Electrophysiological Biomarkers Reflect Target Engagement and Response Using Deep Brain Stimulation for Obsessive-Compulsive Disorder

Abstract

Background

Deep brain stimulation (DBS) of the anterior limb of the internal capsule (ALIC) is an effective treatment for severe, treatment-resistant obsessive-compulsive disorder (OCD). However, optimizing lead placement and stimulation parameters remains a challenge. DBS evoked potentials (EPs) recorded with electroencephalography (EEG) during surgical lead placement could serve as intraoperative biomarkers for target engagement and clinical efficacy.

Methods

We obtained intraoperative EEG recordings on the forehead from 10 patients (2 nonresponders) undergoing ALIC DBS surgery for OCD. Monopolar stimulation at 2 Hz was delivered through all electrode contacts, and EEG EPs were analyzed in relation to stimulation contact, white matter connectivity to the prefrontal cortical regions of interest (assessed via probabilistic tractography), and reduction in symptom severity (assessed with the Yale-Brown Obsessive Compulsive Scale).

Results

We observed consistent DBS EPs with 3 oscillatory peaks (∼35, ∼75, and ∼120 ms) across all patients. EP amplitude varied across contacts, with the largest responses occurring when the location of stimulation overlapped with the preoperatively defined tractographic target. Higher EP amplitudes recorded on the forehead correlated with greater white matter connectivity to the ventromedial prefrontal cortex/orbitofrontal cortex and ventrolateral prefrontal cortex. Treatment nonresponders exhibited less consistent EP waveforms across lead contacts.

Conclusions

These findings suggest that intraoperative EPs provide valuable insights into ALIC DBS target engagement. EP characteristics may serve as biomarkers to refine DBS targeting and predict clinical response, offering a potential tool for optimizing DBS therapy for OCD.

Plain Language Summary

This study shows that stimulation-evoked potentials, cortical responses recorded during deep brain stimulation (DBS) surgery, track with white matter pathways thought to mediate therapeutic effects in obsessive-compulsive disorder. At implantation, these brain signatures differed in patients who did not respond as well to treatment, highlighting the promise of this approach for optimizing care with DBS.

Plain Language Summary

This study shows that stimulation-evoked potentials, cortical responses recorded during deep brain stimulation (DBS) surgery, track with white matter pathways thought to mediate therapeutic effects in obsessive-compulsive disorder. At implantation, these brain signatures differed in patients who did not respond as well to treatment, highlighting the promise of this approach for optimizing care with DBS.

Obsessive-compulsive disorder (OCD) is a chronic neuropsychiatric disorder characterized by persistent distressing thoughts (obsessions) and repetitive behaviors or mental acts (compulsions). OCD affects 1% to 2% of the global population (1,2) and can severely impair daily functioning and quality of life (3). Up to 20% of patients with OCD do not respond to pharmacological treatments and cognitive behavioral therapy (4). For these severe treatment-resistant cases, deep brain stimulation (DBS) has emerged as a promising therapeutic option, with the anterior limb of the internal capsule (ALIC) being the most commonly used target. ALIC DBS has variable response rates between 20% and 70% in an estimated total of 270 patients and requires lengthy trial-and-error parameter optimizations (5, 6, 7, 8), which underscores the need for better methods to optimize targeting and stimulation parameters.

The ALIC serves as the major highway connecting the cortex with subcortical structures including the thalamus, subthalamic nucleus (STN), zona incerta, and midbrain (ventral tegmental area, substantia nigra) (9). High-frequency DBS is thought to disrupt OCD-related hyperactivity within these ALIC connections (10). Meta-analytic approaches have used population-level diffusion tensor imaging (DTI) data to retrospectively derive sub-tracts within the ALIC associated with better DBS response (11). More recently, patient-specific tractography has been prospectively incorporated into the surgical planning process (12). This mirrors the decades-long trajectory of discovery in subcallosal cingulate DBS for depression, where a connectome-defined target ultimately guided successful intervention (13,14).

Tractography defines the anatomical road map for DBS, but direct confirmation of functional or structural pathway engagement with DBS, intraoperatively and longitudinally, calls for a novel, complementary approach. Recent work combining diffusion-weighted imaging and electroencephalography (EEG) with DBS for depression has demonstrated that the brain’s cortical evoked response to stimulation may provide such a dynamic readout of target engagement (15,16). DBS evoked potentials (EPs) are measurable time-locked electrical responses to stimulation. Despite incomplete understanding of their biophysical basis, EPs vary systematically with stimulation site and connectivity strength, suggesting sensitivity to both functional network engagement and the structural pathways (17, 18, 19). These properties support the application of stimulation-induced signals as a potential readout of white matter target engagement for psychiatric DBS.

DBS EPs bridge the gap between tractography-based targeting and functional confirmation, offering a unique opportunity to personalize and optimize DBS therapy. In Parkinson’s disease, EPs have already shown clinical utility. Cortical EPs (cEPs), measurable via scalp EEG, reflect stimulation-evoked network responses and have been used to probe cortical circuit dynamics. In that context, cEPs have captured antidromic activation of the hyperdirect pathway resulting from STN DBS (20,21). These potentials provide insights into cortical circuit dynamics and can be used to guide electrode placement and refine stimulation settings (22, 23, 24).

Despite their promise, EPs have not yet been systematically studied in the context of ALIC DBS for OCD. In this study, we aim to 1) evaluate the feasibility of intraoperative EP recording using scalp EEG during ALIC stimulation, 2) characterize their spatial and temporal features, and 3) examine their relationship to underlying white matter connectivity and clinical response. Together, these findings may support the development of EP-based biomarkers to complement imaging approaches and improve patient-specific targeting and therapy optimization in DBS for OCD.

Methods and Materials

Cohort

We included 10 patients with OCD undergoing awake ALIC DBS implantation surgery at Mount Sinai Hospital (R01MH123542). All patients provided informed consent after receiving all information.

Surgery

Bilateral directional leads (SenSight B33015 1.5 mm spacing, Medtronic) were implanted in the ALIC following the surgical team’s standard clinical practice (12), with patient-specific tractography-based targeting guided by a responder common map (12). In brief, the optimal target location was based on a template map of ALIC connections involved in previous responders to DBS for OCD, including projections to the ventromedial prefrontal cortex (vmPFC)-orbitofrontal cortex (OFC) and ventrolateral PFC (vlPFC) and subcortical projections to the medial and lateral midbrain. The optimal target maximally engaging all these ALIC projections was located around the anteromedial globus pallidus pars externa (GPe). Left- and right-hemisphere electrodes were implanted during separate surgeries 1 month apart. During the procedure, propofol infusion was titrated to achieve deep sedation (50–150 μg/kg/min). Infusions were discontinued at the time of incision, ensuring that propofol levels were negligible during all electrophysiological recordings.

Recording and Stimulation

EEG was recorded from 4 forehead electrodes (FP1, FP2, AF7, AF8) at 22 kHz, referenced to the right mastoid or canula. As depicted in Figure 1, monopolar stimulation was delivered successively from each contact (four 1.5-mm contacts per lead, from ventral to dorsal named C0, C1, C2, C3) including each posterior, medial, and lateral directional segment of the middle 2 contacts (named C1/2 a, b, and c, respectively). Stimuli comprised cathode-first symmetrical biphasic pulses (90-μs pulse width per phase) delivered at a stimulation frequency of 2 Hz for 90 seconds. This stimulation frequency was selected to capture long-latency network responses (up to 500 ms). In 1 patient (P10), stimulation was delivered at 3 Hz due to intraoperative time constraints. Amplitude was set to 3.5 mA for directional and 5.5 mA for ring contacts, respectively, to approximate equal voltage despite impedance differences across contact types (25).

Figure 1.

Intraoperative recording setup. A directional lead was implanted in the anterior limb of the internal capsule. The caudate (purple), globus pallidus externus (GPe), and nucleus accumbens (NAc) (both blue) are visualized as anatomical landmarks. Green fibers represent the connections to the prefrontal cortex. Low-frequency (2-Hz) stimulation was applied from all contacts (2 ring contacts, 6 directional contacts) consecutively for 2 minutes. Brain responses to the stimulation were recorded from forehead electroencephalography (EEG) electrodes (AF8, FP2, FP1, and AF7).

Signal Processing

All data processing was performed in MATLAB (R2024a; The MathWorks, Inc.) with custom scripts. A bipolar referencing scheme was employed using 2 electrode pairs, AF7-AF8 for the left hemisphere and AF8-AF7 for the right hemisphere. Baseline drift was removed using a second-order Butterworth high-pass filter (0.5 Hz) to enable trial-level segmentation. The signal was epoched by time-locking to the stimulation artifact, and EPs were obtained by averaging across trials. To remove the stimulation artifact, the signal was linearly interpolated between −1 ms and 4 ms. Finally, a second-order Butterworth bandpass filter with cutoff frequencies of 5 Hz and 50 Hz was applied to remove high-frequency noise and optimally visualize evoked responses. Given the novelty of the signal and absence of established preprocessing guidelines for ALIC EPs, filtering parameters were chosen based on visual inspection, with the goal of maximally suppressing noise while preserving the observed oscillatory components.

Feature Extraction

To characterize the evoked waveform, peak and trough latencies were extracted from EPs for all stimulation contacts (8 per hemisphere) across all 16 hemispheres using a peak detection algorithm (MATLAB, findpeaks). Peak and trough detection was based on prominence, defined as the amplitude difference between the identified peak or trough and the lowest/highest amplitude of the signal before a peak/trough of a larger magnitude was reached. A threshold for detection was set at a prominence of 5 times the standard deviation of the evoked response within the time window of 400 ms to 450 ms poststimulus. This approach ensured robust EP feature detection while minimizing noise.

The magnitude of the EP was quantified using the area under the curve (AUC) between 10 ms and 100 ms poststimulus. The AUC was calculated from the absolute value of the EP using the trapezoidal method (trapz, MATLAB) to account for the total neural response within this predefined time window. Stimulation was delivered at 5.5 mA for ring contacts and 3.5 mA for segmented (directional) contacts, resulting in significant differences in the volume of tissue activated (VTA). The volumes of the resulting VTA are approximately proportional to the square of the stimulation current when estimating the volume using the SimBio/Fieldtrip method (26). To allow for comparison of EP magnitude between the different contacts, the raw AUC was corrected using the following formula:

$$AU{C}_{corr}=\frac{AUC}{I^2}$$

where AUC_corr is the corrected AUC, AUC is the raw AUC, and I is the stimulation amplitude in mA. This normalization accounts for the differences in stimulation intensity and thus VTA size, allowing for a standardized comparison of the magnitude of evoked neural responses across different contact types.

To evaluate the consistency of the EP shape following stimulation to different contacts within a single lead, the average within-lead correlation was calculated. For each pair of contacts, the correlation coefficient between their respective EP waveforms was computed. These pairwise correlation coefficients were then averaged to obtain a single value representing the overall coherence of the EPs within the lead. This method provided a measure of the similarity in the EP signals across different contacts within the same lead.

Tractography

High-resolution multishell diffusion-weighted magnetic resonance imaging (MRI) data were collected in a 7T Siemens scanner and preprocessed in MRtrix 3 [v.3.0.4 (27)]. Postoperative computed tomography scans and the Lead-DBS toolbox [version 2.6 (28)] were used to localize DBS leads (29). VTAs were estimated via the SimBio/Fieldtrip method (26) with specified stimulation amplitudes and contact configuration (5.5 and 3.5 mA; ring and directional contacts, respectively). Fiber orientation distributions were modeled with BEDPOSTX, and probabilistic tractography [FSL probtractx2 (30)] was performed from each VTA to the vmPFC-OFC, vlPFC. Streamline counts from each VTA to cortical targets were used to evaluate the relationship between structural connectivity and stimulation EPs. More details on tractography methods are provided in the Supplement.

Clinical Measures

OCD symptoms were assessed with the Yale-Brown Obsessive Compulsive Scale (Y-BOCS) (31) both preoperatively and during treatment by a trained rater. Full treatment responsiveness was defined as achieving a reduction of at least 30% in Y-BOCS scores at any point during treatment (32).

Statistical Analysis

For all statistical tests, a significance level of .05 was used. Where necessary, corrections for multiple comparisons were applied using the Bonferroni method, and the Bonferroni-corrected p values are reported. Nonparametric tests were used when the data did not meet the assumptions required for parametric tests.

To compare EP characteristics such as AUC_corr and peak and trough latencies across different stimulation levels, a Friedman test was conducted to assess overall differences. In cases where the Friedman test indicated significance, post hoc pairwise comparisons were performed using the Wilcoxon signed-rank test. To obtain a single value for each level, the average value was calculated across the 3 directional contacts for directional levels.

A linear mixed-effects model was used to evaluate the association between EP magnitude and white matter connectivity to the vmPFC-OFC and VLPFC, respectively. The analysis was conducted using MATLAB’s fitlme function. The outcome variable was the number of streamlines to the respective region of interest, while the predictor variable was the EP magnitude (AUC, not corrected). Random effects included hemisphere, grouped by patient, to account for intersubject and interhemisphere variability.

To compare the average within-lead coherence between responders and nonresponders, a Mann-Whitney U test was performed.

Results

Demographics

A total of 10 patients (4 female/6 male, mean age ± SD = 29.1 ± 6.4 years) were included. Baseline Y-BOCS scores ranged from 24 to 38, with a mean score of 30.2 ± 4.1. Two participants (P03 and P07) did not achieve response criteria (≥30% Y-BOCS reduction) during the follow-up period of 12 months after surgery. Follow-up for P04 was limited to 6 months due to death unrelated to DBS or OCD, and follow-up for P10 was limited to 10 months because the 12-month postoperative time point had not yet been reached. Demographic data is summarized in Table 1. An overview of treatment response over time for all patients can be found in Supplemental S.1.

Table 1.

Cohort Demographics

Patient ID	Sex	Age at Time of Surgery, Years	Hemispheres Tested	Baseline Y-BOCS
P01	Male	40	L R	30
P02	Male	31	L R	26
P03	Male	20	L R	38
P04	Female	33	L	31
P05	Female	25	L R	35
P06	Male	22	L	28
P07	Female	26	L R	24
P08	Female	31	L R	32
P09	Male	26	L R	29
P10	Male	37	L Ra	29
	4 Female/6 Male	29.1 ± 6.44b	10 L/8 R	30.2 ± 4.10b

L, left; R, right; Y-BOCS, Yale-Brown Obsessive Compulsive Scale.

^a3-Hz stimulation.

^bMean ± SD.

ALIC DBS-Evoked Responses

The ALIC DBS EPs recorded from the forehead were characterized by 3 distinct oscillatory peaks with typical timings around 35, 75, and 120 ms (P35, P75, and P120), accompanied by troughs at approximately 20, 60, and 100 ms (N20, N60, and N100). P35 and N60 were most reliably detected, with P35 being observed in 48 of 70 stimulating levels (68.57%) and N60 in 54 of 70 levels (77.14%). Stimulation at either of the middle 2 DBS lead contacts located around the tractography-based target (C1 and C2) generated EP peaks more consistently, accounting for 55.9% of all detected peaks and troughs. This spatial specificity suggests anatomically specific activation of the target pathways.

An example ALIC DBS EP generated by stimulation from contact C3 in P207 is shown in Figure 2A, illustrating the typical waveform pattern observed across participants. Histograms displaying the timing of all detected peaks (Figure 2B) and troughs (Figure 2C) from all contacts across all patients further highlight the consistency in peak timing, suggesting a stable and reproducible EP morphology across different individuals and contacts. All recorded EPs across contacts for all patients can be found in Supplemental S.2.

Figure 2.

Consistent cortical responses to unilateral stimulation of the anterior limb of the internal capsule of patients with obsessive-compulsive disorder are reliably recordable. (A) Example of a recorded evoked potential from patient number 4 (P04) (left hemisphere, contact C3). Distribution of (B) peak latencies and (C) trough latencies from all recorded contacts from all hemispheres (18 hemispheres). Peaks and troughs were detected if their prominence exceeded the threshold of 5 times the standard deviation of the evoked response between 400 ms and 450 ms poststimulus. Peaks are observed consistently at around 35, 75, and 120 ms, and troughs are observed at around 20, 60, and 100 ms.

EP Features Over Stimulating Levels

ALIC EP amplitude and latencies of the 3 consistent peaks and troughs were compared across stimulating levels within individual leads (Figure 3). EP amplitude varied significantly across levels (Friedman test, p = 6.15 × 10⁻⁷), with the largest responses being observed at the 2 middle levels (C1 and C2). While responses at C2 (tractography target) tended to be larger than at C1, this difference did not reach statistical significance (p > .05). These findings suggest that EP amplitude is sensitive to millimeter-scale differences in stimulation location within the target region. To rule out trial count as a confounding factor in this contact-level comparison, a control analysis using equal trial numbers per contact, without averaging over directional contacts, confirmed the same spatial profile of EP magnitude (see Supplemental S.3.2). No significant effect of stimulating contact was found on the latency of any of the peaks (Supplemental S.3.1).

Figure 3.

Evoked potential (EP) amplitude is sensitive to millimeter-scale differences within the target area. (A) Effect of stimulation level on the corrected area under the curve (AUC_corr). EP amplitude varies significantly across stimulating levels (Friedman test, p = 4.10 × 10⁻⁶), with post hoc analysis revealing significant differences between specific contacts (C0 vs. C1: p = .001, C0 vs. C2: p = 4.38 × 10⁻⁴, C1 vs. C3: p = .004, C2 vs. C3: p = 6.42 × 10⁻⁴). Stimulation of contact C2 generally produces the largest response. Panels (B) and (C) show the latency of the trough at around 60 ms poststimulus (N60) and the peak at around 75 ms poststimulus (P75), respectively. These are the most reliably detected peak and trough, with T60 observed in 77.14% (54/70) of the contact levels and P75 in 68.57% (48/70) of the contact levels. No significant effect of stimulation level on latency was observed (Friedman test, p > .05).

Relationship Between EP Magnitude and White Matter Connectivity

We used a linear mixed-effects model to assess whether the magnitude of stimulation-EP magnitude (AUC) was associated with white matter connectivity between the stimulated target region and 2 prefrontal cortical areas, the vmPFC-OFC and the vlPFC. A significant positive association was observed for vmPFC-OFC connectivity (β = 0.60, 95% CI [0.27–0.93], p = .00050; Bonferroni-corrected p = .0010), indicating that larger EP responses were linked to higher structural connectivity with this region. A significant association was also found for vlPFC connectivity (β = 2.79, 95% CI [1.46–4.11], p = 5.22 × 10⁻⁵; Bonferroni-corrected p = 1.04 × 10⁻⁴), with a larger effect size compared with vmPFC-OFC connectivity. These results suggest that EP magnitude may reflect the degree of structural engagement with multiple prefrontal targets, potentially more strongly with the vlPFC in this dataset. The modeled relationships, including predictions, are illustrated in Figure 4.

Figure 4.

Higher evoked potential (EP) amplitudes are correlated with higher connectivity to the ventromedial prefrontal cortex-orbitofrontal cortex (vmPFC-OFC) and ventrolateral prefrontal cortex (vlPFC). Top panels depict regression lines derived from the mixed-effects models for each hemisphere and individual patient data points (solid lines and filled circles represent the right hemisphere; dashed lines and open circles represent the left hemisphere) for EP amplitude across all contacts vs. connectivity to the (A) vmPFC-OFC and (B) vlPFC. Bottom panels show predicted connectivity based on EP amplitude vs. empirical connectivity values based on probabilistic tractography stimulation models for the (C) vmPFC and (D) vlPFC. The diagonal line represents equality between predicted and empirical values. In the mixed-effects models, EP amplitude was significantly associated with connectivity strength to both regions; for the vmPFC-OFC, the regression coefficient was β = 0.60 (95% CI 0.27–0.93), p = .00050 (Bonferroni-corrected p = .0010); for the vlPFC, β = 2.79 (95% CI 1.46–4.11), p = 5.22 × 10⁻⁵ (Bonferroni-corrected p = 1.04 × 10⁻⁴). AUC, area under the curve.

EPs in Responders Versus Nonresponders

Different EP patterns were observed across all DBS lead contacts in responders compared with nonresponders. Two of the 10 patients were clear nonresponders to treatment and also showed divergent EP morphology across stimulation contacts. Specifically, these nonresponders exhibited inconsistent EP patterns across contacts. In contrast, responders generally showed EPs that were more consistently present across lead contacts and exhibited a more uniform shape. To quantify this observed difference in EP uniformity, we calculated the within-lead average pairwise correlation coefficient as a measure of consistency. This analysis revealed a significant difference between responders (n = 8) and nonresponders (n = 2), with higher consistency values in the responder group (0.37 vs. 0.18, p = 3.53 × 10⁻²) (Figure 5A). The differences between an example responder (P09) and nonresponder (P03) are illustrated in Figure 5C and Figure 5D, respectively. While the small number of nonresponders limits firm conclusions, these preliminary findings suggest that nonuniform or inconsistent EP morphology may be associated with poor clinical response and warrant further investigation in larger cohorts.

Figure 5.

Comparison of within-lead correlation and evoked potentials (EPs) between treatment-responsive and nonresponsive patients. Panel (A) shows average within-lead correlation of EPs for patients with and without treatment response, defined as achieving a ≥30% reduction in Yale-Brown Obsessive Compulsive Scale scores within 12 months of deep brain stimulation. Responders showed significantly higher average within-lead correlation (p = 3.53 × 10⁻²). Panel (B) provides schematic representation of the lead contact configuration, with numbered contacts corresponding to those referenced in panels (C, D). Panels (C) and (D) show EPs recorded from all contacts on the leads of a treatment-responsive patient (P09, right hemisphere) and a nonresponsive patient (P03, left hemisphere), respectively. Each waveform represents a recording from an individual contact, highlighting differences in signal consistency and morphology between the 2 cases.

Discussion

Electrophysiological biomarkers for lead placement and clinical response have the potential to play a pivotal role in improving the clinical efficacy of ALIC DBS for OCD. We aimed to investigate whether intraoperative EP recordings could help confirm optimal lead placement and offer a preliminary indication of later treatment outcomes. The results show that ALIC DBS EPs, characterized by distinct oscillatory peaks and troughs, are reliably recorded during surgery using an easily applicable, limited EEG forehead montage. These fast and reliable recordings can be readily integrated into existing surgical procedures, making them a practical and valuable tool. EP amplitude was correlated with DBS-related white matter connectivity to regions implicated in OCD DBS treatment efficacy, such as the vmPFC-OFC and vlPFC, and responders showed more consistent and uniform EPs than nonresponders, together suggesting that EPs can be utilized intraoperatively to optimize DBS targeting and treatment response.

A critical priority of this study was to provide sufficient characterization of cortical ALIC DBS EPs to disambiguate the measure in future work. While ventral internal capsule/ventral striatum (VC/VS) DBS EPs have been recorded from PFC and OFC regions using temporarily implanted stereoencephalography electrodes during a 10-day intracranial monitoring period (33), our method may be more accessible. We consistently observed a robust EP waveform with a rhythmic structure characterized by alternating peaks and troughs at ∼25 Hz, suggesting a stereotyped oscillatory response to ALIC stimulation. This frequency range falls within the high-beta band (21–30 Hz), which has been increasingly implicated in OCD pathophysiology.

In OCD, a blunted increase in prefrontal beta power was reported during the post-trial period of a visuospatial working memory task, a phase associated with removing no-longer-relevant information from working memory (34,35). Moreover, the magnitude of beta increase was inversely correlated with symptom severity, suggesting that reduced high-beta reactivity may reflect impaired inhibitory control and contribute to persistent intrusive thoughts (34). In a separate study, increased beta-gamma phase-amplitude coupling was observed in frontocentral EEG sensors of patients with OCD compared with healthy control participants. This abnormal synchrony, likely originating from ventromedial prefrontal regions, was reduced by DBS of the nucleus accumbens, indicating that high-beta dynamics may reflect a therapeutic mechanism of action (36). Further supporting this, a case report showed that stimulation of the VC/VS has been shown to enhance high-beta activity in the PFC, suggesting that beta-band modulation is not only a feature of the disorder but also a potential marker of effective engagement of target circuits (37). This suggests that these EPs may offer a physiologically meaningful signal worth exploring further as a tool to guide and optimize DBS therapy.

A key question that remains is whether these repeating peaks and troughs reflect distinct processing stages or arise from a unified, rhythmic response. The consistent spacing between peaks and their apparent phase alignment suggest that they may reflect a recurrent process, similar in form to evoked resonant neural activity (ERNA) observed in the STN during STN DBS (38,39). In that context, ERNA has been shown to arise from reciprocal activity within the STN–globus pallidus externus (GPe) loop, producing a decaying oscillatory pattern at approximately 300 Hz (40). While the frequency and anatomical substrate differ, the structured, oscillatory response that we observed may reflect a similarly organized loop-based mechanism involving cortical and subcortical nodes within the ALIC-frontal circuitry. Based on the location of stimulation within the ALIC and the frontal recording sites, one plausible interpretation is that the observed EP waveform reflects engagement of the cortico-thalamo-striatal loop. This circuit is not only anatomically connected via the ALIC but also functionally implicated in the pathophysiology of OCD. The structured timing of the EP peaks may reflect sequential activation or recurrent interactions within this loop. However, this interpretation remains speculative and requires further confirmation, ideally through intracranial recordings from other nodes of the circuit or high-definition EEG-based cortical source analysis, determination of conduction delays based on known fiber properties, and computational modeling of network-level dynamics.

A significant effect of stimulation location on EP amplitude was observed, highlighting variability of the cortical response to stimulation across different levels on the DBS lead. Notably, stimulation at the middle 2 contacts (C1 and C2) elicited the largest responses, with the third contact from ventral to dorsal (C2) generating the largest EP amplitude. This finding is particularly important because surgical planning aims to place C2 at the tractographic sweet spot associated with treatment response in our previous patients (12). Furthermore, 9 of 10 patients were programmed on one of these middle contacts at the 6-month follow-up, with 5 of 10 on contact 2. Importantly, an effective target engagement biomarker must demonstrate sensitivity to millimeter-scale differences within the target area. The observed variability in EP responses across DBS lead contacts highlights their possible utility in refining lead placement and optimizing alignment with one of the key therapeutic white matter pathways.

The magnitude of the DBS EPs across the forehead was used to predict white matter connectivity from the stimulated ALIC target region to treatment-relevant areas of the PFC. This finding supports the notion that noninvasive electrophysiological recordings can serve as a readout of white matter perturbation within the target region. This is consistent with previous findings in depression, where EPs in response to stimulation have been shown to reflect engagement of structurally connected areas (15,16). Connectivity to both the vmPFC-OFC and vlPFC was significantly predicted by EP magnitude, indicating that the stimulation-evoked response reflects engagement of multiple prefrontal pathways. The regression coefficient was larger for the vlPFC, suggesting a stronger association in this region. However, both models showed robust fits, supporting the utility of EPs as a marker of structural connectivity strength.

However, the underlying neurophysiological basis of this association between EPs and white matter connectivity remains to be clarified. It is not yet known whether the EPs reflect direct activation of afferent pathways, local network synchronization, or other mechanisms. Furthermore, the spatial sensitivity of the EP signal may still be influenced by recording electrode placement. For example, the midline positioning of intraoperative EEG electrodes may favor detection of signals from ventromedial regions, while responses from more lateral areas such as the vlPFC could be less readily captured. Extending coverage to more lateral scalp regions remains challenging intraoperatively due to constraints imposed by the stereotactic frame and sterile field. Future work in less constrained postoperative environments could help clarify the topographical specificity of these evoked responses.

Distinct patterns of EPs were observed in responders and nonresponders, defined as patients whose symptom scores decreased by 30% from baseline in severity and those who failed to cross this threshold, respectively. In patients meeting the response threshold, EPs across different contacts within the lead were more coherent, as confirmed by a significantly higher average within-lead correlation. The characteristic 3 peak/trough EP pattern was absent in most contacts of patients whose scores did not meet the threshold and, if present, was typically limited to the upper contacts. While the small number of nonresponders limits our ability to draw firm conclusions, this exploratory analysis suggests that more consistent and uniform EPs across contact levels may be worth further investigation as a potential biomarker of optimal lead placement and alignment with key white matter tracts in larger datasets.

Limitations and Future Directions

Several important considerations should guide interpretation of these findings. The observed differences in waveform consistency between responders and nonresponders must be viewed with caution. While the high responder rate likely reflects the use of a predefined DTI-based targeting strategy, the low number of nonresponders constrains our ability to draw firm conclusions about the predictive value of EP characteristics for clinical outcomes. While a clear relationship between specific white matter targets and clinical outcomes has been established (13,14,41, 42, 43), a target engagement biomarker does not necessarily equate to a response biomarker. Thus, this work identifies physiological features relevant to target engagement, and the link to treatment response illustrates potential clinical relevance. Larger, more balanced cohorts and prospective designs with preregistered analysis plans will be critical for validating whether these physiological signatures have prognostic utility.

It is not yet possible to conclude that EPs truly reflect dynamic engagement of the connected pathway or whether the observed correlation with tractographic “sweet spots” is coincidental. A factor confounding the current results stems from the use of 2 different stimulation amplitudes, 3.5 mA for directional contacts and 5.5 mA for ring contacts. This limits the interpretability of EP differences despite linear correction because the response of neural tissue to increasing electrical stimulation is likely sigmoidal, with threshold and saturation effects (44,45). Moreover, the EEG montage used in this study limited our ability to examine targets in the OCD circuit beyond the frontal lobe or to assess the laterality of the evoked responses. Prior research suggests that the response is predominantly ipsilateral (33), a finding that was supported in our cohort; in cases where a mastoid reference was available, the signal appeared strongest in the same hemisphere as the stimulation before re-referencing. Future studies would benefit from the use of a consistent reference and additional validation strategies, such as stimulation paradigms with varied frequency and amplitude to probe response specificity, biophysical modeling of current spread and fiber activation and concurrent functional MRI-EEG multimodal integration, to establish a stronger causal link between stimulation, circuit engagement, and cortical response.