Effects of personalized transcranial magnetic stimulation on social cognitive network functional connectivity in schizophrenia spectrum disorders: A double-blind, randomized, sham-controlled target engagement trial Authors: Lindsay D. Oliver, Daniel M. Blumberger, Zhi-De Deng, Colin Hawco, Erin W. Dickie, Julia Gallucci, Jerrold Jeyachandra, Salim Mansour, Stephanie M. Hare, James M. Gold, George Foussias, Miklos Argyelan, Zafiris J. Daskalakis, Robert W. Buchanan, Anil K. Malhotra, Aristotle N. Voineskos Published: 2026 ABSTRACT Background:Schizophrenia spectrum disorders (SSDs) often feature debilitating social cognitive impairments, with limited treatment options. This study used a target engagement approach to examine the effects of individually-targeted repetitive transcranial magnetic stimulation (rTMS) to the dorsomedial prefrontal cortex (DMPFC) on social cognitive network functional connectivity (i.e., mentalizing and simulation networks) in individuals with SSDs.Methods:We conducted a double-blind, randomized, sham-controlled trial in adults with SSDs (NCT04418011). Participants received two weeks of DMPFC stimulation with 10 Hz rTMS, intermittent theta burst stimulation (iTBS), or sham stimulation. They completed pre- and post-treatment functional magnetic resonance imaging, including a social processing task. Treatment was personalized using functional connectivity mapping and electric field modelling. Primary outcomes were changes in individualized within-mentalizing and between mentalizing-simulation network functional connectivity after active stimulation versus sham, compared using effect sizes. Secondary outcomes were safety and tolerability.Results:Analyses included 46 participants who completed treatment (10 Hz rTMS: n = 15; iTBS: n = 15; sham: n = 16). iTBS produced a significantly greater increase in within-mentalizing network connectivity, with a large effect size, while 10 Hz rTMS did not differ from sham. Effects on between-network connectivity were negligible-to-small for both active conditions. Participants receiving iTBS reported higher pain ratings, but adverse event and discontinuation rates did not differ across groups.Conclusions:Individualized DMPFC-targeted iTBS demonstrated target engagement via increased mentalizing network connectivity relative to sham, and was feasible, safe, and well-tolerated in adults with SSDs. A larger and longer clinical trial powered to assess change in social cognitive outcomes is ongoing (NCT06118268). Introduction Impairments in social cognition significantly contribute to poor functional outcomes and reduced quality of life in individuals with schizophrenia spectrum disorders (SSDs) [1–3]. Despite their clinical significance, there are currently no approved medical interventions for social cognitive deficits. These deficits span both lower- and higher-level processes [4–6], thought to rely on at least two partially dissociable but interacting social cognitive neural systems [7–9]. The first is a frontoparietal and insular mirroring or simulation network, which is implicated in lower-level emotion recognition and action understanding [10–12]. The second is a midline and lateral temporal mentalizing network, which supports higher-level theory of mind and complex mental state inference [13,14]. Functional magnetic resonance imaging (fMRI) studies have reported atypical functional connectivity between regions of the mentalizing and simulation networks in people with SSDs compared to healthy controls [15,16]. Our previous fMRI work using the Empathic Accuracy Task across over 250 people with SSDs and healthy individuals found that better social cognitive performers exhibited greater negative functional connectivity between social cognitive networks [17]. Other fMRI studies have demonstrated associations between mentalizing network connectivity, including the dorsomedial prefrontal cortex (DMPFC), and social performance or functioning in SSDs [18–22]. Modulating brain circuits involved in these processes presents a promising therapeutic approach to enhance social cognitive performance and, ultimately, improve social functioning [23]. As a key node of the mentalizing network, the DMPFC may be a viable target to engage social cognitive circuitry. Repetitive transcranial magnetic stimulation (rTMS) is a non-invasive neuromodulation technique approved for the treatment of depression [24] and obsessive compulsive disorder [25], with recent data supporting its safety and potential efficacy for symptoms in SSDs [26,27]. Both standard 10 Hz rTMS and intermittent theta burst stimulation (iTBS) can safely target the DMPFC [28], but iTBS uses a more rapid and frequency-specific pulse pattern designed to engage synaptic plasticity more efficiently and potentially drive distinct network-level effects [29,30]. iTBS also reduces treatment time and cost, and our group has established its non-inferiority to 10 Hz rTMS in depression [24], but data comparing these protocols in SSDs is limited. Although TMS can modulate activity in specific brain regions and associated networks [31], clinical response remains highly variable. Individual differences in brain functional organization [32–34], including in SSDs [35], contribute to this variability [36]. Variations in cortical geometry and TMS coil positioning further influence the resulting electric field and stimulation dose [37,38]. Consequently, the conventional approach of targeting a common region using fixed scalp-based heuristics and coil placement may be suboptimal, stimulating different regions or networks across individuals [39,40]. Early prospective evidence suggests that personalized functionally-defined TMS targets may improve treatment response in depression [41–44], and very recently in SSDs for negative symptoms [45]. TMS-induced electric field modelling is also starting to be incorporated in clinical trials to account for structural brain variation and coil positioning [46]. However, to our knowledge, no TMS trial in SSDs has prospectively included individualized targeting integrating both functional mapping and electric field modelling. We conducted a double-blind, randomized, sham-controlled target engagement clinical trial using individualized 10 Hz rTMS and iTBS to target social cognitive networks in SSDs. Individually-optimized targets and coil positioning were determined using our team’s recently published pipeline, incorporating personalized functional connectivity mapping and electric field modelling [47]. Our primary aim was to determine whether individualized DMPFC active 10 Hz rTMS or iTBS would demonstrate engagement of our neural target (i.e., target engagement) versus sham stimulation via change in functional connectivity 1) within the mentalizing network, and 2) between the mentalizing and simulation networks, during social cognitive processing. We hypothesized that active DMPFC stimulation would increase 1) within-mentalizing network positive connectivity, and 2) between-network negative connectivity, from pre-to post-rTMS compared to sham stimulation. Our secondary aim was to examine the comparative safety and tolerability of the different treatments. We also explored changes in social and non-social cognitive performance, and clinical symptoms. Methods Participants and study design Participants with SSDs aged 18–55 were recruited for the NeuroModulation of Social Cognitive Circuitry in People with Schizophrenia Spectrum Disorders (ModSoCCS) study from the Centre for Addiction and Mental Health (CAMH), Zucker Hillside Hospital (ZHH), and the Maryland Psychiatric Research Center (MPRC). Participants met diagnostic criteria for a SSD based on the Structured Clinical Interview for DSM-5 (SCID-5), were English-speaking, and clinically stable (see Supplementary Material for details and exclusion criteria). Participation was not dependent on a prespecified level of baseline social cognitive impairment. All participants signed an informed consent agreement. The protocol was approved by the respective research ethics and institutional review boards and was registered with clinicaltrials.gov (NCT04418011). All research was conducted in accordance with the Declaration of Helsinki. Clinical, cognitive, and safety assessments Participants completed measures of global psychopathology, negative symptoms, and depressive symptoms pre- and post-treatment, and the Quality of Life Scale [48] pre-treatment. A range of social and non-social cognitive tasks were conducted both pre- and post-treatment. The Empathic Accuracy (EA) task was completed during fMRI [49,50]. Participants viewed six videos (110-170s each), across two fMRI runs, of target individuals describing emotional autobiographical events, and provided continuous ratings of how positive or negative the targets felt throughout (see Supplementary Material for details). The Verbal Analogue Pain Scale was completed after each rTMS session, wherein participants were asked to rate their pain/discomfort verbally on a scale from 0 (no pain) to 10 (worst pain). Adverse events were assessed via an open-ended query for any adverse event or physical discomfort experienced following each treatment session and recorded in a treatment log by the rTMS technician. MRI data acquisition Pre-treatment MRI scans were acquired prior to randomization and used to identify personalized DMPFC targets, and we attempted to acquire post-treatment scans within one week of the final TMS treatment. MRI scans were collected on three 3T Siemens Prisma scanners using matched scanning parameters, including an anatomical T1-weighted scan, two eyes-closed resting-state fMRI scans, and two EA runs (see Supplementary Material for details and MRI quality control information). Individualized TMS targeting Treatment was individually-optimized using our open-source automated Bayesian Optimization Of NeuroStimulation (BOONStim) pipeline (see Ref. [47] for details; Fig. 1). Briefly, pre-treatment T1-weighted and resting-state fMRI data were preprocessed using fMRIPrep [51] and Ciftify [52] (see details below and in Supplementary Material). SimNIBS [53,54] and Freesurfer [55] generated participant-specific head models to run electric field simulations, which were registered to each participant’s fMRI surface [58]. Individualized DMPFC regions of interest (ROIs) were derived from resting-state functional connectivity. A Neurosynth meta-analytic map for ‘mentalizing’ [56] was projected to the cortex; the DMPFC parcel was removed, and the remaining network served as a seed to compute each participant’s mentalizing-to-DMPFC connectivity. These weighted DMPFC maps were resampled to the SimNIBS surface. A Bayesian optimization procedure then selected the coil position and orientation that maximized modelled dose over the target for each participant [53, 54,57]. The resulting individualized coordinates guided TMS delivery. Treatment procedures DMPFC-rTMS was administered using a MagVenture X100 or R30 stimulator equipped with a Cool-B70 A/P coil and Qooler fluid-cooling device (MagVenture, Farum, Denmark), positioned under MRI guidance using the Brainsight neuronavigation system (Rogue Resolutions, Montreal, Canada) at CAMH and MPRC, and the Localite system (Localite GmbH, Germany) at ZHH. Stimulation parameters Before initiating treatment, each participant’s resting motor threshold (RMT) of the hand region of the motor cortex was determined in accordance with standard clinical practice [59,60]. 10 Hz rTMS was applied using conventional FDA-approved parameters, and iTBS was delivered at the same site and intensity (120% RMT), differing only in stimulation pattern and number of pulses (see Supplementary Material). Electrical stimulators were used during both active and sham stimulation to mask the tactile and auditory sensation of active stimulation. Those in the sham group were randomly assigned to either 10 Hz rTMS or iTBS sham with the same parameters as the active conditions. Course Participants who completed the trial received a total of 10 treatments (rTMS, iTBS, or sham rTMS/iTBS) over two weeks conducted five days per week on weekdays. Randomization and integrity of the blind Participants were randomly assigned to one of the three arms using a permuted block randomization system. To assess blinding integrity, participants were asked to complete a conventional guess form after the first treatment, which asked whether they believed they received active or sham treatment. Participant blinding was quantified using the Simple Blinding Index [61] (see Supplementary Material for details). MRI processing and individualized connectivity metrics MRI data were preprocessed using standard pipelines as in our prior work [62,63], including fMRIPrep [51] and the Ciftify toolbox [52] for projection to the cortical surface. For the preprocessed EA task fMRI data, a GLM was run including event regressors, a parametric modulator based on EA ratings for each video, and head-motion and tissue-specific nuisance regressors [17,62,64]. The residual activation (i.e., removing task effects and noise) was retained for background connectivity analysis. The focus is on state-related functional connectivity rather than stimulus-driven coactivation, which is ideal for continuous tasks such as the EA task [65,66]. See Supplementary Material for details. Individualized ROIs for the mentalizing and simulation networks were generated based on the resting-state fMRI data. We focused on a small number of connections between the DMPFC and canonical regions of these networks based on the literature [10–14,67,68] and clustered meta-analytic maps given the limited sample size (see Supplementary Material for details). Briefly, individualized ROIs were defined for the DMPFC (the TMS target site, as described above), the left and right temporoparietal junction (TPJ; mentalizing network), and the left and right premotor cortex into the inferior frontal cortex (IFC; simulation network). Fisher’s Z-transformed Pearson correlations were calculated between the EA residual time courses of the DMPFC and the left and right TPJ for within-mentalizing network connectivity, and between the DMPFC and the left and right IFC for between mentalizing-simulation connectivity, both pre- and post-treatment. Statistical analyses Data were analyzed using R 4.1.0 [69]. Baseline demographic and clinical variables were compared across groups using nonparametric Kruskal–Wallis tests for continuous variables due to non-normal distributions and Fisher’s Exact tests for categorical variables. Baseline connectivity metrics were compared using ANOVAs, whereas baseline cognitive variables were compared using Kruskal–Wallis tests due to non-normal distributions. Given that this was a target engagement trial, the focus was primarily on generating effect sizes for modulation of the intended neural target (i.e., change in functional connectivity), and combining these with tolerability data to guide decisions about proceeding with a larger clinical trial and its design. For transparency, effect sizes, statistical test results, and p-values are all reported. Primary Analyses: Prespecified linear regression models were used to compare post-pre treatment change in within-mentalizing network connectivity (DMPFC and the left and right TPJ) and between mentalizing-simulation connectivity (DMPFC and the left and right IFC) across treatment conditions, with sham stimulation as the reference. Models were first built with only days between final treatment and post-treatment MRI as a covariate and then also with age, sex, and site as covariates. Pairwise comparisons were conducted using estimated marginal means to compare each active treatment condition against the sham condition, with FDR-correction for multiple comparisons. Effect sizes were calculated using Cohen’s d. Secondary Analyses: Tolerability was evaluated using an intent-to-treat (ITT) approach by examining group differences in mean pain scores across all treatments on the Verbal Analogue Pain Scale using an ANOVA, overall incidence of adverse events using Fisher’s exact tests, and exploring drop-out rates. Exploratory Analyses: Nonparametric Kruskal–Wallis tests were used to compare post-pre treatment change in social cognitive, non-social cognitive, and clinical variables across treatment conditions, given largely non-normal distributions according to visual inspection and Shapiro–Wilk normality tests (p < 0.05). Effect sizes were calculated between each active treatment versus the sham condition using Cliff’s delta, which can be interpreted as the degree of distributional non-overlap between two distributions from −1 to 1, where 0 indicates complete overlap [70,71]. Code used for these analyses can be found here: https://github.com/loliver4/ModSoCCS. Results Participant enrolment and data collection occurred between November 2020–February 2023. Seventy-seven participants consented to the study, 51 were randomized, 49 started treatment, and 46 completed treatment (10 Hz rTMS n = 15, iTBS n = 15, sham n = 16), demonstrating 94% retention (Fig. 2). Baseline participant demographic and clinical characteristics for those who started treatment appear in Table 1. Treatment groups did not differ significantly on any demographic or clinical features, aside from mixed race representation being greater in the iTBS group. For the pooled active arms versus sham, the Simple Blinding Index indicated successful blinding (0.07, 95% CI [−0.13, 0.33], p = 0.53). Further, no between-group differences were found for percentage of participants believing they had received active treatment (rTMS = 75%, iTBS = 100%, sham = 80%; p = 0.13), or RMT expressed as percentage of maximum stimulator output (mean ± SD: rTMS = 38.44% ± 7.3%, iTBS = 40.31% ± 11.8%, sham = 36.09% ± 8.31%; p = 0.43). Of the 46 participants who completed treatment, two participants completed their post-treatment MRI more than one week (both 17 days) after their final treatment and one participant failed to respond to more than one video during the EA task pre-treatment. Three participants demonstrated motion above a mean framewise displacement (FD) of 0.5 mm (max 0.64 mm) in either MRI session. Sensitivity analyses were conducted after excluding these participants in a reduced sample of 40 participants (10 Hz rTMS n = 14, iTBS n = 15, sham n = 11) for the connectivity metrics, mirroring the analyses in the full sample. Primary Outcomes: One outlier was detected for the post-pre within-mentalizing connectivity metrics (DMPFC and the left and right TPJ) based on a threshold of ± 3 standard deviations from the mean and was excluded from the within-mentalizing connectivity analyses. Change scores for the post-pre treatment connectivity metrics showed relatively normal distributions according to visual inspection and Shapiro-Wilk normality tests (all p > 0.1). At baseline, groups did not differ significantly in within-mentalizing or between mentalizing-simulation connectivity (Table S1), nor the number of days from final treatment to MRI (mean ± SD days: rTMS = 1.47 ± 2.2, iTBS = 1.07 ± 1.2, sham = 3.12 ± 5.6; p = 0.24). In the analyses adjusted for days since final treatment, age, sex, and site, for within-mentalizing network connectivity, pairwise comparisons showed a large effect and significant difference between iTBS and sham in DMPFC-left TPJ connectivity change (d = 1.03, 95% CI [−1.8, −0.23]; pFDR = 0.025), but not rTMS and sham (d = 0.21, 95% CI [−0.98, 0.57]; pFDR = 0.59; Fig. 3a). The regression model revealed a significant effect of treatment group on change in DMPFC-left TPJ connectivity (F(7,37) = 3.56, p = 0.0051). Relative to the sham group, the iTBS group showed a significantly greater increase in connectivity (β = 0.166, SE = 0.063, t = 2.62, p = 0.013), whereas the 10 Hz rTMS group did not differ significantly (β = 0.033, SE = 0.061, t = 0.541, p = 0.59). For DMPFC-right TPJ, pairwise comparisons showed a medium effect and non-significantly greater increase in connectivity in the iTBS condition versus sham (d = 0.59, 95% CI [−.4,0.20]; pFDR = 0.28), and a small, non-significant difference between rTMS and sham (d = 0.19, 95% CI [−0.96,0.58]; pFDR = 0.61; Fig. 3b). There was no significant effect of treatment group on change in DMPFC-right TPJ connectivity (F(7,37) = 0.905, p = 0.51; iTBS: β = 0.126, SE = 0.083, t = 1.51, p = 0.14; rTMS: β = 0.041, SE = 0.081, t = 0.506, p = 0.62). For between mentalizing-simulation connectivity, pairwise comparisons showed negligible to small non-significant differences in DMPFC-left IFC connectivity change for iTBS (d = −0.08, 95% CI [−0.68,0.85]; pFDR = 0.83) and rTMS (d = −0.44, 95% CI [−0.31,1.2]; pFDR = 0.48; Fig. 3c) compared to sham. There was no significant effect of treatment group on change in DMPFC-left IFC connectivity (F (7,38) = 0.616, p = 0.74; iTBS: β = −0.020, SE = 0.092, t = −0.219, p = 0.83; rTMS: β = −0.108, SE = 0.091, t = −1.19, p = 0.24). For DMPFC-right IFC connectivity change, pairwise comparisons revealed negligible to small non-significant differences for iTBS (d = 0.11, 95% CI [−0.88,0.65]; pFDR = 0.77) and rTMS (d = −0.23, 95% CI [−0.53,0.98]; pFDR = 0.77; Fig. 3d) versus sham. There was no significant effect of treatment group on change in DMPFC-right IFC connectivity (F(7,38) = 0.660, p = 0.70; iTBS: β = 0.025, SE = 0.084, t = 0.299, p = 0.77; rTMS: β = −0.051, SE = 0.083, t = −0.612, p = 0.54). There were no significant effects of age, sex, or site in the primary regression models (all p > 0.05), though the study was not powered to examine the effects of covariates. The same pattern of significant connectivity findings was observed in the sensitivity analyses with the smaller sample (see Supplementary Results). Secondary Outcomes: Number of adverse events and average pain ratings by treatment group appear in Table 2. A total of 8/16 (0.5), 8/16 (0.5), and 6/17 (0.35) participants in the 10 Hz rTMS, iTBS, and sham groups, respectively, experienced any adverse event across treatment sessions (risk difference active groups versus sham = 0.15; effect size, calculated as relative risk of adverse event in sham/active group = 0.70). Adverse events experienced across groups included tingling/prickling, headache, anxiety, nausea, back/neck pain, and facial/dental pain, with no significant differences observed across treatment groups. Mean pain scores differed significantly by treatment group, with higher ratings in the iTBS versus sham group, though average pain ratings were below 5/10 across groups. For those who started treatment, discontinuation rates did not differ across groups (Fig. 2; one from each arm), including one who was withdrawn from the rTMS group due to a medication change (no longer eligible), one who withdrew consent from the iTBS group due to scheduling conflicts, and only one who withdrew consent due to treatment intolerance, after one treatment session, from the sham group. No serious adverse events were reported. Exploratory Outcomes: At baseline, groups did not differ significantly in social cognitive, non-social cognitive, or global psychopathology, negative, or depressive symptom total scores (Table S2; all p > 0.05). The majority of participants (33/46; 71.7%) also demonstrated baseline social cognitive performance at or below the lower quartile of a large healthy adult sample on a representative measure of higher-level social cognition, and performance across social cognitive measures was comparable to a large clinically stable SSD sample (see Supplementary Results; [72]). There was a reduction in TASIT 3 sarcasm scores after iTBS versus sham stimulation and an improvement in verbal learning domain scores after rTMS versus sham, both with medium effect sizes. There were no significant differences between treatment groups in post-pre treatment change in social cognitive, non-social cognitive, or clinical scores (Table S3). Discussion The present study used individualized approaches to target brain networks associated with social cognitive processing in SSDs, while also comparing the efficacy and tolerability of both conventional 10 Hz rTMS and iTBS to sham stimulation. Our results demonstrated neural target engagement with a large effect size via significantly increased connectivity within the mentalizing network following individually-targeted iTBS, but not 10 Hz rTMS, versus sham stimulation. We did not see promising effect sizes for connectivity change between the mentalizing and simulation networks in active versus sham groups. The high trial retention, lack of treatment group-based differences in adverse events and discontinuation, and low mean pain ratings across groups support the feasibility, safety, and tolerability of our approach. The increased DMPFC-left TPJ connectivity following iTBS versus sham stimulation demonstrates target engagement of the mentalizing network, supporting the first part of our primary hypotheses. We also observed a medium non-significant effect for increased DMPFC-right TPJ connectivity following iTBS versus sham, indicative of potential broader bilateral network-level modulation, with both effects maintained in our sensitivity analyses. The DMPFC and TPJ are core nodes of the mentalizing network implicated in higher-level social cognition, and the most consistently engaged regions across theory of mind tasks [13, 14,67,68]. While both the left and right TPJ are typically considered part of this network [13,14], some evidence suggests that the right TPJ is more selectively involved in mental state inference and theory of mind, and the left TPJ in linguistic and semantic aspects of social processing [73,74]. The EA task involves emotional inference, including linguistic and semantic processing, and task performance has been associated with bilateral TPJ activation [75], consistent with bilateral involvement. Meta-analyses and fMRI studies have shown that people with SSDs exhibit mentalizing network dysfunction [76,77], with connectivity of mentalizing regions associated with social cognition and functioning [19,21,22]. Previous work also suggests that the DMPFC may represent a disconnected hub in this network in SSDs, particularly with the bilateral TPJ [18]. In contrast, we found little evidence of treatment-related changes in mentalizing-simulation network connectivity. Though our prior work linked social cognitive performance to negative connectivity between social cognitive networks [17], our main criterion for proceeding to a larger trial was change in within-mentalizing network connectivity given the strong association between higher-level mentalizing processes and functional outcomes in SSDs [2–4,78]. Our current findings likely reflect the use of the mentalizing network to define our individualized DMPFC targets (including bilateral TPJ), the EA task context, and the network-specific effects of rTMS [31], which may be heightened by personalized targeting approaches [79]. The finding that iTBS produced the largest effect versus sham is clinically meaningful, as it requires just over 3 min to administer, improving efficiency and accessibility for patients. However, the lack of efficacy of 10 Hz rTMS was somewhat unexpected. iTBS mimics endogenous theta rhythms that promote long-term potentiation and plasticity in cortical circuits [29,30]. This may be especially relevant for the mentalizing network given magnetoencephalography evidence that perspective-taking and theory of mind processes involve local theta-band oscillations and theta-band synchrony between the TPJ and MPFC [80,81]. Preliminary evidence from interleaved TMS-fMRI studies in healthy adults also indicate that iTBS and 10 Hz rTMS evoke distinct prefrontal activation patterns [82] and that iTBS may produce greater cumulative effects [83]. Additionally, in a trial comparing four DLPFC rTMS protocols for negative symptoms in people with SSDs, the TBS protocol yielded larger symptom reductions than 10 Hz rTMS [84], though further work is needed to confirm these effects and elucidate underlying mechanisms. Our results also support the safety and tolerability of personalized 10 Hz rTMS and iTBS targeting the DMPFC in SSDs. Although the iTBS group reported higher pain ratings, adverse event and discontinuation rates did not differ significantly and mean pain scores remained below 5/10 across groups, indicating good tolerability. Exploratory analyses showed medium effect sizes for reduced sarcasm detection after iTBS and enhanced verbal learning after 10 Hz rTMS relative to sham, but no significant between-group differences in cognitive or clinical outcomes. The study was not powered to detect small-to-moderate behavioral effects, and its two-week duration may have been insufficient for such changes to emerge [85]. However, recent meta-analytic evidence suggests that TMS shows promise for improving social cognition in neuropsychiatric disorders, although most studies targeted the dorsolateral PFC (DLPFC) [86]. In healthy adults, some studies indicate that rTMS to the MPFC can transiently modulate both theory of mind [87,88] and emotion recognition [89] abilities. There is also early evidence in SSDs suggesting that transcranial direct current stimulation to the DMPFC may improve mentalizing performance [90]. An ongoing, larger clinical trial, initiated following the positive neural target engagement findings from this trial, is adequately powered to detect behavioral effects and has a longer duration, with all out-of-scanner social cognitive measures serving as primary rather than exploratory outcomes to evaluate treatment-related changes in social cognition (NCT06118268). To maximize target engagement and improve the precision and interpretability of our findings, we used individualized TMS targeting and ROI definitions to account for individual differences in brain structure and function [33–35,37,38]. This trial marks the first prospective clinical application of our BOONStim pipeline [47] for personalized target identification and stimulation optimization, demonstrating its feasibility and potential efficacy. Growing evidence supports the therapeutic value of individualized TMS approaches, including both personalized connectivity mapping and electric field modelling [40,46,91,92], though prospective studies remain relatively limited and methodologically diverse. In depression, such approaches have been associated with improved response rates [41–44]. One study in SSDs used prospective personalized functional connectivity-based DLPFC targeting with accelerated iTBS, showing a reduction in negative symptoms versus the sham group, and associated connectivity changes [45]. Building on this, our findings suggest that individualized TMS targeting of the DMPFC may enhance engagement of social cognitive networks in people with SSDs. Despite the noted strengths of our approach, this was a target engagement study with some limitations. The sample size was powered to detect effects related to target engagement, but this trial was not designed to detect treatment effects on social cognitive performance nor the potential influence of sex, gender identity, and other factors, such as medication. Though illness chronicity, and the relatively wide age range, may also contribute to variability in treatment response, we observed no significant exploratory effects of age and covaried for this variable across our primary analyses. The ongoing subsequent clinical trial focused on social cognitive performance is being conducted in an earlier-phase population, which may help mitigate potential effects of illness duration and antipsychotic exposure. Across groups, belief in active treatment was similarly high, which may reflect the use of electrical stimulators across treatment arms, the 2:1 active-to-sham randomization ratio, or elevated TMS trial expectations [93]. Concerns have also been raised about the validity of personalized treatment targets based on fMRI data from a single timepoint [94,95], though recent evidence suggests that reliable and accurate individualized connectivity-guided DLPFC targets can be identified with 15–20 min of resting-state data [96]. Participants also exhibited relatively low baseline clinical symptoms due to the clinical stability inclusion criterion, which may have reduced sensitivity to detect changes in these measures. Lastly, heterogeneity of social cognitive impairments in SSDs [97] may have influenced treatment response, particularly as participation was not restricted based on the presence or severity of impairment. However, comparisons with a large cohort study [72] suggest substantial social cognitive impairment in our sample and broad representativeness of clinically stable SSD populations. Examining the effects of baseline social cognition and functional connectivity on treatment outcomes, and stratifying individuals based on level of impairment or more homogeneous biotypes may prove useful for future targeted trials and mental health resource conservation [23,98]. This target engagement study provides the first evidence that individually-targeted iTBS to the DMPFC can modulate mentalizing network connectivity in individuals with SSDs, using our personalized TMS targeting pipeline. Exploratory analyses showed non-significant, largely negligible to small, differences in social cognitive performance following active versus sham stimulation. Our findings support the feasibility, safety, and tolerability of this individualized approach. A larger and longer efficacy trial is underway to evaluate the effects of individualized iTBS to the DMPFC on social cognitive performance and functional outcomes (NCT06118268). Supplementary Material