magnetic resonance imaging; STN: subthalamic nucleus; NAc: nucleus accumbens; fMRI: functional MRI; TMS: transcranial magnetic stimulation; T1: T1w anatomical scan.
We performed whole-brain tractography with second-order integration over FODs probabilistic algorithm (Tournier et al., 2010) to reconstruct a tractogram of 10 million streamlines, using dynamic seeding (Smith et al., 2015) in 0.7 step size, and imposing a maximum track length of 250 mm.We additionally performed unidirectional targeted tractography, seeding at random within the bilateral STN and NAc, keeping all parameters at default or same as in whole-brain tractography.To improve the biological accuracy of the reconstructions, for both whole-brain and targeted approaches we employed anatomically-constrained tractography, using tissue priors to inform biologically realistic streamlines generation and ending (Smith et al., 2012).To allow quantitative inferences on the reconstructed streamlines, spherical-deconvolution informed filtering of tractograms (SIFT2) was used on the combined whole-brain and targeted tracking data to proportionally match the streamline density to the estimated density of each fiber population in every voxel of the image (Smith et al., 2015).
The sum of streamline weights obtained from this step was then quantified in a connectome matrix (Smith et al., 2015), using 210 cortical and 34 subcortical regions from the Brainnetome Atlas (Fan et al., 2016), alongside the manually segmented bilateral STN and NAc.All regions were coregistered to native DWI space as follow.The T1w image was registered to the MNI152 1 mm template using Advanced Normalization Tools (ANTs v3.0, http://stnava.github.io/ANTs/)SyN registration (Avants et al., 2008), then applying the inverse of the warp fields and generic affine matrix to the parcellation image.The T1w image was coregistered to the T2w image using ANT's registration tools with specific parameters for partial slab registration (https://github.com/ntustison/PartialSlabEpiT1ImageRegistration), and the inverse transform was then applied to the STN segmented volumes.A two-step rigid body registration was finally used to align the T1w image to the DWI image using the FMRIB Software Library (FSL v6.0; http://www.fmrib.ox.ac.uk/fsl) boundary-based registration (Greve and Fischl, 2009), and the inverse of the rigid transform was applied to all T1 space parcel images.
Based on the individual connectivity matrix, two regions were selected: the cortical region displaying the strongest connectivity (i.e.highest streamline weights) with the STN (target-1), and the cortical region displaying the strongest connectivity (i.e.highest summed streamline weights among the nodes in the 80th percentile of the connectivity distribution of both nuclei) with the STN and NAc (target-2).Both right and left nuclei were considered, and the hemisphere displaying the strongest connectivity was chosen per individual.We extracted from the combined tracking data all tracks traversing the STN and NAc, we mapped the streamline endpoints to an image (Calamante et al., 2012), and we selected as individual stimulation target the highest-intensity voxel within the previously identified most-connected cortical region.To aid visualization and TMS targeting, the final mask was dilated by 3 voxels.

TMS protocol
TMS was delivered using a figure-of-eight coil (MCB70) connected to a X100 MagVenture stimulator (MagVenture, Farum, Denmark).For precise coil placement and stimulation, neuronavigation was performed using Localite TMS Navigator software (LOCALITE Biomedical Visualization Systems GmbH, Sankt Augustin, Germany).When possible, the entry point automatically calculated at the shortest distance from the target was chosen.In case the entry point would fall on unreachable locations (e.g. the TMS coil would cover the participant's eye), or raise discomfort and safety concerns, we calculated different entry points at a larger distance from the target, and moved the coil upward to either option until an acceptably comfortable position was reached.In order to reduce the inter-scan time interval and overall duration of the experimental session, we used a cTBS protocol, delivering 50 Hz triplets of pulses 5 times a second for 40 s (600 pulses in total).Stimulation intensity was set at 100 % resting motor threshold, determined by finding the right motor cortex and adjusting stimulation intensity until observing a visible movement in the contralateral finger in half of the trials (Rossini et al., 2015).A purpose-built placebo TMS coil (MC-P-B70, MagVenture, Farum, Denmark) with strong attenuation of the magnetic field was placed over target-1 as control condition.During the stimulations, the coil was held tangentially to the scalp with a handle orientation at a 45-degree angle to the midline.In a posteriori check, we performed e-field modeling to estimate the magnitude of electrical field at the intended target.This was done using SimNIBS v4.0.1 (https: //simnibs.github.io),following the charm segmentation pipeline for head model generation (Puonti et al., 2020) and employing for e-field modeling the default parameters associated with the MCB70 Magventure coil, part of the list of validated coils available within SimNIBS (Drakaki et al., 2022).Using the get_fields_coordinates function, we then extracted the precise magnitude of vector fields at the target coordinates and compared it with the magnitude determining the top 99.9, 99 or percentiles of the generated field.

Rs-fMRI preprocessing and functional connectivity analysis
Rs-fMRI preprocessing was performed following the minimal processing pipelines for the Human Connectome Project, described in (Glasser et al., 2013).Briefly, anatomical images were minimally preprocessed (bias field correction), segmented (using Freesurfer v5.2, https://surfer.nmr.mgh.harvard.edu/)and registered to standard space, producing GM and WM masks necessary for functional preprocessing.This included motion and EPI-induced distortions correction, bias field correction, intensity normalization and registration to the T1w and MNI volume spaces, all applied in a single resampling step.FSL's independent component analysis (ICA)-based Xnoiseifer (Salimi-Khorshidi et al., 2014;Griffanti et al., 2014) was then used to regress out motion timeseries and artifact ICA components.Preprocessing and analysis of the fMRI data was completed using the CONN toolbox (Nieto-Castanon, 2021) within Matlab.Potential outlier scans were detected based on framewise displacement above 0.9 mm.We applied spatial convolution with a Gaussian kernel at 6 mm full width half maximum (FWHM).Potential confounding effects were regressed out of the data, removing noise components from WM and CSF (Behzadi et al., 2007), estimated subject-motion parameters (Friston et al., 1996) and the identified outlier scans (Power et al., 2014).Temporal frequencies below 0.01 Hz or above 0.1 Hz were removed from the signal.
In our first-level analyses, we computed the following voxel-level measures within the STN and NAc: amplitude of low frequency fluctuations (ALFF), estimating the variability of BOLD signal power within the defined frequency band (Yang et al., 2007), and local correlation (LCOR) maps, estimating voxel-level local coherence as the strength and sign of connectivity between a voxel and its local neighborhoods (Deshpande et al., 2009), defined using a 4 mm FWHM Gaussian kernel.We then computed seed-based connectivity maps as the Fisher-transformed bivariate correlation coefficients between the right STN/NAc timeseries and each individual voxel's timeseries.To investigate the temporal dynamics of resting-state FC changes, we decomposed the time-course signal in sliding windows of 100 s (shifted in steps of s) to compute the dynamic variability in seed-based connectivity.For all mentioned FC analyses, the manually segmented masks of the STN and NAc were used as seeds, limitedly to the hemisphere displaying the strongest connectivity as identified in Section 2.5.General linear models were used to investigate whether a difference in FC values from pre-to post-stimulation (at 5 and 25 min) would differ in the active vs. sham conditions.Statistical maps were thresholded using parametric cluster-based inferences (voxel p-uncorrected < 0.05, cluster p-FDR< 0.01).Individual FC values were extracted to investigate the direction of FC changes across conditions, and planned contrasts with paired sample t-test were performed to check for a significant increase or decrease in FC in the sham and TMS conditions separately.Finally, we calculated Pearson correlation coefficients between individual difference values in FC pre-and post-TMS, and the recorded distance between stimulation target and actual coil placement, in order to evaluate the potential impact of increased distance entry-target on observed effects.

TMS target selection and procedure
Manually segmented nuclei had an average size largely consistent with what previously reported (Massey et al., 2012;Lauer et al., 2001;Seifert et al., 2015;Lee et al., 2020;Zwirner et al., 2017), of 99.9(±9.5)mm 3 for the right STN, 102(±11.2) mm 3 for the left STN, 420.2(±55.3)mm 3 for the right and 466(±71.9)mm 3 for the left NAc.For all our participants, the fiber tracking and target selection procedure yielded two stimulation targets located in the right hemisphere, connected to the right STN and right NAc.Accordingly, subsequent analyses were conducted limitedly to the right STN and NAc as seeds of interest.Individual MNI millimeter coordinates and atlas label of target-1 and target-2 are reported in Table 1 and displayed in Fig. 2. Target-1 was mostly located in the precentral and postcentral gyrus, although three participants showed STN strongest connectivity to the superior frontal gyrus.Given the relative ease of reach, all participants received stimulation at the shortest possible distance from the target, as optimally computed by the neuronavigation software (Mean(±SD): 19.7 mm (±2.68),Table 1).Subsequent e-field modeling showed that, for all participants, target-1 (Mean(±SD): 1.19 V/m (±0.3)) fell within the top 99 percentile of the generated magnitude of electric field (Mean(±SD): 0.74 V/m (±0.06)).Target-2 was located in the middle frontal gyrus for all participants except one (in the orbital gyrus).Stimulation of target-2 at the shortest calculated distance was unfeasible, with the participant's eye being partially or fully covered by the TMS coil.We therefore moved the coil upward to the differently calculated entry points until an acceptably comfortable position was reached.For stimulation of target-2, we consequently registered a higher average and inter-subject variability distance entry-to-target (Mean(±SD): 33.6 mm (±9.41),Table 1].E-field modeling showed that, for all subjects, target-2 (Mean(±SD): 0.85 V/m (±0.3)) fell within the top 95 percentile of the generated magnitude of electric field (Mean(±SD): 0.37 V/m (±0.05)).Resting motor threshold was calculated before each stimulation session, with small intra-individual variation (Mean(±SD): 0.02 (± 0.009)).Across participants, resting motor threshold ranged from 26 % to 53 % of machine stimulator output, with an average of 37 %.When asked about their experience across the different sessions, 8 out of the 9 participants provided us with their impressions.All of them indicated target-2 stimulation as the most uncomfortable/unpleasant, rating the comfort with an average score of 28.75(±18.2),and the tolerability with an average score of 37.5(±18.8).All except 2 participants did not report any noticeable difference between sham stimulation and stimulation of target-1.

Modulating functional connectivity
We did not find a significant difference in the ALFF and LCOR of either right STN or NAc following stimulation of target-1 nor target-2 as compared to sham.Hereafter, we report all clusters that significantly differed (with a p-FDR<0.001) in the observed FC change (from pre-to post-) across conditions (sham vs. active stimulation).We report the direction of FC change (increase vs. decrease) and associated p-value from planned contrast testing, and we explicitely mention when the TMS-induced FC change was smaller than what observed in the sham condition.

Table 1
Individual stimulation targets.From left to right, we report the MNI millirniter coordinates of target-1 and target-2 stimulation, as identified in Part 1 of the experienient.the corresponding label from the Brainnetoine Atlas, and the recorded distance in millimeter between the identified target and the entry point (i.e.coil position).PrG: precentral gyrus; SFG: superior frontal gyrus; PoG: postcentral gyrus; MFG: middle frontal gyms; OrG: orbital gyrus; ttru: trunk reeion; dl: dorsolateral area; m: medial area; 1: lateral area.

Correlation with distance entry-target
Pearson correlation coefficients were calculated between observed pre-post active TMS FC change and the recorded distance between stimulation target and actual coil placement.We only found a significant negative correlation for observed changes in FC between the right STN and right cuneal cortex at 25 min after target-2 stimulation (r = 0.659, p = .05),and between the right NAc and left ventrolateral prefrontal cortex at 5 min after target-1 stimulation (r = 0.779, p = .001).

Discussion
The present study set out to evaluate the feasibility and effect on functional connectivity of a personalized, connectivity-based procedure to TMS targeting, aiming to identify an accessible cortical stimulation target displaying the strongest connectivity to deep nuclei otherwise non-invasively out of reach.As a proof-of-concept for refractory OCD treatment, we focused on DBS targets of the known efficacy in relieving obsessive-compulsive symptoms: the STN and the NAc.Our data-driven approach relying on individual structural connectivity yielded two cortical stimulation sites quite homogenously distributed around the precentral and middle frontal gyrus.Owing to the extensive body of literature proving remote effects of non-invasive brain stimulation, we hypothesized that TMS over the individually-defined cortical entry point would modulate the resting-state FC of the connected deep targets towards the rest of the brain.Our sham-controlled, offline TMS-fMRI procedure indeed revealed both static and dynamic signatures of FC changes of both STN and NAc at two time points following stimulation.We here discuss the potential relevance of these findings to the treatment of refractory OCD by interpreting the obtained results against existing literature on OCD dysfunctional neurocircuits.We further articulate on a few feasibility-related considerations for connectivitybased approaches to TMS target selection, and interpretation of remote TMS effects.

Modulating functional connectivity networks
The STN and NAc are critical nodes of the cortico-striatal-thalamocortical (CSTC) loops, which decades of research convincingly linked to OCD pathology (Graybiel and Rauch, 2000;Milad and Rauch, 2012).
Both target-1 and target-2 were chosen based on their structural connectivity with the right STN.Consistent with our hypothesis, we observed significant changes in the FC of the STN, which were however detected only (for target-2 stimulation) or primarily (for target-1 stimulation) between 25-to-35 min following TMS.The STN primarily participates in the sensorimotor circuit (with regions involved in the generation and control of motor behaviors and integration of sensory information), and in the ventral cognitive circuit (alongside prefrontal and thalamic regions involved in self-regulatory behaviors) (van den Heuvel et al., 2016).Alterations in these networks are thought to drive dysfunctional habit formation and faulty inhibition responses, presumably explaining the persistence of maladaptive obsessive thoughts and compulsive behaviors, despite knowledge of their irrationality and associated negative consequences (for a review, see Shephard et al., 2021).TMS of the supplementary/pre-supplementary motor area (SMA/pre-SMA) and DBS of the STN likely act on these circuits and consequently improve the experienced symptoms.Here, stimulation of the cortical region individually most connected with the STN increased   Individual functional connectivity values are extracted and plotted for each significant cluster across the four contrasted conditions, represented as line-connected dots.The color of the x-axis signals the corresponding cluster on the brain surface, and peak voxel coordinates are reported.Violin plots represent the observed preto-post difference in the sham condition (light green) and in the active stimulation condition (salmon).We star where the TMS-induced difference was significantly smaller than the sham-induced difference.L: left; R: right; NAc: nucleus accumbens; FC: functional connectivity; TMS: transcranial magnetic stimulation. .FC with a cluster of voxels in the primary motor and sensorimotor cortex, suggesting that this procedure could critically engage nodes of the sensorimotor circuit.However, despite a topographical organization is certainly present, the integration and overlap between the different CSTC loops is acknowledged (Milad and Rauch, 2012;Shephard et al., 2021), and it is thus likely that stimulation of one circuit affects the functioning of others.For example, the STN also critically participates in the control of emotional and motivational behaviors (van Wijk et al., 2020).Our results indeed show that TMS modulates the temporal variability of FC measures with a cluster in the paracingulate cortex (target-1) and ventral orbitofrontal cortex (OFC) (target-2), key components of the fronto-limbic circuit to generate, evaluate and regulate emotional responses (Kohn et al., 2014), and with the medial frontal cortex (target-1), potentially influencing executive functions important for goal-directed behavior (van den Heuvel et al., 2016).
Modulation of the FC of the NAc unfolded following a different temporal dynamic, with most stable changes being recorded in the first 5-to-15 min after stimulation.The NAc is a key component of the ventral affective circuit, together with regions involved in reward functions (OFC, ventral striatum and thalamus) (Haber and Knutson, 2010).OCD neurobiological theories suggest that a lowered sensitivity to reward as opposed to an enhanced sensitivity to, or an aversion of, punishment might drive some avoidant behaviors (Figee et al., 2013) or feelings of anhedonia (Abramovitch et al., 2014) that can be common in OCD patients.Studies for example report increased FC between NAc and other reward circuit regions at rest, and alterations during anticipation of reward and punishment (Jung et al., 2013), as well as poorer performance on reward-based decision-making and reward-learning tasks (Grassi et al., 2015;Grassi et al., 2020).Remarkably, our results show a significant decrease in the FC of the NAc with a large cluster in the OFC.Despite only target-2 was chosen based on its connectivity with the NAc, we observed the same effect following target-1 stimulation, once again suggesting that brain stimulation is more likely to elicit a cascade of neural events rather than targeted effects only.Consistent with the knowledge that the NAc also participates to other circuitries (Yin and Knowlton, 2006), we also observed increased variability of FC measures with sensorimotor regions following target-2 stimulation.Likewise, other regions participate to the functioning of the ventral affective circuit, such as components of the fronto-limbic and ventral cognitive circuit (thus including the STN) (van Wijk et al., 2020).In regard to the latter, we indeed recorded both static and dynamic modulation of NAc FC with ventral cognitive components (inferior frontal gyrus and ventrolateral prefrontal cortex) following target-1 stimulation.
It should be mentioned that we here only discussed FC modulations within the different CSTC loops, but research has suggested that OCD dysfunctional circuits are not restricted therein (Milad and Rauch, 2012;Shephard et al., 2021).Different lines of evidence have indeed suggested for example altered dynamics within and between intrinsic resting-state networks (Gürsel et al., 2018;Gürsel et al., 2020;Liu et al., 2021;Luo et al., 2021), or the special role of certain hub regions, such as the inferior parietal lobule (Picó-Pérez et al., 2019), the anterior cingulate cortex (Tang et al., 2019) or the precuneus (Jones and Bhattacharya, 2014;Fajnerova et al., 2020).Our results also show modulation across visual, parietal and precuneal areas, suggesting that this procedure could also critically modulate these nodes, with potentially secondary effects on the networks in which they participate.
To conclude, we observed the modulation of nodes of both overlapping and segregated functional networks following TMS at two cortical sites differently structurally connected to the STN and the NAc.Although we did not aim to compare targets in terms of their potential therapeutic superiority, it could be argued that one entry point might be more effective than another, depending on the prevalent functional and/ or behavioral dysfunction of the individual patient.A neurocircuit-based taxonomy of OCD has in fact been recently suggested (Shephard et al., 2021), advancing the hypothesis that patients might be clustered based on dominant dysfunction in one (or more) neurocircuit and specific phenotypic manifestations of the disorder.However, a validation of this taxonomy within the context of brain stimulation target selection still needs to be addressed, and explicit comparisons between the efficacy of stimulating different targets should be addressed in appropriately powered clinical studies.

Feasibility of individualized, connectivity-based target definition
Connectivity-based selection procedures for TMS target definition have been established primarily in the context of depression treatment (MD Fox et al., 2012;Cash et al., 2019;Fox et al., 2013;Cash et al., 2021;Cash et al., 2021).While most attempts at personalizing target selection relied on FC patterns, recent results suggest the potential superiority of diffusion-based targeting (Luber et al., 2022), although formal comparisons of the two approaches in patients populations are still lacking.When it comes to fiber tractography, the field is particularly aware of potential biases inherent to either deterministic or probabilistic approaches (Jbabdi and Johansen-Berg, 2011;Schilling et al., 2018;Sarwar et al., 2019).However, improvements in the quantitative assessment of streamline reconstructions now allow more precise and reliable use of structural connectivity measures (Smith et al., 2015;Smith et al., 2012), for example for the purpose of TMS target selection.The STN might be particularly suited to this procedure, displaying strong and widespread anatomical connectivity to the cortex (Brunenberg et al., 2012).Non-human and human studies have distinguished three functionally different parts within the STN, differentiating by their connectivity with motor, associative or limbic areas (Hamani et al., 2004;Temel et al., 2005).The sensorimotor part constitutes as the largest, exhibiting direct connections to primary motor, premotor, SMA and somatosensory cortex, via the so-called "hyperdirect" pathway (Brunenberg et al., 2012;Hamani et al., 2004;Nambu et al., 2000).In line with this body of literature, in the present study the cortical targets individually most connected to the STN spread across the postcentral, precentral, and superior frontal gyri.Remarkably, despite the inter-individual variability in exact localization, TMS induced group-level effects on STN FC.Of note, low-frequency stimulation of the SMA/pre-SMA is ranked among the most effective TMS targets for OCD treatment, despite the inter-individual variability in clinical response remaining relatively large (Fitzsimmons et al., 2022).We advance the hypothesis that another region within the sensorimotor network might lead to better results, if individually showing stronger connectivity to its subcortical components than a group target selected a priori.
In keeping with recent advances in understanding DBS mechanisms (Brunoni et al., 2017;Zhou et al., 2017;Rehn et al., 2018;Fitzsimmons et al., 2022), we additionally selected the cortical region displaying the strongest connectivity to both the STN and NAc.Independent lines of evidence have indeed suggested that the closer the DBS electrode to a fiber pathway connecting the thalamus to the lateral and medial prefrontal cortex, the better the clinical response, reporting this association in different cohorts of patients targeted in either the NAc-ventral capsule or the STN (Baldermann et al., 2021;Li et al., 2020;Smith et al., 2021;van der Vlis et al., 2021).In line with what was reported, our procedure localized this region in the middle frontal gyrus, homogeneously across individuals.The OCD TMS literature counts only two studies targeting the nearby right or left OFC (Ruffini et al., 2009;Nauczyciel et al., 2014), hence not allowing any conclusion to be made on efficacy in OCD patients.It should be noted that stimulation of this area comes with a number of challenges; in the present study, participants poorly rated the tolerability and comfort of this particular stimulation session, due to stimulation of facial nerves and relatively strong muscle twitches.Furthermore, placing the TMS coil at the shortest distance from target was not possible.These aspects should also be critically considered when discussing the advantage of one stimulation target over another.
To conclude, our data-driven target selection approach partly aligns with what is encountered across the OCD brain stimulation literature.The idea of using neuroimaging to improve precise spatial localization and account for anatomical individual differences has long been introduced, and measurable differences on the effects of stimulation reported (Sack et al., 2009;Fitzgerald et al., 2009;Rusjan et al., 2010).It remains to be confirmed whether the large observed inter-individual variability in response to TMS for OCD can be reduced by targeting cortical stimulation sites selected based on individual structural connectivity patterns to subcortical nuclei.

Interpreting remote TMS aftereffects
Remote effects of TMS have primarily been established using interleaved TMS-fMRI paradigms, assessing the online cortical and subcortical impact of motor/sensorimotor (Bestmann et al., 2004;Denslow et al., 2005;Bestmann et al., 2003;Bestmann et al., 2005;Peters et al., 2020), parietal (Sack et al., 2007) and prefrontal cortex stimulation (Luber et al., 2022;Li et al., 2004;Cho and Strafella, 2009;Hanlon et al., 2017;Fonzo et al., 2017;Wang et al., 2018;Dowdle et al., 2018;Vink et al., 2018;Oathes et al., 2021).Other offline studies add to this literature reporting TMS-induced modulation of widespread, connected networks (Davis et al., 2017;Tik et al., 2017;Paus et al., 2001;Fox et al., 2012;Liston et al., 2014), establishing the idea that TMS could be used to access layers lying deeper than what directly targeted at the 3 cm reach of standard coils.However, understanding how these effects arise is non-trivial.Neuronal activation at the stimulation site transsynaptically spreads to remote connected regions, but it is unclear whether this happens via changes in FC, or via activation-induced synaptic plasticity at the remote site itself during stimulation (Bergmann and Hartwigsen, 2021).Additionally, a number of potential confounding or covarying factors potentially play a role, and a scenario where remote effects are actually induced by e.g.unintended costimulation of non-target regions and networks is difficult to exclude (for a review and further considerations, see Bergmann and Hartwigsen, 2021).Of note, we considered FC changes of remote sites towards other regions of the brain, placing our outcome measure even further down the causal chain of stimulation effects, and thus further away from straightforward interpretations.Besides the problematic understanding of how these remote aftereffects arise, characterizing how they manifest is equally challenging.Distinguishing stimulation protocols based on their inhibitory vs. facilitatory effect on synaptic plasticity is admittedly oversimplified (Lefaucheur et al., 2020), even more so when considering remote effects, the direction of which is difficult to predict.Additionally, even assuming an a priori understanding of these effects, influencing the neural activity of remote regions does not imply influencing their functional connectivity in the same expected direction, as these concepts cannot be considered as a sign of one another.Our results indeed show region-specific modulations, with instances of increases or decreases in FC across different network nodes.Of note, in some cases we recorded smaller changes following active as compared to sham stimulation, which could be seen either as a sign that our sham procedure failed to control for secondary effects of stimulation (Duecker and Sack, 2013;Duecker and Sack, 2015), or as an instance of a true inhibitory effect on the dynamic fluctuations in FC that can be normally expected (Shehzad et al., 2009;Liu et al., 2010;Meindl et al., 2010;Van Dijk et al., 2012).
In light of all these considerations, notwithstanding the importance of basic, combined TMS-neuroimaging studies like ours, any observed increase or decrease in the FC of (remote) regions has limited interpretability in the absence of a context where communication between these regions becomes relevant.To our knowledge, this is the first attempt at investigating targets used in brain stimulation treatment for OCD.Our proof-of-concept study established that individualized, connectivity-based TMS can modulate the FC of OCD-relevant deep targets.However, we underscore the importance of taking the next necessary steps, asking whether this TMS procedure elicits the same observed changes in OCD patients, and whether these changes then translate to a cognitive, emotional, or behavioral shift to functional patterns.Once and if this is established, this procedure could be used as standalone treatment, or in the evaluative phase of DBS to judge which network is more easily engaged in the individual patient, and thus more likely to benefit from invasive electrode placement.

Limitations
Several limitations constrain the interpretation of our results.First, we conducted a proof-of-concept study in a small sample of healthy volunteers.Conforming to its preliminary and exploratory nature, when assessing the results, we deliberately used a liberal statistical threshold that could have thus inflated the reported group effects.Replication across wider, patients cohorts is thus mandatory.Second, it is well established that the TMS-induced magnetic field decays exponentially at increasing distance from the coil (Thielscher and Kammer, 2004), and that this can be compensated by adjusting stimulation intensity to the estimated scalp-cortex distance (Stokes et al., 2005).This would have been particularly advisable for stimulation of target-2.However, given the low-rated comfort and tolerability, we believe that stimulating at an intensity higher than 100 % resting-motor-threshold would have been unfeasible.The lack of FC changes detected at different time points following target-2 stimulation might be explained by the increased distance between the stimulating coil and the intended cortical target, which we were not able to appropriately compensate.However, correlation analyses suggest that variability in the entry-target distance did not majorly impact the FC changes that were detected.Furthermore, the e-field modeling results suggest that a reasonably strong field might have still hit target-2.Nonetheless, in keeping with considerations on remote TMS aftereffects discussed above, we cannot exclude that other factors confounding or covarying with the effects of stimulation might have influenced the observed results.

Conclusions
We employed a data-driven approach to identify cortical sites accessible non-invasively based on individual structural connectivity of OCD-relevant deep targets; the STN and NAc.This procedure yielded two cortical sites quite homogeneously distributed around the precentral and middle frontal gyrus.These results partly align with what is encountered across the OCD brain stimulation literature, thus suggesting the feasibility and anatomical reliability of this approach.Our shamcontrolled, offline TMS-fMRI procedure revealed that TMS over these identified cortical sites can modulate both static and dynamic signatures of FC of the intended deep targets, with overlap and differences in the engaged networks across stimulation sites.Given the relevance of these networks to OCD pathology, we deem an individualized, connectivitybased TMS procedure worth investigating further, by evaluating its effects in the OCD patient population, by comparing its efficacy to standard TMS target selection, and by testing it within the context of a neurocircuit-based taxonomy to guide target selection.

Ethics statement
The experiment was approved by the Ethics Review Committee Psychology and Neuroscience (ERCPN) of Maastricht University, and was carried out in accordance with the Code of Ethics of the World Medical Association (Declaration of Helsinki).All participants provided written informed consent.

Data and code availability
The data and the code that support the findings of this study can be shared upon reasonable request.Researchers wishing to obtain access must contact the corresponding author and last author providing a concrete project outline on the use of the data.A formal data sharing agreement will be eventually arranged.

Fig. 1 .
Fig.1.Schematic representation of study design.First, participants underwent 7T MRI scanning, acquiring anatomical and diffusion-weighted imaging data.In each individual brain, we used probabilistic tractography to reconstruct fiber tracts traversing manually segmented STN/NAc and reaching the cortex.The cortical region mostly connected to the STN (target-1) and that mostly connected to both STN and NAc (target-2) were selected as TMS targets (PART 1).In a within-subject crossover design with separate sessions, continuous theta burst stimulation was administered over the two individually-defined targets.Additionally, a sham coil over target-1 was used as control condition.The order was randomized and counterbalanced across participants.A 10-minute resting-state fMRI protocol was acquired before and at two time points after stimulation, interleaved by an anatomy scan (PART 2).yo: years old; MRI: magnetic resonance imaging; STN: subthalamic nucleus; NAc: nucleus accumbens; fMRI: functional MRI; TMS: transcranial magnetic stimulation; T1: T1w anatomical scan.

Fig. 2 .
Fig.2.Individual TMS stimulation targets, defined as the (voxel) connectivity hotspot within the cortical region most connected to the STN (target-1) and that most connected to both STN and NAc (target-2).For all participants, the right hemisphere displayed the strongest connectivity patterns.L: left; R: right; TMS: transcranial magnetic stimulation.

Fig. 3 .
Fig. 3. Connectivity maps of significant functional connectivity changes of the right STN following active as compared to sham stimulation.Significant group-level clusters from second-level general linear models are projected on a brain surface using the CONN toolbox at 5 min (A,C) and at 25 min (B,D) post target-1 (A,B) and target-2 (C,D) stimulation.Individual functional connectivity values are extracted and plotted for each significant cluster across the four contrasted conditions, represented as line-connected dots.The color of the x-axis signals the corresponding cluster on the brain surface, and peak voxel coordinates are reported.Violin plots represent the observed pre-to-post difference in the sham condition (light green) and in the active stimulation condition (salmon).We star where the TMS-induced difference was significantly smaller than the sham-induced difference.L: left; R: right; STN: subthalamic nucleus; FC: functional connectivity; TMS: transcranial magnetic stimulation. .

Fig. 4 .
Fig. 4. Dynamic connectivity maps of significant functional connectivity changes of the right STN following active as compared to sham stimulation.Significant group-level clusters from second-level general linear models are separately projected on a brain surface at 25 min post target-1 (A) and target-2 (B) stimulation.Peak voxel coordinates are reported.Group average functional connectivity values are extracted and plotted across time (i.e.sliding windows) for the different contrasted conditions; the left side of each graph represents the pre-stimulation condition (sham in green, active TMS in blue), whereas the right side of each graph represents the post-stimulation condition (sham in orange, active TMS in yellow).The group average temporal variability of FC across time is represented as the bar graph in the middle.We star where the TMS-induced difference was significantly smaller than the sham-induced difference.L: left; R: right; STN: subthalamic nucleus; FC: functional connectivity; TMS: transcranial magnetic stimulation.

Fig. 5 .
Fig. 5. Right NAc connectivity maps of significant functional connectivity changes following active as compared to sham stimulation.Significant group-level clusters from second-level general linear models are projected on a brain surface at 5 min (A,C) and at 25 min (B,D) post target-1 (A,B) and target-2 (C,D) stimulation.Individual functional connectivity values are extracted and plotted for each significant cluster across the four contrasted conditions, represented as line-connected dots.The color of the x-axis signals the corresponding cluster on the brain surface, and peak voxel coordinates are reported.Violin plots represent the observed preto-post difference in the sham condition (light green) and in the active stimulation condition (salmon).We star where the TMS-induced difference was significantly smaller than the sham-induced difference.L: left; R: right; NAc: nucleus accumbens; FC: functional connectivity; TMS: transcranial magnetic stimulation. .

Fig. 6 .
Fig. 6.Dynamic connectivity maps of significant functional connectivity changes of the right NAc following active as compared to sham stimulation.Significant group-level clusters from second-level general linear models are separately projected on a brain surface at 5 min (A,C) and at 25 min (B,D) post target-1 (A,B) and target-2 (C,D) stimulation.Peak voxel coordinates are reported.Group average functional connectivity values are extracted and plotted across time (i.e.sliding windows) for the different contrasted conditions; the left side of each graph represents the pre-stimulation condition (sham in green, active TMS in blue), whereas the right side of each graph represents the post-stimulation condition (sham in orange, active TMS in yellow).The group average temporal variability of FC across time is represented as the bar graph in the middle.We star where the TMS-induced difference was significantly smaller than the sham-induced difference.L: left; R: right; NAc: nucleus accumbens; FC: functional connectivity; TMS: transcranial magnetic stimulation.