Accelerated intermittent theta burst stimulation targeting personalized fronto-parietal control network improves core symptoms of autism spectrum disorder: a double-blind, randomized controlled trial Authors: Qi Liu, Jiawei Zhang, Xinyu Duan, Ping Zhang, Yang Yang, Guanzhong Yao, Bulideerqing Jiaerheng, Xiao-Jing Shou, Yong He, Kaiyue Han, Meixiang Jia, Lifang Wang, Weijun Gong, Wuxiang Xie, Kai Sun, Danhong Wang, Xiao-Dan Wu, Hua Cao, Hao Zhang, Hesheng Liu Published: 2026 ABSTRACT Autism Spectrum Disorder (ASD) presents a substantial global challenge, yet no pharmacological treatments effectively target its core symptoms, especially in individuals with severe cognitive or adaptive impairments. This double-blind, sham-controlled, randomized clinical trial assessed accelerated intermittent theta burst stimulation (iTBS) targeting the personalized fronto-parietal network (FPN) in ASD. Participants (6–30 years) were randomized in a 2:1 ratio to active or sham iTBS (three daily sessions, 1800 pulses/session) over 12 weeks (324k pulses) alongside behavioral training. The primary outcome was defined as the response rate, charaterized by a ≥ 1-point reduction in ADOS-2 SA at week 12. Of 132 individuals screened, 67 were randomized (mean age 10.04 ± 4.22 years; 88.1% male; all with cognitive/adaptive delays), with 59 completing the study. Active iTBS resulted in a significantly higher response rate (55% vs. 29%) and higher symptom improvement than sham (cohen’s d = −0.53), with mild local pain in only 5% of iTBS group. In the profound autism subgroup, the active group exhibited language improvement alongside amelioration of core symptoms. These findings suggest that prolonged, accelerated FPN-targeted iTBS is a safe and efficacious intervention for severe ASD, offering a promising therapeutic approach.Registration ClinicalTrials.gov Identifier: NCT05890846 Introduction Autism spectrum disorder (ASD), a neurodevelopmental condition hallmarked by early-onset social communication deficits and restricted repetitive behaviors [1], affects approximately 1% of the global population with marked regional heterogeneity (prevalence range: 0.01% to 4.36%) [2]. Convergent evidence indicated polygenic susceptibility interacting with environmental factors [3–5], culminating in circuit-level neurodevelopmental dysfunction such as excitation-inhibition (E–I) imbalance [6–8]. While early intensive behavioral interventions in preschool children [9] and social cognitive training in mildly impaired adolescents [10] show promise in improving social communication ability, and pharmacological agents demonstrate limited benefits for comorbid symptoms (e.g., irritability) [11, 12], no psychopharmacological treatment has shown robust efficacy for core symptoms in ASD [13, 14]. This treatment gap is particularly pronounced in school-aged and older individuals, especially those with profound autism and significant functional impairments, underscoring an urgent clinical need for targeted, mechanism-based interventions [15]. Repetitive transcranial magnetic stimulation (rTMS), a non-invasive technique for modulating neural activity [16], has demonstrated clinical efficacy in treating various mental disorders, including FDA-approved applications for depression [17, 18]. The efficacy of neuromodulation depends critically on targeting functionally relevant neural circuits [19, 20]. However, substantial interindividual variability in functional organization makes standardized anatomical targeting suboptimal, as identical stereotaxic coordinates may engage distinct functional networks across different subjects [21]. In ASD pathophysiology, dysfunction in the fronto-parietal network (FPN), a system governing goal-directed behavior and cognitive flexibility [22, 23], has been mechanistically linked to core symptom domains [24, 25]. Specifically, multi-level FPN dysfunction, including atypical task activation, static and dynamic functional connectivity and disruption to its hierarchical organization, has been associated with ASD [25–28]. Critically, FPN demonstrates the highest degree of interindividual variability among all large-scale functional networks, including differences in network location, size, and topology [21, 29]. Emerging evidence indicates that individualized targeting guided by functional neuroimaging enhances both spatial precision and therapeutic efficacy [30, 31]. Building on our prior individualized functional mapping technique [29, 32, 33], we developed a precision targeting protocol that identifies the target in each individual’s FPN within the middle dorsolateral prefrontal cortex (DLPFC), a node critical for contextual adaptation [34]. Given the high contextual adaptation is critical for dynamically shifting social interactions, the individualized approach directly targets the circuit-level dysfunction and enables testing its potential in treating ASD. Although rTMS shows potential for improving core symptoms of ASD, current evidence remains inconclusive due to methodological limitations, including heterogeneous protocols, small sample sizes, and reliance on subjective outcome measures [35, 36]. These constraints underscore the need for innovative protocols to address ASD’s unique neurobiology. Specifically, the chronic neurodevelopmental trajectory of ASD necessitates sustained therapeutic interventions [37], as transient effects from short-term rTMS protocols have demonstrated limited clinical durability. Intensive, optimized rTMS regimens may therefore offer superior and more enduring treatment efficacy. Moreover, optimized rTMS protocols, including theta-burst stimulation [38], higher pulse doses [39] and condensed schedules [40] have demonstrated improved therapeutic efficacy across multiple neuropsychiatric conditions [41]. For ASD, translating these approaches would require tailoring both spatial (precise targeting) and temporal (extended duration) parameters to drive lasting network changes [42, 43]. In this randomized, double-blind, sham-controlled clinical trial, we investigated whether a 12-week course of personalized, accelerated intermittent theta burst stimulation (iTBS, 324k pulses in total) targeting individually-defined FPN nodes, combined with standard behavioral therapies, would significantly alleviate core autism symptoms. Treatment efficacy was objectively assessed by research-reliable examiners using the Autism Diagnostic Observation Scale, 2nd Edition (ADOS-2) [44], featuring standardized direct observation and operationalized ASD behavior criteria. Materials and methods Study design This double-blind, randomized clinical trial employed a parallel-group design with a 2:1 allocation ratio (active iTBS vs. sham). The study protocol was approved by the Institutional Review Boards of both the China Rehabilitation Research Center (CRRC) and Henan Provincial People’s Hospital (HPPH). Signed informed consent have be obtained from all participants or their legal guardians. The study protocol is provided in Supplement 1, and the trial design is schematically illustrated in Fig. 1. This study was registered at ClinicalTrials.gov (NCT05890846).Fig. 1Trial procedure.A Personalized FPN target identification involved (i) baseline structural and functional MRI scans under sedation for each participant, followed by (ii) individualized functional network parcellation using an approach previously reported, (iii) determining the target search area by integrating FPN confidence values (top panel) and structural accessibility (bottom panel, warm colors denoting sulci), and (iv) selecting the optimal target (green dot, marked by red arrow) before (v) participants were randomized (blinded to group allocation) to iTBS or sham groups. B The clinical trial workflow began with informed consent from legal guardians, followed by screening, pre-treatment clinical evaluation, and sedated MRI scanning for recruited ASD youth; after personalized FPN target identification, participants were randomized to receive active/sham iTBS combined with standard behavioral therapy over 12 weeks (60 days). Post-treatment assessments included an sedated MRI scan and two clinical evaluations at 8 and 12 weeks. C The daily treatment schedule included 3 iTBS sessions (1 session/hour, 1800 pulses/session at 100% RMT, triplet 50-Hz bursts, repeated at 5 Hz), interspersed with two 30-minute standard 1-on-1 behavioral therapy sessions. BOLD blood oxygenation level-dependent, FPN frontal parietal network, iTBS intermittent theta-burst stimulation. Brain maps display the lateral left hemisphere of a random participant. Trial procedure. A Personalized FPN target identification involved (i) baseline structural and functional MRI scans under sedation for each participant, followed by (ii) individualized functional network parcellation using an approach previously reported, (iii) determining the target search area by integrating FPN confidence values (top panel) and structural accessibility (bottom panel, warm colors denoting sulci), and (iv) selecting the optimal target (green dot, marked by red arrow) before (v) participants were randomized (blinded to group allocation) to iTBS or sham groups. B The clinical trial workflow began with informed consent from legal guardians, followed by screening, pre-treatment clinical evaluation, and sedated MRI scanning for recruited ASD youth; after personalized FPN target identification, participants were randomized to receive active/sham iTBS combined with standard behavioral therapy over 12 weeks (60 days). Post-treatment assessments included an sedated MRI scan and two clinical evaluations at 8 and 12 weeks. C The daily treatment schedule included 3 iTBS sessions (1 session/hour, 1800 pulses/session at 100% RMT, triplet 50-Hz bursts, repeated at 5 Hz), interspersed with two 30-minute standard 1-on-1 behavioral therapy sessions. BOLD blood oxygenation level-dependent, FPN frontal parietal network, iTBS intermittent theta-burst stimulation. Brain maps display the lateral left hemisphere of a random participant. Participants ASD patients aged 6 to 30 years, with ADOS-2 score above the ASD cut-offs (module algorithm-specific cutoffs of 7, 8, or 11 points) [44] were recruited. Exclusion criteria were comorbidity with psychotic disorders, seizures, genetic disorders, severe physical diseases, metal implants, severe self-injury in the past year, receiving neuromodulation treatments in the past 3-months, and enrolling in other clinical trials. Clinical assessments and outcomes Clinical evaluations were performed at three timepoints: baseline, week 8 (mid-treatment), and week 12 (post-treatment). The primary efficacy endpoint was treatment response rate in social communication, defined as achieving a minimal clinically important difference (MCID) of a ≥ 1-point reduction in the ADOS-2 Social Affect (SA) raw score at Week 12. This MCID criterion was adapted from a large-scale clinical trial of 5-month music therapy in ASD [45]. Secondary outcomes included: (1) response rate at week 8, and (2) longitudinal changes in ADOS-2 SA calibrated severity scores (CSS) across all assessment timepoints. ADOS-2 administrations were performed by certified examiners (XJS, DDW, QL). Secondary exploratory measures included treatment-related changes in: ADOS-2 CSS (total and RRB domains), Social Communication Questionnaire (SCQ), Social Responsiveness Scale-2 (SRS-2), Repetitive Behavior Scale-Revised (RBS-R), Child Behavior Checklist (CBCL), and standardized language assessments. Blinding Non-blinded researcher (GY) randomized participants to active iTBS and sham group using stratified design to minimize imbalance in ADOS-2 modules (module 1/2 vs. 3/4) and symptom severity (ADOS-2 total CSS ≤ 7 vs. > 7), based on a randomization table generated by an independent statistician (WX). GY also assigned participants to either an active or sham stimulator (number-labeled). The clinical examiners, participants, caregivers, and other researchers were blinded. Neuroimaging The MRI data were acquired using a 3 T Philips Ingenia scanner (32-channel head coil at CRRC) and a 3 T Siemens Prisma scanner (64-channel head coil at HPPH) at baseline and post-treatment, including one 1 mm isotropic T1 scan and 30 min of functional MRI scan under sedation given limited cooperation of participants (Supplement 1). We have confirmed that sedation does not affect personalized target identification as whole-brain functional fingerprint of each individual remains identifiable during sedation [46, 47]. Personalized target localization The personalized FPN targeting methodology is illustrated in Fig. 1A. Following preprocessing of the MRI data, fMRI runs with mean head motion exceeding 0.5 mm, maximal head motion greater than 5 mm, or maximal rotational motion greater than 5 degrees were excluded from further analysis. Participants with fewer than 3 high-quality runs were excluded from the study. Then we parcellated the cerebral cortex into fine-grained functional regions [32] and then identified the left DLPFC FPN parcel as the target search area. The final stimulation target was chosen by weighing both functional relevance and practical feasibility. Functional relevance was based on how closely a voxel’s connectivity profile matched the topology of FPN derived from the parcellation, while feasibility was determined by sulcal depth to ensure the site was accessible for TMS. Intervention protocol Participants received 12 weeks of active or sham accelerated iTBS (3 daily sessions, 50 min interval between two sessions, 1800 pulses/session at 100% RMT) using neuronavigated TMS stimulators (MT20A, Neural Galaxy Inc, Beijing) equipped with double-sided coil, plus two 30-min standard behavioral therapy sessions between stimulations (Fig. 1B,C). The therapies consisted of applied behavior analysis (ABA) and speech therapy, both administered according to standard clinical protocols. Although the specific content was tailored to each patient within these guidelines, treatment fidelity was ensured through record review by a supervising therapist. Randomization was stratified by center, and all behavior therapists remained blinded to the group assignments. Statistical analyses The sample size calculation was based on pilot data, which indicated response rates of 71% in the active group and an estimated 30% in the sham group. A response was defined as a score change of ≥ 1 point (normal distribution, SD = 2). Using PASS 2021, we determined that 60 participants were needed for a two-proportion test (Fisher’s exact test, α = 0.05, 80% power, 2:1 allocation). To accommodate a 10% attrition rate, the planned enrollment was increased to 67. Efficacy analyses were performed on the modified intention-to-treat (mITT) population comprising participants who completed week-12 primary outcome assessments. Chi-square tests were used for comparing response rates (primary and secondary outcomes). Linear mixed-effects models were employed for assessing longitudinal changes in ADOS-2 SA CSS and other clinical scores. The model included three fixed effect variables (group, time and group × time interactive term), one random effect variable (participant), and no covariates. The regression coefficient for the interactive term reflects the estimated differences in symptom severity changes between groups. Sensitivity analyses were conducted in the per-protocol (PP) population, defined as participants who received at least 80% (48 days) of treatment, and completed all three assessments within 1 week of the designated date, with evaluations performed by the same ADOS-2 examiner. Mean and standard deviation were reported, if not specified. All analyses were performed using R software (version 4.2.2), with statistical significance defined as two-sided p < 0.05. Post-hoc analyses Post-hoc analyses evaluated the robustness of treatment effects and explored efficacy in individuals with profound autism (requiring very substantial support), who were understudied in previous research. The efficacy robustness analyses included additional efficacy evaluations: (1) across different ADOS-2 modules, (2) regarding the effects of potential covariates; (3) in exploratory outcome subdomains; and (4) among all participants who completed the week-8 assessment. Potential covariates, including age, sex, study centers, ADOS-2 examiners, symptom severity (ADOS-2 CSS) and baseline IQ, were first assessed using mixed models with interaction term for group and time. Significant covariates were further examined in stratified analyses; To investigate symptomatological changes induced by iTBS, item-level efficacy analyses of ADOS-2 were performed. Specifically, overlapping items across different modules were merged, and only items with at least 5 data points in either group were included. Profound autism participants were identified as those with nonverbal/minimal verbal communication (administered ADOS Module 1) or severe cognitive impairment (IQ < 50) [48]. The efficacy analyses were repeated in the profound subgroup following the main analyses. Results Between June 2022 and March 2024, from 132 screened participants with ASD, 67 met eligibility criteria and were randomized in a 2:1 ratio to active iTBS (n = 43) or sham control (n = 24) groups (CONSORT diagram in Fig. 2). All participants were severely affected by ASD, with 93% exhibiting cognitive impairment (IQ < 70) and 95% showing adaptive impairment (GAC < 70), and 88% having both. Eight participants withdrew after randomization. After quality control of the fMRI data, two participants did not pass screening because they had fewer than three high-quality fMRI runs. Additionally, one participant had one run excluded, and another participant had two runs excluded, both due to excessive head motion. All participants completed all assigned behavioral therapy sessions. The trial demonstrated excellent retention, with comparable completion rates between groups (iTBS: 88.4%; sham: 87.5%, Supplementary Table 1 presents demographic characteristics and baseline clinical measures of completed and dropout participants). Treatment allocation guesses from either ADOS-2 examiners or caregivers did not differ between active iTBS and sham groups (Supplementary Table 2). Demographic characteristics and baseline clinical measures between two groups are shown in Table 1. The personalized targets were mapped onto a standard cortical atlas (Fig. 3A), where they exhibited an average inter-individual spacing of 12.86 ± 8.33 mm (range: 0–41.92 mm).Fig. 2CONSORT flowchart.Table 1Demographic and clinical characteristics of the sample at baseline.iTBS Group (n = 43)Sham Group (n = 24)pAge(years)10.30 ± 4.379.58 ± 3.980.51Sex(Male:Female)40:319:50.12IQ( < 50:50-69: ≥ 70)31:9:318:4:20.91GAC( < 50:50-69: ≥ 70)a19:21:215:7:10.27Parent education ( 0.05). Data are presented as mean ± SD for continuous variable and number:number of categorical variables.IQ intelligence quotient, GAC general adaptive composite, ADOS-2 autism diagnostic observation schedule second edition, ADOS-2 CSS ADOS-2 the total of the comparison score, ADOS-2 SA CSS ADOS-2 social affect comparison score, ADOS-2 RRB CSS ADOS-2 restricted and repetitive behavior comparison score, SCQ social communication questionnaire, SRS-2 social responsiveness scale 2 version, RBS-R repetitive behaviors scale - revised, Language combined score of receptive and expressive language, CBCL child behavior checklist.adata missing, 1 from iTBS group and 1 from sham group.Fig. 3The iTBS treatment of FPN induced improvement in social communication symptoms in ASD.A Personalized stimulation targets (dark green dots) and the overlap of their corresponding E-fields from the iTBS (upper) and sham group (lower) were displayed on a standard left hemisphere surface. The E-field map was estimated on each participant’s individual surfaces and then projected onto a common surface to assess the overlap among participants. A relatively small proportion of areas show significant E-field overlap, indicated in yellow. B The percentage of participants who responded after 12 weeks of treatment was significantly higher in the iTBS group (55%, 21/38) than in the sham group (29%, 6/21, χ2 = 3.88, p = 0.049), but not after 8 weeks of treatment (iTBS, 45%, 17/38; sham, 38%, 8/21; χ2 = 0.24, p = 0.62). C Mean score changes of ADOS-2 SA CSS are depicted for the iTBS (red line) and the sham group (blue line) at 8 weeks and 12 weeks.The iTBS group showed a significantly faster overall improvement compared to the sham group (Group × Time interaction β = −0.3, 95% CI, −0.57 to −0.04, p = 0.03; d = 0.53), indicating efficacy in the active stimulation group. Error bars represent the standard error. CONSORT flowchart. Table 1Demographic and clinical characteristics of the sample at baseline.iTBS Group (n = 43)Sham Group (n = 24)pAge(years)10.30 ± 4.379.58 ± 3.980.51Sex(Male:Female)40:319:50.12IQ( < 50:50-69: ≥ 70)31:9:318:4:20.91GAC( < 50:50-69: ≥ 70)a19:21:215:7:10.27Parent education ( 0.05). Data are presented as mean ± SD for continuous variable and number:number of categorical variables.IQ intelligence quotient, GAC general adaptive composite, ADOS-2 autism diagnostic observation schedule second edition, ADOS-2 CSS ADOS-2 the total of the comparison score, ADOS-2 SA CSS ADOS-2 social affect comparison score, ADOS-2 RRB CSS ADOS-2 restricted and repetitive behavior comparison score, SCQ social communication questionnaire, SRS-2 social responsiveness scale 2 version, RBS-R repetitive behaviors scale - revised, Language combined score of receptive and expressive language, CBCL child behavior checklist.adata missing, 1 from iTBS group and 1 from sham group. Demographic and clinical characteristics of the sample at baseline. Variance were comparable between groups for each t-test (Levene’s test, all p > 0.05). Data are presented as mean ± SD for continuous variable and number:number of categorical variables. IQ intelligence quotient, GAC general adaptive composite, ADOS-2 autism diagnostic observation schedule second edition, ADOS-2 CSS ADOS-2 the total of the comparison score, ADOS-2 SA CSS ADOS-2 social affect comparison score, ADOS-2 RRB CSS ADOS-2 restricted and repetitive behavior comparison score, SCQ social communication questionnaire, SRS-2 social responsiveness scale 2 version, RBS-R repetitive behaviors scale - revised, Language combined score of receptive and expressive language, CBCL child behavior checklist. adata missing, 1 from iTBS group and 1 from sham group. The iTBS treatment of FPN induced improvement in social communication symptoms in ASD. A Personalized stimulation targets (dark green dots) and the overlap of their corresponding E-fields from the iTBS (upper) and sham group (lower) were displayed on a standard left hemisphere surface. The E-field map was estimated on each participant’s individual surfaces and then projected onto a common surface to assess the overlap among participants. A relatively small proportion of areas show significant E-field overlap, indicated in yellow. B The percentage of participants who responded after 12 weeks of treatment was significantly higher in the iTBS group (55%, 21/38) than in the sham group (29%, 6/21, χ2 = 3.88, p = 0.049), but not after 8 weeks of treatment (iTBS, 45%, 17/38; sham, 38%, 8/21; χ2 = 0.24, p = 0.62). C Mean score changes of ADOS-2 SA CSS are depicted for the iTBS (red line) and the sham group (blue line) at 8 weeks and 12 weeks.The iTBS group showed a significantly faster overall improvement compared to the sham group (Group × Time interaction β = −0.3, 95% CI, −0.57 to −0.04, p = 0.03; d = 0.53), indicating efficacy in the active stimulation group. Error bars represent the standard error. Efficacy For the primary endpoint, defined as ≥1-point reduction in ADOS-2 SA at week-12, the active iTBS group showed a significantly greater response rate than sham controls (Fig. 3B; 55% vs. 29%; χ² = 3.88, p = 0.049). The interim analysis at week-8 revealed no between-group difference in response rates (45% vs. 38%; χ² = 0.24, p = 0.62). However, longitudinal mixed-effects modeling demonstrated significantly accelerated improvement in the iTBS group, as indicated by ADOS-2 SA CSS reduction slopes (Fig. 3C, Table 2; Group × Time interaction β = −0.3, 95% CI −0.57 to −0.04, p = 0.03; d = −0.53). Exploratory outcomes (SCQ, SRS-2, RBS-R, CBCL, language) showed no significant group differences (Table 2), with consistent results in PP analyses (Supplement 2, Supplementary Figure 1, Supplementary Table 3).Table 2Secondary and exploratory outcomes measured at baseline, 8 weeks and 12 weeks treatment.groupnBaseline8 weeks12 weeksGroup × Time β (95% CI)dpADOS-2 SA CSSiTBS387.5 ± 1.567.13 ± 1.767.05 ± 1.83−0.3 (−0.56, −0.04)−0.530.03sham217.0 ± 1.107.52 ± 1.817.38 ± 1.24ADOS-2 CSSiTBS387.82 ± 1.527.79 ± 1.47.82 ± 1.61−0.04 (−0.3, 0.22)−0.070.76sham217.86 ± 1.628 ± 1.827.95 ± 1.24ADOS-2 RRB CSSiTBS388.74 ± 1.379.08 ± 1.129 ± 1.210.17 (−0.08, 0.43)0.330.18sham218.86 ± 1.428.57 ± 1.298.67 ± 1.46SCQiTBS3721.24 ± 6.4420.38 ± 7.1319.89 ± 6.47−0.65 (−1.52, 0.22)−0.380.14sham2022.55 ± 6.2223.4 ± 5.8923.05 ± 7.12SRS-2iTBS3776.22 ± 9.2175.65 ± 10.4175.3 ± 8.930.93 (−0.22, 2.08)0.370.11sham2180.48 ± 8.3876.95 ± 9.2877.05 ± 8.1RBS-RiTBS3717.95 ± 11.1317.27 ± 11.5916.16 ± 12.690.91 (−0.81, 2.64)0.220.29sham2123.43 ± 14.9520.52 ± 14.1619 ± 16.1LanguageiTBS3421.41 ± 16.9223.03 ± 16.5423.62 ± 17.40.67 (−0.27, 1.6)0.370.16sham2022.45 ± 17.8822.55 ± 18.5822.7 ± 19.22CBCLiTBS3660.56 ± 7.4959.69 ± 8.4057.94 ± 9.540.12 (−1.1, 1.34)00.84sham2063.5 ± 5.4160.9 ± 6.9160.9 ± 7.93 Table 2Secondary and exploratory outcomes measured at baseline, 8 weeks and 12 weeks treatment.groupnBaseline8 weeks12 weeksGroup × Time β (95% CI)dpADOS-2 SA CSSiTBS387.5 ± 1.567.13 ± 1.767.05 ± 1.83−0.3 (−0.56, −0.04)−0.530.03sham217.0 ± 1.107.52 ± 1.817.38 ± 1.24ADOS-2 CSSiTBS387.82 ± 1.527.79 ± 1.47.82 ± 1.61−0.04 (−0.3, 0.22)−0.070.76sham217.86 ± 1.628 ± 1.827.95 ± 1.24ADOS-2 RRB CSSiTBS388.74 ± 1.379.08 ± 1.129 ± 1.210.17 (−0.08, 0.43)0.330.18sham218.86 ± 1.428.57 ± 1.298.67 ± 1.46SCQiTBS3721.24 ± 6.4420.38 ± 7.1319.89 ± 6.47−0.65 (−1.52, 0.22)−0.380.14sham2022.55 ± 6.2223.4 ± 5.8923.05 ± 7.12SRS-2iTBS3776.22 ± 9.2175.65 ± 10.4175.3 ± 8.930.93 (−0.22, 2.08)0.370.11sham2180.48 ± 8.3876.95 ± 9.2877.05 ± 8.1RBS-RiTBS3717.95 ± 11.1317.27 ± 11.5916.16 ± 12.690.91 (−0.81, 2.64)0.220.29sham2123.43 ± 14.9520.52 ± 14.1619 ± 16.1LanguageiTBS3421.41 ± 16.9223.03 ± 16.5423.62 ± 17.40.67 (−0.27, 1.6)0.370.16sham2022.45 ± 17.8822.55 ± 18.5822.7 ± 19.22CBCLiTBS3660.56 ± 7.4959.69 ± 8.4057.94 ± 9.540.12 (−1.1, 1.34)00.84sham2063.5 ± 5.4160.9 ± 6.9160.9 ± 7.93 Secondary and exploratory outcomes measured at baseline, 8 weeks and 12 weeks treatment. Safety Safety findings included: one severe adverse event (SAE) unrelated to treatment (hospitalization for pneumonia in the active iTBS group), local pain (5% participants in the active iTBS group), and unexpected emotional problems in both groups (iTBS: 7% vs sham: 8%; p = 0.83). See Supplementary Table 4 for complete AE data. Post-hoc analysis Module-specific analyses (M1: n = 39, M2: n = 12, M3: n = 7; M4 excluded for n = 1) demonstrated efficacy exclusively in M1: week-12 response rate (67% vs 26%; χ² = 5.91, p = 0.02) and SA CSS reductions (Group × Time interaction, p = 0.02; d = −0.73) both favored iTBS (Supplementary Tables 5,6). Potential covariates showed no significant heterogeneity (Supplementary Table 7, all p > 0.05), precluding stratified analysis. All exploratory outcome subdomains (SCQ, SRS-2, RBS-R, CBCL, language) showed non-significant group differences, except worsen SRS-2 communication scores in the iTBS group (Supplementary Table 8). Additional efficacy analyses revealed comparable week-8 response rates between active iTBS (43%, n = 40) and sham groups (38%, n = 24; χ² = 0.16, p = 0.69). However, active iTBS showed significantly greater ADOS-2 SA CSS reductions versus sham group (Group × Time interaction, p = 0.02; d = −0.6), despite no between-group differences in exploratory outcomes (Supplementary Table 9). In item-level efficacy analyses, response to joint attention (RJA; Group × Time interaction: p = 0.004, d = −0.83) and level of engagement (ENG; Group × Time interaction: p = 0.03, d = −0.65) were significantly reduced following iTBS intervention, with medium-to-large effect sizes (Supplementary Figure 2, Supplementary Table 10 for a full list of 37 items). Subgroup analysis in profound autism We further analyzed the patients with profound autism [48]. The demographic and baseline clinical measures of profound autism patients were balanced between the active iTBS (n = 34) and sham (n = 18) groups (Supplementary Table 11). Profound autism cases (n = 52, 87% of whole sample, Supplementary Table 12) showed milder clinician-rated ADOS-2 symptoms but higher parent-reported SCQ impairment compared to non-profound ASD (n = 15), with comparable retention (88.46% vs 86.67%). The iTBS treatment showed significant efficacy in patients with profound autism, as evidenced by response rate at week-12 (iTBS 60% vs sham 25%, χ² = 5.12, p = 0.02) and ADOS-2 SA CSS reductions (Group × Time interaction, p = 0.03; d = −0.6; Supplementary Table 13). Additional efficacy were found in parent-reported SCQ core symptom (Group × Time interaction, p = 0.04; d = −0.63) and language ability (Group × Time interaction, p = 0.01; d = −1.01). The treatment effects were unrelated to age (Supplementary Table 14). Discussion This trial demonstrated that a 12-week accelerated iTBS treatment targeting the personalized FPN was well-tolerated and significantly improved core symptoms in ASD patients with cognitive/adaptive impairments. This randomized controlled trial, which is among the longest-treatment TMS studies in ASD to date [35, 49], revealed that significant response rates were achieved at 12 weeks but not at 8 weeks. This finding implies that a sustained treatment course is critical for this population. Focusing exclusively on individuals with co-occurring functional impairments, the present study targeted the clinically most challenging subgroup marked by pressing therapeutic gaps and significant research hurdles [13]. While parental dedication tends to increase with clinical severity, the risk of dropout also rises among the most severely affected individuals (Supplementary Table 1). Moreover, profound autism introduced unique methodological challenges, including intensive staffing requirements, low participant cooperation, and severe behavioral challenges. These difficulties are exemplified by the necessity of monitored sedation during MRI scans, a costly and labor-intensive procedure. Standard research protocols often fail to accommodate such needs, leading to a disproportionate focus on higher-functioning individuals in neuroimaging and non-invasive neuromodulation studies. As a result, the therapeutic potential of TMS in patients with profound autism remains uncertain. For instance, a 4-week iTBS pilot RCT in this population was discontinued early due to lack of efficacy, despite demonstrating the feasibility of daily rTMS [50]. In contrast, our trial provides the first robust evidence of rTMS efficacy for core symptoms in this high-burden cohort, offering a promising treatment avenue. The post hoc findings of efficacy in minimally verbal participants (ADOS-2 M1) and clinically meaningful parent-reported core symptom improvements within the profound autism subgroup suggest that non-invasive neuromodulation may be able to restore brain functions even in these difficult cases. Neuromodulation efficacy depends on precise circuit targeting. The present study focused on the FPN node within the middle DLPFC, a hub implicated in contextual adaptation [34]. As a high-order, late-maturing functional network [51, 52], the FPN exhibits abnormalities in various neurodevelopmental disorders [53]. ASD-related FPN dysfunction manifest as atypical task activation [25], hypoconnectivity [28], impaired dynamic switching [26], and restricted functional gradient expression during cross-modality processing [27]. The FPN retains neuroplasticity through early adulthood [54], thus neuromodulation of this network may mitigate symptoms by inducing synaptic plasticity changes and rebalancing its local excitatory-inhibitory activity [43, 55, 56]. Critically, the FPN exhibits greater inter-individual variability than any other cortical functional network [21]. To account for this variability and ensure targeting precision, we have implemented personalized FPN-guided neuronavigation in this trial (Fig. 1A). A recent RCT using continuous theta-burst stimulation targeting amygdala network reported social improvements in young children with ASD but showed a limited efficacy, probably due to the suboptimal selection of functional target [57]. Despite both protocols engaged left DLPFC, they modulated distinct circuits, highlighting the DLPFC’s complex role in ASD pathogenesis and the need to further explore its neurofunctional trajectories across development. Stimulation protocol optimization is critical to neuromodulation efficacy. Our accelerated iTBS protocol integrated precision targeting, condensed sessions, and behavioral interventions, factors potentially contributing to observed benefits. In the sham group, we did not observe clinically-meaningful improvement in core symptoms, indicating that routine behavioral training alone confers limited benefit for low-functioning ASD though its synergistic role with iTBS requires further investigation. Compared to conventional rTMS protocols, iTBS enhances cortical excitability and plasticity, potentially improving stimulation efficiency and reducing treatment duration [40, 58]. The high daily iTBS dosage (5400 pulses/day; 9 × standard protocols) and intensive repetitive sessions have demonstrated efficacy in other disorders (e.g., depression) [30, 38, 39] and was adopted in this trial. Nevertheless, the exact effect of stimulation dosage in ASD should be tested in future experiments. Other brain networks, such as the default mode network, which supports social cognition and associated ASD core deficits, also deserve further exploration. In this trial, symptom alleviation required 12 weeks to achieve statistical significance, with improvements emerging at week 8 but response rates only reaching significance at week 12. This pattern suggests cumulative neuroplastic effects necessitate sustained duration for clinical translation [59], aligning with ASD’s chronic neurodevelopmental trajectory [37] and parallels with intensive behavioral therapies requiring extended implementation [35, 60]. Prior inconsistent neuromodulation outcomes may be due to inadequate treatment duration. For example, a meta-analysis of 16 published non-invasive neuromodulation trials showed that none of the trials with low session numbers yielded core symptom improvement, except for a tDCS protocol delivering ≥ 10 sessions over 2–4 weeks [35]. Our condensed high-frequency regimen (180 sessions over 12 weeks) likely augmented neural plasticity [61] and contributed to the observed treatment effects, underscoring the need to optimize both protocol parameters and duration for neurodevelopmental disorders. Finally, it is worth mentioning that rigorous behavioral measures are crucial in the clinical research of ASD. In our trial, efficacy was measured using the clinician-rated ADOS-2, which has been validated as a primary outcome in landmark ASD psychosocial RCTs due to its standardized administration and rigorous global research-reliability framework [45, 62]. Recent noninvasive neuromodulation meta-analyses indicated that clinician-rated measurement is rarely used as an outcome measure [35, 49]. This is probably due to the lengthy administration (≥ 1.5 h/assessment) and evaluator training requirement of ADOS-2. However, clinician-rated assessments offer superior reliability to subjective parent-reported measures [14, 63], as the latter may be confounded by contextual factors, parental expectations, or symptom focus [64]. Given the challenges in implementing clinician-rated measures and instability of subjective scales, development of objective biomarkers (e.g., neural functional markers, eye-tracking) is critical for advancing mechanistic understanding and elucidating symptom changes. Limitations Several limitations warrant consideration. First, the restricted sample size and high heterogeneity may constrain the generalizability of our findings. The modest effect size observed here suggests that efficacy may be limited in this severe population. Therefore, a key future direction is to evaluate this paradigm in patients with less severe symptoms to better elucidate its potential. Second, COVID-19 restrictions prevented follow-up data collection, precluding assessment of efficacy durability. Third, routine 1-on-1 therapies were highly individualized to participant performance levels and extended treatment duration, necessitating intensive commitment from families and clinical teams. Four, multiple comparisons in exploratory outcome analyses and post-hoc analyses may increase risk of false positives, so caution should be exercised when interpreting the results. Five, the broad age range of participants could confound results due to age-related neuroplasticity differences. While our results indicated that the treatment response was uncorrelated with age, neuroplasticity variations across age groups should be further explored. Finally, the scanning-related confounding variables (e.g., sedative drugs and scanning sites) may also introduce confounds in MRI data and influence target identification [46, 47]. Conclusion In summary, this RCT demonstrates that 12 weeks of accelerated iTBS targeting the individualized FPN is safe and effective for improving core symptoms of ASD, suggesting a promising neuromodulation approach for ASD patients with co-occurring functional impairments. Confirmatory trials with larger samples, long-term follow-up, and focus on severely affected ASD individuals are essential to validate efficacy and durability. Supplementary information Trial protocol Supplementary methods, results, figures and tables Trial protocol Supplementary methods, results, figures and tables