Fourteen patients with BPD (3 men; age: M = 23.57, SD = 4.73 years) underwent a prefrontal rTMS protocol. The inclusion criteria were: 5 out of 9 criteria for BPD according to DSM-5 (1), age between 20 and 45, and stable medication for at least 6 weeks before the stimulation and until the end of the stimulation. The exclusion criteria were addiction (except nicotine), bipolar I disorder, and major depressive disorder (current or in the history of the patient), current acute psychotic state, a schizophrenia-spectrum disorder, and contraindications preventing MRI or rTMS.

Self-reported measures were included to assess the clinical effect of the treatment in areas that were hypothesized as possibly improvable by rTMS. Borderline symptoms were measured by the Borderline Symptom List 23 (BSL-23), an established measure for BPD symptom severity (29). Impulsivity was measured by the UPPS-P scale (30, 31), specifically the subscales Lack of Premeditation, Lack of Perseverance, Negative Urgency, and Positive Urgency. Sensation Seeking was left out since it does not seem to be altered in BPD (32). Emotion regulation was measured by the Difficulties in Emotion Regulation Scale [DERS; (33)], and anxiety was measured by the Zung Self-Rating Anxiety Scale [SAS; (34)]. Depression symptoms were measured by the Montgomery and Åsberg Depression Rating Scale [MADRS; (35)] assessed by a clinical psychologist who did not participate in the rTMS patients treatment. The questionnaires were administered to the patients before and after the stimulation protocol. Possible side effects of rTMS were evaluated by research staff before and after each stimulation session, based on the recommendations of McClintock et al. (36).

Functional MRI was done immediately before and after the rTMS protocol with 3T Magnetom Siemens Prisma machine (Siemens Healthcare GmbH, Erlangen, Germany) at the Central European Institute of Technology (CEITEC) in Brno, Czech Republic. The acquisition sequence was three-dimensional magnetization-prepared rapid gradient echo [3D inversion recovery; anatomical T1; 3D MP-RAGE; Repetition time (TR) = 2,300 ms, Echo time (TE) = 2.33 ms, voxel size = 1 × 1 × 1 mm³, FoV (224 × 224 mm), Flip Angle (FA) = 8°, 240 sagittal slices]. We used a gradient-echo echoplanar-imaging sequence sensitive to BOLD contrast for acquisition of one task-free and four task fMRI sessions (GNG task) with the following task-free/task parameters: TR = 750/2,280 ms, TE = 35 ms, voxel size = 3 × 3 × 3 mm³, FoV (192 × 192 mm), FA = 46°/75°, multiband factor = 4/1, 39/40 transversal slices, 380/153 scans.

We adapted the GNG task according to the task design from Albares et al. (37). In the beginning of each trial, a variable duration fixation cross was presented for 2–6 s, followed by a Go or NoGo stimulus lasting 0.2 s, and finally followed by a post-trial black screen for 2 s. We used white letters A and B on a black background as the Go and NoGo stimuli. The fixation cross was green in 1/3 of cases and red in 2/3 s of cases. The patients were instructed that a Go or NoGo stimulus can appear after the red cross, while only the Go stimulus can appear after the green cross (Go condition, NoGo condition, and Go-control condition; Figure 1). The ratio of Go and NoGo stimuli occurring after the red cross was equal. The patients were further instructed to press a button as quickly as possible when a Go stimulus appeared, but not to press the button when a NoGo stimulus appeared (i.e., to perform inhibition). The task contained 4 blocks of 54 trials each with breaks between the blocks; the whole task lasted ~25 min. NoGo vs. Go contrast was used to analyze the neural correlates of inhibition.

The rDLPFC was targeted individually by stereotactic neuronavigation (software Brainsight, v. 2.02), fMRI data analysis for neuronavigation was performed in SPM12. The rDLPFC anatomical mask was derived from the Destrieux neuroanatomic atlas (38) from FreeSurfer software (39) as a conjunction of the sulcus frontalis inferior, sulcus frontalis medius, and gyrus frontalis medius. Regions of interests were further specified by creating a sphere mask with radius 15 mm around the maximum in rDLPFC derived for the contrast between NoGo and Go conditions and by generating the mask for an area up to 20 mm under the skull. Within the brain area defined by combining those masks, the point of individual peak BOLD signal in the contrast between NoGo and Go conditions was used as rTMS target.

We used the DuoMAG XT magnetic stimulator (Deymed Diagnostic) with a 70BF air-cooled coil. The patients underwent 15 stimulation sessions at 110% of their individual resting motor threshold (RMT) over a period of 3 weeks with one session each working day. One stimulation session contained 1,500 pulses (total 22,500 pulses during the whole procedure) divided into 10 trains with 10 Hz frequency (train interval lasted 10 s, inter-train interval lasted 30 s). The RMT was measured before the first stimulation session and defined according to the Rossini-Rothwell method (40).

Behavioral and self-reported data analysis was performed in jamovi software v1.6 (41). Differences before and after the rTMS treatment in self-report questionnaires and MADRS scores and in the relative frequency of mistakes after NoGo stimuli in the GNG task (NoGo errors) were analyzed by paired t-tests and non-parametric Wilcoxon rank test due to small sample size.

According to the self-reported measures (Table 1), rTMS led to significant reduction in Lack of Premeditation and Lack of Perseverance in the UPPS-P impulsivity scale, improvement in emotion regulation (DERS), reduction of depressive symptoms (MADRS), and borderline significant reduction of anxiety symptoms (SAS). There was no significant change in BSL-23, negative urgency and positive urgency, and no significant change in the percentage of NoGo errors in the GNG task after vs. before the treatment. Parametric and non-parametric tests yielded similar results.

In the resting state after rTMS vs. before rTMS, we found a decrease in connectivity of the left amygdala with a cluster comprising the right precuneus and superior parietal lobule [peak 27 −70 37 (t = 5.00, p = 0.025), 78 voxels; Figure 2A]. No significant changes in the right amygdala, insula, or right DLPFC connectivity were observed.

In the GNG task, specifically in the NoGo vs. Go contrast and after vs. before rTMS, we found decreased connectivity of the left amygdala with a cluster comprising the left parietal lobule, precuneus, and middle and posterior parts of cingulate gyrus [peak −15 −67 13 (t = 5.99, p = 0.000), 374 voxels; Figure 2B]; and a decrease in the connectivity of the left insula with the left precuneus [peak −9 −67 19 (t = 5.96, p = 0.017), 67 voxels; Figure 2C], with the right precuneus [peak 12 −64 25 (t = 5.41, p = 0.014), 70 voxels; Figure 2D], and with the posterior cingulate gyrus [peak 6 −40 28 (t = 5.11, p = 0.010), 74 voxels; Figure 2E]. No significant changes in the right amygdala, right insula, or right DLPFC connectivity were observed.

We observed two trend associations of connectivity change and symptom change (Table 2). Decreased connectivity in the left amygdala with a cluster comprising the right precuneus and right superior parietal lobule in the resting state was positively correlated with decreased depression symptoms (MADRS; r_s= 0.52, p = 0.067) and Lack of Premeditation (UPPSP-PRE; r_s= 0.53, p = 0.064). We observed another negative correlation of decreased Lack of Perseverance (UPPSP-PER; r_s= –.49, p = 0.076) and decreased left insula connectivity with the right precuneus in the inhibition condition.

Regarding the performance in inhibition in the GNG task, the number of NoGo errors was low both before and after rTMS, and it did not significantly change. A small number of mistakes is normal in equiprobable tasks such as the one used in our study (28, 49). Regarding the self-reported results, we observed some improvement in all measured variables after the rTMS protocol. The improvement reached statistical significance in emotion regulation (DERS), depression (MADRS), and two UPPS-P impulsivity subscales—Lack of Premeditation and Lack of Perseverance (in other words, the abilities to think before acting and to persist in activities and decisions) and borderline significance in anxiety (SAS). High effects sizes were observed for the changes in Lack of Premeditation and emotion dysregulation, medium effect sizes were observed for changes in Lack of Perseverance and in anxiety and depression levels. The results are consistent with previous studies, specifically the decreased anxiety (19, 22), depression (14, 15, 18–20, 22), impulsivity (15, 19, 20) and affective instability (16). Overall, rTMS treatment had positive effects on some BPD symptoms including anxiety, depression, emotion dysregulation, and impulsivity. Interestingly, only one study (16) used high-frequency stimulation of the right DLPFC such as in our study; other studies used low-frequency stimulation of the same area (19), or high-frequency stimulation of the left DLPFC (14, 15, 19, 20), or DMPFC stimulation (18, 22). Our results are consistent across the previous studies (23). At least two interpretations might be considered here: either the clinical effects are caused by factors other than active stimulation, which might indicate a placebo effect, or the parameters of the stimulation protocol might not play a crucial role in the results, and the rTMS might instead act as a more global prefrontal intervention (thanks to cortico-cortical connections).

In both the resting state and inhibition conditions, we found decreased left amygdala connectivity with the precuneus (PCN), posterior cingulate cortex (PCC), and parietal lobules (PL) after rTMS. These regions are parts of the posterior default mode network (pDMN), which is associated with self-related (personal) and future-oriented reflection and thinking and emotional and attributional evaluation of situations (50, 51).

Previous studies reported increased involvement of pDMN in patients with BPD as compared to healthy controls, suggesting higher self-referential thinking (6, 7). Increased connectivity of the amygdala with pDMN areas has been found to be associated with a number of negatively experienced phenomena, such as depressive rumination [PCC (52)], sleep deprivation [PCC, PCN (53)], complicated grief [PCN (54)], and also history of abuse [PCC, PCN (55)] or risk-taking [PCC, PCN (56)]. Increased connectivity of the amygdala with PCN has been also identified in adolescents who attempted suicide or had high suicidal ideation as compared to adolescents with low suicidal ideation and healthy controls (57).

Importantly, decreased amygdala connectivity with pDMN (in resting state only) was positively correlated with decreased depressive symptoms and with decreased lack of thinking before acting and planning (Lack of Premeditation). This means that the larger the decrease in amygdala-pDMN connectivity in the resting state after rTMS as compared to before rTMS, the greater the improvement in depressive symptoms and premeditation ability. Thus, our results are consistent with previous literature reporting positive associations of amygdala-pDMN connectivity and borderline or depressive symptoms. Decreasing amygdala-pDMN connectivity seems to be beneficial in patients with BPD and was achieved after the prefrontal rTMS protocol in our study.

In the inhibition condition, we found decreased connectivity of the PCN and PCC with the left insula after rTMS as compared to before rTMS. Again, there was a decrease of pDMN areas connectivity. The insula processes interoceptive information to engage externally oriented attention and internal cognitive control, helps with integrating information from the limbic system, and is a major part of the inhibition network (24, 58, 59). The PCC, PCN, and insula have been found to be associated with an increased sense of personal agency, with the insula correlating specifically with personal agency in negative situations (50). Higher interconnectedness in the PCN and insula has been found in people with a history of childhood maltreatment (60). In another study, increased connectivity of the insula and PCN distinguished posttraumatic stress disorder patients with and without dissociation symptoms (61). The pDMN has been found to be deactivated during pain, while the insula is activated by pain (62). Increased pDMN-insula connectivity has been reported in patients with BPD during pain processing (63); while decreased pDMN-insula connectivity has been observed in healthy people (64), which might reflect a disturbance in pain processing in BPD and suggest more self-referential experiencing of pain in BPD (63).

In our study, we found a negative correlation of decreased insula-pDMN connectivity in the inhibition condition and symptom reduction—specifically with reduced Lack of Perseverance. Though previous studies suggest that decreasing insula-pDMN connectivity could be beneficial in patients with BPD, our data did not support the hypothesis that reduction in this connectivity is positively associated with symptom changes in patients with BPD. Moreover, it should be noted that we found the decrease of pDMN-insula connectivity only in inhibition condition and not in resting state. Thus, the effect might be more situation specific or task related than general, while previous studies usually report increased pDMN-insula connectivity in resting state or during pain processing.

Overall, previous studies suggest that increased connectivity of pDMN with the amygdala and insula is associated with increased self-referential and self-attributional thinking, especially in negatively perceived situations. These connectivity increases are then associated with symptoms commonly associated with BPD, such as depressive mood, risk-taking, and suicidality with pDMN-amygdala connectivity, and dissociation and disturbed pain processing with pDMN-insula connectivity. Increased connectivity in both cases have been linked to adverse experiences in childhood. Our study reported decreased pDMN-amygdala and pDMN-insula connectivity and decreased BPD symptoms. We observed decreased pDMN-amygdala connectivity both in the resting state and during an inhibition task. This result suggests that negative self-referential thinking was decreased during resting and during a task activity focused on impulse control. We found that symptom reduction was positively correlated with decreased amygdala-pDMN connectivity, but negatively correlated with insula-pDMN connectivity. Thus, the reduction of negative self-referential thinking associated with reduced pDMN connectivity with the amygdala seems to be a beneficial effect for patients with BPD and a candidate mechanism of the rTMS effect in patients with BPD.

In contrast to our initial hypothesis, we did not find any changes in the connectivity of the dorsolateral prefrontal area stimulated by rTMS. Thus, the mechanism of how prefrontal stimulation led to the connectivity and symptom changes observed in this study remains unclear. However, changes in DMN connectivity after high-frequency prefrontal stimulation were found in previous studies in healthy volunteers and in patients with depression and PTSD (65–67). Another study examining the neural effects of high-frequency DLPFC stimulation in 60 healthy people found no changes in DMN connectivity, but multiple increases in the connectivity of the cingulate and frontoparietal areas (68); this was not observed in our study. Several possible interpretations might be drawn concerning how prefrontal stimulation can induce distant connectivity changes without influencing the stimulation area connectivity per se. First, it is possible that the effects are induced by factors other than the intervention, including a placebo effect. Double-blind sham-controlled studies are needed to shed more light on this possibility. Second, prefrontal rTMS might actually induce distant beneficial brain changes, which would correspond with the hypothesis that the precise stimulation site might not be a crucial parameter at all. Third, it is possible that the current study did not have sufficient power to show prefrontal connectivity changes, and replication on a higher sample is needed. Moreover, medication in patients with BPD might influence the results since limbic hyperactivity has been found to be moderated by medication status in BPD patients (11). Overall, no existing data currently support the hypothesis that prefrontal rTMS leads to symptom reduction by increasing the prefrontal-limbic connectivity, but symptom reduction might be achieved by prefrontal rTMS through distant brain connectivity changes, such as the decrease of pDMN-amygdala connectivity observed in this study.

The major limitation of our study is the small sample size and absence of a control group. The placebo effect is usually relatively high in rTMS studies (69), and there is a risk that this study was influenced as discussed above. On the other hand, this is the first study to report the neural effects of rTMS in BPD patients. Our results should be taken as exploratory and hypotheses-generating; they should be verified in a placebo-controlled study.

Another possible limit is the choice of behavioral paradigm for individual neuronavigation. We chose this task to identify a DLPFC target most strongly related to impulse control within the aim to reduce impulsivity. However, patients with BPD typically show impulsive behavior under the influence of emotions (32, 70, 71). A different task activating inhibition under emotions or emotion regulation could be more appropriate for targeting rTMS in patients with BPD. On the other hand, given the existing studies of rTMS in BPD, the rTMS target does not seem to play any crucial role in achieving symptom improvement. We only observed changes in connectivity from the right DLPFC, but we did not analyze changes in connectivity within the whole prefrontal cortex. This could be also the reason we observed only distant changes in brain connectivity.