Deep brain stimulation induces white matter remodeling and functional changes to brain-wide networks

Abstract

Deep brain stimulation (DBS) is an emerging therapy for treatment-resistant neurological and psychiatric disorders. Despite this, little is known about the anatomical and functional mechanisms that underlie this therapy. We targeted DBS to white matter adjacent to the subcallosal anterior cingulate cortex (SCC-DBS) in macaques, modeling the approach proven effective for depression in humans. SCC-DBS caused a selective increase in fractional anisotropy, linked to white matter microstructure, in the cingulum bundle. At the cellular level, this was associated with an increase in both myelinated oligodendrocytes and the degree of myelination in the mid-cingulum bundle. SCC-DBS also altered brain-wide functional connectivity, changing interactions between the SCC and multiple brain networks, most notably the default mode network that has been implicated in depression. Overall, our data indicate that white matter remodeling as well as selective changes in multiple brain networks may contribute to DBS’s therapeutic efficacy.

In a nonhuman primate model, Fujimoto et al. show that deep brain stimulation promotes white matter remodeling and reorganizes brain-wide functional networks, detailing a mechanism through which this neuromodulation therapy may treat depression.

Main

Despite advances in available treatments, many people with psychiatric and neurological disorders do not respond to pharmacological, behavioral or physical therapies¹. In these cases, deep brain stimulation (DBS) is increasingly being used to reduce symptoms. First developed for Parkinson’s disease², this neuromodulation therapy is now being adapted for other neurodegenerative and psychiatric disorders, particularly depression³. Importantly, unlike the approach of targeting the gray matter in the subthalamic nucleus or parts of the motor thalamus, to alleviate movement control deficits^2,3, DBS for depression instead targets white matter pathways in different parts of the brain^4–6. In particular, DBS delivered to the white matter adjacent to subcallosal anterior cingulate cortex (SCC-DBS)⁴ targets the confluence of three white matter tracts: the cingulum bundle (CB), the forceps minor (FM) and the uncinate fasciculus (UF)⁷. DBS to this location has led to sustained long-term symptom reduction in approximately 60–75% of treatment-resistant patients⁸. Despite the clinical efficacy of SCC-DBS for depression⁹, the neurobiological mechanisms engaged by this treatment are still being fully established. Revealing the mechanisms of DBS has the potential to provide improved and new therapeutic targets as well as insight into the underlying pathophysiology of depression. In this study, we investigated the mechanisms by which SCC-DBS modulates brain structure in addition to function, using healthy nonhuman primates to isolate the direct effects of stimulation without the confounding influence of preexisting pathology.

It has been theorized that SCC-DBS directed to white matter has its beneficial effects by rapidly normalizing activity and functional communication across distributed brain circuits¹⁰. Although there is evidence to support this functional account of the therapeutic effects^11,12, the available data are still limited due to the challenges with imaging patients with implanted DBS devices. Often overlooked is the possibility that in addition to its rapid functional effects on neural activity, DBS may concurrently promote changes to white matter microstructure. This idea is supported by the findings that DBS-mediated improvement in mood often takes weeks to fully appear^4,7; and there is a relationship between white matter integrity before starting treatment and the extent of improvement in depression ratings following SCC-DBS¹³. If DBS causes anatomical remodeling of white matter tracts, it would reveal a previously underappreciated mechanism of therapeutic action and a role for white matter dysfunction in psychiatric disease in general, particularly in depression.

Results

SCC-DBS induces selective macroscale remodeling in the CB

Matching the location in the brain that has been used for treatment-resistant depression, we targeted mini-DBS electrodes to the confluence of the CB, the UF and the FM in two macaque monkeys (Fig. 1a, top). DBS leads were implanted unilaterally, allowing for the other hemisphere to act as an internal control to investigate the direct effects of white matter stimulation compared to indirect cross-hemisphere effects. Post-surgery computed tomography (CT) confirmed the DBS lead locations relative to the confluence of the three white matter tracts, and this was used to determine the active contact for stimulation (Fig. 1a, bottom, and Fig. 1b). Following a four-week period after implantation during which no stimulation was delivered, chronic stimulation (130 Hz, 5 mA, 90 μsec) was delivered for six weeks (Fig. 1c). This stimulation regime was used as it mirrors the approach taken in human patients^4,7. Further, it was used because on average, the rate of change in depression rating symptoms asymptotes after 6 weeks of stimulation^4,6,7. Both before lead implantation and after 6 weeks of stimulation, we assessed the animals’ home cage behavior to see if there were any changes in movement or foraging behaviors (Fig. 1c). Time spent moving and foraging both increased after SCC-DBS, with concordant effects observed across subjects (one-way analysis of variance (ANOVA); movement: F_1,236 = 34.61, P < 0.001; foraging: F_1,236 = 17.35, P < 0.001; Extended Data Fig. 1, monkeys N and T). No neurological deficits were observed after SCC-DBS. In addition, we did not observe any rotational tendencies or other changes in behavior that were indicative of asymmetries in motor function. Hand preference remained consistent in both monkeys before and after DBS (pre-DBS: right hand; post-DBS: right hand). Thus, despite both animals being healthy and unilateral stimulation, SCC-DBS induced a slight behavioral change, confirming that it was having an effect on the brain.

Fig. 1. A nonhuman primate model of SCC-DBS.

a, Identification of white matter tracts and confirmation of SCC-DBS lead location. Top: visualization of the confluence of the CB, the FM and the UF. The three white matter tracts were reconstructed using probabilistic tractography based on DWI and overlaid on a T1w pre-DBS structural image (monkey T). Pink circle denotes the confluence of the three white matter tracts: CB (yellow), UF (blue), FM (red). Bottom: post-DBS surgery CT confirmation of DBS lead location. The post-DBS implantation CT image is overlaid on a pre-DBS implantation T1w image (monkey T). b, The scheme of SCC-DBS implantation. A miniaturized custom-made DBS lead was implanted into the confluence through a small craniotomy and connected to a human DBS system. Bottom: CT scout view of SCC-DBS implanted macaque head. c, Experimental timeline. Baseline T1, DWIs and rs-fMRI were collected for each subject. Using the baseline DWIs, the SCC-DBS target—that is, the confluence of the CB, the UF and the FM—was visualized. Using a stereotactic approach, a SCC-DBS lead was implanted to target the confluence in one hemisphere. After a 4-week recovery period, chronic stimulation for 6 weeks was delivered. Just after removal of SCC-DBS, T1, diffusion and functional MRIs were collected. On the day after MRI, the animal’s brain was extracted and prepared for histological assessment. Stim., stimulation.

Extended Data Fig. 1. Naturalistic behavioral changes after SCC-DBS stimulation.

The probability of movement (left) or foraging (right) observed in a 10-second bin among 60 bins (10 minutes) during either the pre or post SCC-DBS sessions. Symbols indicate probability of observed behavior for each animal (red circle, monkey N, red triangle, monkey T, black square, monkey D (non-stimulation control animal). Error bars show SEM. Each data point represent time-bin–level observations (10-second bins) within each session. Data from monkeys N and T were analyzed using 1-way ANOVA. Number of time-bin–level measurements (n): Movement, n = 120 (pre), 120 (post); Foraging, n = 120 (pre), 120 (post). Statistical comparisons assessed differences between pre- and post-DBS conditions. All tests were two-sided. Tukey–Kramer post-hoc methods to corrected for multiple comparisons. The main effect of condition was significant (Movement, F_1,236 = 34.61, P = 1.37e-08, ηp² = 0.13; Foraging, F_1,236 = 17.35, P = 4.37e-05, ηp² = 0.068). *** P < 0.001.

Source data

Fractional anisotropy (FA), derived from diffusion-weighted images (DWIs), is a measure of white matter integrity and has been shown to correlate with changes in white matter structure in humans and animals^14,15. We therefore obtained whole-brain DWI before SCC-DBS lead implantation and after 6 weeks of stimulation (Fig. 1c). We extracted FA values from the probabilistic reconstructions of the CB, the UF and the FM to quantify in vivo changes in white matter integrity following SCC-DBS. We then directly compared FA values within the tracts by subtracting the presurgery FA from post-stimulation FA values (Fig. 2a).

Fig. 2. SCC-DBS induces selective macroscale remodeling in CB.

a, top: FA in the three white matter tracts before and after DBS. Extracted FA values are shown from the CB (left), UF (middle) and FM (right). Top row: white matter tract masks. Second row: pre-DBS FA value. Third row: post-DBS FA value. Fourth row: subtraction image of post-DBS FA and pre-DBS FA values. The color indicates FA value. The sagittal images show the stimulated hemisphere. Bottom: the bar graph shows mean FA values (±s.e.m.) before (pre-DBS) and after (post-DBS) stimulation for each white matter tract in each hemisphere. Colored bars indicate the stimulated hemisphere (CB, yellow; UF, blue; FM, red), and gray bars indicate the control hemisphere. Lighter colors represent pre-DBS, and darker colors represent post-DBS. Each FA value represents voxel-wise measurements within each tract mask, pooled across two monkeys. Number of voxel-wise measurements (n): stimulation side: CB, n = 380 (pre), 380 (post); UF, n = 252 (pre), 252 (post); FM, n = 105 (pre), 105 (post); control side: CB, n = 363 (pre), 363 (post); UF, n = 367 (pre), 367 (post); FM, n = 133 (pre), 133 (post). FA values were compared using three-way ANOVA. Statistical comparisons assessed differences between pre- and post-DBS conditions within each hemisphere and tract. All tests were two-sided. Tukey–Kramer post hoc tests corrected for multiple comparisons. There was a significant interaction between PrePost and side (F_1,3187 = 4.64, P = 0.0312, η_p² = 0.0015). A significant PrePost difference was observed in the stimulated CB (P = 0.022). *P < 0.05. b, FA changes after SCC-DBS in subregions of the CB. Mean (±s.e.m.) FA value in four subregions (SCC, dACC, MCC and PCC portion) before (pre-DBS) and after (post-DBS) stimulation in the stimulated (top graph) and control hemisphere (bottom graph). Gray color shows pre-DBS, and red shows post-DBS. Symbols represent individual subjects (circle/triangle, monkey N/T, respectively). Each FA value represents voxel-wise measurements in each CB subregion mask, pooled across two monkeys. Number of voxel-wise measurements (n): stimulation side: SCC, n = 54 (pre), 54 (post); dACC, n = 137 (pre), 137 (post); MCC, n = 55 (pre), 55 (post); PCC, n = 80 (pre), 80 (post); control side: SCC, n = 81 (pre), 81 (post); dACC, n = 158 (pre), 158 (post); MCC, n = 50 (pre), 50 (post); PCC, n = 72 (pre), 72 (post). FA values were compared using three-way ANOVA. Statistical comparisons assessed differences between pre- and post-DBS conditions within each hemisphere (stimulated versus control) and CB subregion. All tests were two-sided. Tukey–Kramer post hoc methods corrected for multiple comparisons. There was a significant interaction between PrePost and side (F_1,1357 = 5.42, P ₌ 0.020, η_p² = 0.0040). A significant PrePost difference was observed in the stimulated MCC-CB (P = 0.0017). **P < 0.01.

Source data

We found there was a selective increase in FA in the CB in the stimulated hemisphere, particularly the mid-cingulate portion, after 6 weeks of SCC-DBS (three-way ANOVA, F_1,3187 = 4.64, P = 0.0312; post hoc tests on tracts, Tukey–Kramer test, P = 0.022; Fig. 2a, left column, and Extended Data Fig. 2, left column). Importantly, both monkeys exhibited changes in the same direction (Extended Data Fig. 2). No consistent changes were apparent in either the stimulated UF or the FM (Fig. 2a, middle and right columns, and Extended Data Fig. 2, middle and right columns) or in the same tracts within the control hemisphere (Fig. 2a, bottom graphs, and Extended Data Figs. 3 and 4). Based on these findings, we next aimed to determine whether specific subregions of the CB showed FA changes after SCC-DBS. Comparing the FA values before and after 6 weeks of DBS in four CB subregions revealed that only the mid-cingulate region on the stimulated side showed an increase in FA in both animals (three-way ANOVA, F_1,1357 = 5.42, P = 0.020; post hoc analyses, Tukey–Kramer test, P = 0.0017; Fig. 2b and Extended Data Fig. 4a).

Extended Data Fig. 2. Fractional anisotropy changes in individual subjects.

Fractional anisotropy (FA) value in the cingulum bundle (CB, left), uncinate fasciculus (UF, middle), and forceps minor (FM, right) in each subject. The extracted FA value in each white matter tract of interest is shown in NMT space. The top row is pre-DBS FA value, the middle row is post-DBS FA value and the third row is the subtraction image of FA value (post-DBS FA value – pre-DBS FA value) in each panel. The color indicates FA value. The sagittal images show the stimulated hemisphere.

Extended Data Fig. 3. Fractional anisotropy in unstimulated control hemisphere.

Fractional anisotropy (FA) value in the cingulum bundle (CB, A, top) and the uncinate fasciculus (UF, B, bottom) in the control hemisphere. The extracted FA value in each white matter tract of interest is shown in NMT space. The top row is pre-DBS FA value, the middle row is post-DBS FA value, and the third row is the subtraction image of FA value (post-DBS FA value – pre-DBS FA value) in each panel. The color indicates FA value.

Extended Data Fig. 4. Fine-grained analysis of FA in CB, UF and FM.

FA value in the cingulum bundle (CB), the uncinate fasciculus (UF), and the forceps minor (FM) extracted from each plane within the tract masks for monkey N (top) and monkey T (bottom). Each FA value represents voxel-wise measurements, collected from two monkeys. Number of voxel-wise measurements (n) are shown in Supplementary Table 5 (CB), Supplementary Table 6 (UF) and Supplementary Table 7 (FM). Blue circle/line; pre-DBS mean FA in the stimulated hemisphere, red circle/line; post-DBS mean FA in the stimulated hemisphere, cyan circle/line; pre-DBS mean FA in the control hemisphere, magenta circle/line; post-DBS mean FA in the control hemisphere. The error bars are SEM. A) The mean FA values per slice in CB. The left image is CB mask (yellow) displayed in NMT space. The CB mask was divided into two parts based on its C shape; one part consisted of the SCC, and the other contained the remaining CB regions. For the SCC part, defined as area 25 and area 32 based on the macaque D99 atlas, the mean FA value in each axial slice (the average of FA values extracted from each voxel with the same z coordinates in the individual pre-DBS T1 space) was calculated and plotted. For the remaining part, which includes the dACC, MCC and PCC portion of CB, the mean FA value in each coronal slice (the average of FA values extracted from each voxel which has the same y coordinates) was calculated and plotted. B) The mean FA values per slice in the UF. The left image is UF mask (blue) displayed in NMT space. The UF mask was divided into two parts, the temporal part and the frontal-insula part because of its C shape. For the temporal part, defined as the part of the UF running through the temporal lobe based on macaque D99 atlas, the mean FA value in each axial slice (the average of FA values extracted from each voxel which has the same z coordinates in the individual pre-DBS T1 space) was calculated and plotted in the graphs. For the frontal-insula part, the mean FA in each coronal slice (the average of FA values extracted from each voxel which has the same y coordinates) was calculated and plotted. C) The mean FA values per slice in the FM. The left image is the FM mask (red) displayed in NMT space. The FM mask was divided into left and right hemispheres. For each hemisphere, the mean FA in each coronal slice (the average of FA values extracted from each voxel which has the same y coordinates) was calculated and plotted.

Source data

In summary, in vivo neuroimaging indicates that SCC-DBS modified the white matter integrity of the CB after 6 weeks of stimulation. This anatomical change was specific to the mid-cingulate portion of the CB (mid-cingulum bundle (MCC-CB)), which is spatially remote from where stimulation was delivered. Thus, SCC-DBS is associated with a selective macrolevel remodeling of white matter distant from the stimulation site in a tract that connects the SCC with dorsal and posterior parts of the cingulate cortex¹⁶.

SCC-DBS induces increased myelinating oligodendrocytes and the degree of myelination

Prior work has shown that FA is associated with underlying white matter microstructure^14,15; however, such macrolevel estimates are unable to reveal what specifically underlies this measure¹⁷. Thus, we next sought to establish whether, and how, the SCC-DBS–induced FA change was related to cellular-level changes in white matter in the mid-cingulate cortex. First, we investigated how stimulation influenced oligodendrocyte proliferation, especially myelinating oligodendrocytes in the mid-cingulate portion of CB where FA increases after SCC-DBS were found (Fig. 3a), by conducting immunolabeling against CC1, an antibody that preferentially labels myelinating oligodendrocytes. The proportion of CC1 positive cells in the whole DAPI stained population was higher in the mid-cingulate portion of the CB on the stimulated side compared to the control side in both animals (CC1/DAPI ratio: one-way ANOVA, F_1,108 = 9, P = 0.0034; Fig. 3b,c and Extended Data Fig. 5). No changes were observed in the anterior portion of CB, consistent with the lack of FA alterations there (Extended Data Fig. 6a). Similarly, the region where the FM and UF overlap showed no oligodendrocyte changes (Extended Data Fig. 6b). Thus, SCC-DBS was associated with selective increase in the proliferation of oligodendrocytes specifically in the MCC-CB at the same location where we found a change in FA and distant to where stimulation was delivered.

Fig. 3. SCC-DBS induces increased myelinating oligodendrocytes and the degree of myelination.

a, Coronal slice for histology: sagittal and coronal MRI images showing where changes in cellular level markers of white matter structures were assessed. b, Oligodendrocyte change. Double-stained images with CC1 (marker for myelinated oligodendrocytes, red) and DAPI (marker for nuclei, blue) from monkeys N and T (top and bottom rows, respectively) in the MCC-CB for control (left) and stimulated hemispheres (right) (×20 magnification). Scale bars: 100 μm. c, Box plot of the ratio of CC1- and DAPI-positive cells (%). Symbols represent each subject (circle/triangle, monkey N/T, respectively). The box represents the interquartile range (IQR; 25th–75th percentile). Green line indicates the median. Whiskers extend to 1.5 × IQR. Outliers are not displayed. Each data point represents measurements from individual fields (seven fields per hemisphere per section) across four coronal sections per subject, pooled across two monkeys. The number of field-wise measurements (n) used for statistical analysis was as follows: stimulated hemisphere, n = 56 (7 fields × 4 sections × 2 subjects); control hemisphere, n = 56. CC1/DAPI ratios were compared using one-way ANOVA. Statistical comparisons assessed differences between stimulation side and control side. All tests were two-sided. The main effect of side was significant (F_1,108 = 9, P = 0.0034, η_p² = 0.077). **P < 0.01. d, Electron microscopy images of myelinated axons in control (left) and stimulated hemispheres (right) from the mid-cingulate portion of CB in monkey T. Scale bars: 800 nm. e, Scatter plot of g-ratio measurements from control (cyan) and stimulated hemispheres (magenta) for each animal (top, monkey N; bottom, monkey T). The bottom-right image shows the g-ratio measurement in a representative myelinated axon. f, Box plot of g-ratio measurements. Symbols represent individual subjects (circle/triangle, monkey N/T, respectively). The box represents the IQR (25th–75th percentile). Green line is the median. Whiskers extend to 1.5 × IQR. Outliers are not displayed. Each data point represents measurements from individual fields (stimulated hemisphere, n = 350; control hemisphere, n = 350), pooled across two monkeys. g-ratio ratios were compared using one-way ANOVA. Statistical comparisons assessed differences between stimulation side and control side. All tests were two-sided. The main effect of side was significant (F_1,696 = 58.98, P = 5.42 × 10⁻¹⁴, η_p² = 0.078). ***P < 0.001.

Source data

Extended Data Fig. 5. The number of myelinated oligodendrocytes increased in the mid-cingulate portion of the CB.

Left box plot shows the number of CC1 positive cells in the mid-cingulate portion of the CB and the right graph is the box plot demonstrating the number of DAPI positive cells. The box represents the interquartile range (IQR; 25th–75th percentile). Gray line indicates the median. Whiskers extend to 1.5 × IQR. Outliers are not displayed. Symbols represent average values for individual animals (circle, monkey N, triangle, monkey T). Each data point represents measurements from individual fields (7 fields per hemisphere per section) across four coronal sections per subject, pooled across two monkeys. The number of field-wise measurements (n) used for statistical analysis was as follows: stimulated hemisphere, n = 56 (7 fields × 4 sections × 2 subjects); control hemisphere, n = 56. The number of CC1 positive cells and the number of DAPI positive cells were compared using 1-way ANOVA. Statistical comparisons assessed differences between stimulation side and control side. All tests were two-sided. The main effect of side for the number of CC1 positive cells was significant (F_1,108 = 6.22, P = 0.0142, ηp² = 0.054), while no significant difference was observed for the number of DAPI positive cells (F_1,108 = 0.55, P = 0.4617, ηp² = 0.0050). * P < 0.05.

Source data

Extended Data Fig. 6. The number of myelinated oligodendrocytes in the dACC portion of the CB and in the overlapping region of the UF and FM.

A) Left top MRI images illustrate the anatomical location of the histologically examined region in the dACC portion of the CB (Yellow; CB mask, Pink; dACC ROI, Green line on the sagittal image; the plane of the coronal slice assessed, Green square on the coronal image; the area selected for histological analysis). Left bottom images are representative immunofluorescence images (Red; CC1 positive cells, Blue; DAP positive cells; Yellow scale bar; 100 μm). Left box plot displays the ratio of CC1-positive to DAPI-positive cells in the dACC portion of the CB. The top right graph shows the number of CC1 positive cells and the bottom right graph is a box plot demonstrating the number of DAPI positive cells in the dACC portion of the CB. Symbols indicate average values for individual animals (circles: monkey N; triangles: monkey T). Box represents the interquartile range (IQR; 25^th–75^th percentile). Green line is the median. Whiskers extend to 1.5 × IQR. Outliers are not displayed. For three box plots, each data point represents measurements from individual fields (7 fields per hemisphere per section) across four coronal sections, collected from two monkeys. The number of field-wise measurements (n) used for statistical analysis was as follows: stimulated hemisphere, n = 56 (7 fields × 4 sections × 2 subject); control hemisphere, n = 56. CC1/DAPI ratio, number of CC1 positive cells and number of DAPI positive cells were compared between hemispheres using 1-way ANOVA. Statistical comparisons assessed differences between stimulation side and control side. All tests were two-sided. The main effect of side were not significant (CC1/DAPI ratio, F_1,108 = 2.93, P = 0.0899, ηp² = 0.026, #CC1, F_1,108 = 0, P = 0.9974, ηp² = 9.73e-08, #DAPI, F_1,108 = 1.22, P = 0.2718, ηp² = 0.011). B) Left top MRI images illustrate the anatomical location of the histologically examined region in the overlapped region of the uncinate fasciculus (UF) and the forceps minor (FM) (Blue; UF mask, Red; FM mask, Green square on the coronal image; the area selected for histological analysis). Left bottom images are representative immunofluorescence images (Red; CC1 positive cells, Blue; DAPI positive cells, Yellow scale bar; 100 μm). Left box plot displays the ratio of CC1-positive to DAPI-positive cells in the overlapped region of UF and FM. The top right graph shows the number of CC1 positive cells and the bottom right graph is a box plot demonstrating the number of DAPI positive cells in the overlapped region of UF and FM. Symbols indicate average values for individual animals (circles: monkey N; triangles: monkey T). Box represents the interquartile range (IQR; 25^th–75^th percentile). Green line is the median. Whiskers extend to 1.5 × IQR. Outliers are not displayed. For three box plots, each data point represents measurements from individual fields (5 fields per hemisphere per section) across four coronal sections, collected from two monkeys. The number of field-wise measurements (n) used for statistical analysis was as follows: stimulated hemisphere, n = 40 (5 fields × 4 sections × 2 subject); control hemisphere, n = 40. CC1/DAPI ratio, number of CC1 positive cells and number of DAPI positive cells were compared between hemispheres using 1-way ANOVA. Statistical comparisons assessed differences between stimulation side and control side. All tests were two-sided. The main effect of side were not significant (CC1/DAPI ratio, F_1,76 = 1.45, P = 0.2319, ηp² = 0.019, #CC1, F_1,76 = 0.79, P = 0.3778, ηp² = 0.010, #DAPI, F_1,76 = 0, P = 0.9611, ηp² = 4.97e-05).

Source data

Next, we sought to establish whether SCC-DBS also changed the morphology of myelination within this region. Alterations in the thickness of the myelin sheath around an axon impact the conduction of signals along that axon, as well as neural synchrony^18,19. If stimulation increases FA in the MCC-CB, this should be associated with increased myelin thickness¹⁵. Using electron microscopy, we measured the ratio between the inner and outer diameter of a random selection of myelinated axons within the mid-cingulate (Fig. 3d). This ratio, known as the g-ratio (Fig. 3e, bottom-right), allows for the degree of myelination within a white matter tract to be quantified. In the stimulated hemisphere, the MCC-CB g-ratio was decreased compared to the control hemisphere in both animals (one-way ANOVA, F_1,696 = 58.98, P < 0.001; Fig. 3d–f), meaning that the degree of myelination was increased compared to the unstimulated control hemisphere.

Taken together, these findings provide histological confirmation that SCC-DBS enhances myelin remodeling in the mid-cingulate portion of the CB and provides a direct biological basis for the increased FA seen after 6 weeks of SCC-DBS.

Whole-brain functional connectivity changes following SCC-DBS

Our macro- and microlevel structural analyses of white matter indicate that the CB was remodeled by SCC-DBS. The degree of myelination within a white matter tract is known to affect the conduction velocity of signals in that tract as well as the level of neural synchrony between areas that connect through that tract^18,19. Therefore, we next examined the functional alterations induced by SCC-DBS.

We examined how SCC-DBS alters resting-state functional magnetic resonance imaging (rs-fMRI) signals across the brain by generating a whole-brain connectivity matrix. In contrast to the localized structural alterations observed in white matter, SCC-DBS induced widespread functional changes (Fig. 4). Broadly, we observed a reduction in cortico–cortical functional connectivity and a concurrent increase in subcortico–subcortical functional connectivity. This is evident when the pre- is subtracted from the post-SCC-DBS functional connectivity (Fig. 4a, top-right column, and Fig. 4b, top). When the functional connectivity was sorted based on cortical networks membership, it was clear that connectivity within and across networks was reduced (Fig. 4c, left). To determine whether these changes were specifically associated with SCC-DBS, we also analyzed rs-fMRI data from a nonsurgical control group (three male rhesus macaques that had not received DBS surgery). In these control animals, a comparison conducted between scans acquired 6–7 weeks apart did not reveal a systematic pattern of functional changes, suggesting that the observed alterations in the experimental animals were indeed specific to SCC-DBS (Fig. 4a,b, bottom, and Fig. 4c, right). These data suggest that although SCC-DBS induces a local structural change, it has a much broader effect on functional connections.

Fig. 4. Whole-brain functional connectivity changes in SCC-DBS and nonsurgical control groups.

a, Brain-wide functional connectivity in cortical and subcortical areas. Confusion matrices represent the functional connectome in pre scans (left column), post scans (middle column) and the difference between post and pre scans (right column), respectively. Inset color bar represents Z-transformed correlation coefficients. Note that the post-DBS scans in the DBS group (top panels) were acquired 6 weeks after stimulation started, whereas the nonsurgical control group (bottom panels) underwent two functional scans 6–7 weeks apart without DBS. Labels on the side of the right column represent ROIs as listed in the bottom-right panel (CHARM and SARM atlas, level 4). b, Box plot of functional connectivity change (Z-score, post − pre) for each ROI of the CHARM and SARM atlas for SCC-DBS group (top) and nonsurgical control group (bottom), respectively. The box represents the IQR (25th–75th percentile). Central line is the median. Whiskers extend to 1.5 × IQR. Outliers are not displayed. Each data point represents measurements from individual ROI (pre, n = 122; post, n = 122), pooled across two monkeys. Rank-sum test was conducted to determine statistical significance (P < 0.01 with Bonferroni correction). Statistical comparisons assessed differences between pre and post conditions. All tests were two-sided. P values are provided in Supplementary Table 1 (DBS group) and Supplementary Table 2 (nonsurgical control group). Effect sizes (rank–biserial correlation, r_rb) are shown Supplementary Table 3 (DBS group) and Supplementary Table 4 (nonsurgical control group). Colors represent significant negative (blue) and positive changes (orange) in functional connectivity from zero in each ROI. The dotted lines divide cortical (left) and subcortical (right) ROIs. c, Circular plots showing functional connectivity changes for SCC-DBS (left) and control (right) groups. The color and width of lines indicate the direction and strength of changes (blue, Z < −0.1; orange; Z > 0.1). The dotted lines divide cortical and subcortical ROIs. RSC, retrosplenial cortex; STGr, rostral superior temporal gyrus; STGc, caudal superior temporal gyrus; ParaHipp, parahippocampal cortex; ERh, entorhinal cortex; PRh, perirhinal cortex; rostral DLP, rostral dorsolateral prefrontal cortex; rostral VLP, rostral ventrolateral prefrontal cortex; Ins/Pi, insula and parainsula; Ri, retroinsula; SⅠ, primary somatosensory cortex; SⅡ, secondary somatosensory cortex; SMA/preSMA, medial supplementary motor areas; M1, primary motor cortex; PM, premotor cortex.

Source data

To look more closely at the direct effects of SCC stimulation, we conducted a seed-based ROI analysis comparing functional connectivity between the stimulated SCC (area 25) and the rest of the brain. Before stimulation, the SCC showed positive connectivity to the ventromedial frontal cortex (vmFC), dorsal anterior cingulate cortex (dACC), posterior cingulate cortex (PCC), superior temporal gyrus (STG), hippocampus (HIP), amygdala (AMY) and dorsolateral prefrontal cortex (DLPFC), as well as negative connectivity to the insula and somatosensory cortex (Fig. 5a, top row, Extended Data Fig. 7, top row).

Fig. 5. Whole-brain functional connectivity changes following SCC-DBS.

a, The mean whole-brain functional connectivity with SCC shown on sagittal (left) and axial (right) images before (top row) and after (second row) 6 weeks of SCC-DBS stimulation, as well as the difference between post- and pre-SCC-DBS stimulation (third row). Sagittal images of functional connectivity from the stimulated hemisphere are shown. The bottom row shows functional connectivity changes with the right SCC in the nonsurgical control group over a 6-week period without any intervention (right hemisphere is shown). Images are thresholded for functional connectivity with P < 0.01 and cluster size >20 (cluster-level corrected P < 0.01, two-sided). Regions outlined in black indicate statistically significant clusters. The green circle represents the location of the active contact of SCC-DBS lead. b, Functional connectivity with stimulated SCC changes induced by SCC-DBS in the DMN and SAN hubs. Bars represent the mean value of functional connectivity normalized Z-score between SCC and each brain network hub ROI. Each data point represents voxel-wise measurements within each target ROI, pooled across two monkeys. Colored bars are the mean value of functional connectivity between stimulated SCC and target ROI in the stimulated hemisphere (red for the DMN hubs and yellow for the salience network (SAN) hubs). The gray bars show the mean functional connectivity with stimulated SCC in the control hemisphere. Lighter color or lighter gray bars indicate pre-DBS functional connectivity Z-score with stimulated SCC, and the darker color or darker gray bars represent post-DBS scores. The gray circle represents monkey N, and the gray triangle represents monkey T. Error bars show s.e.m. Number of voxel-wise measurements (n): stimulation side: vmFC, n = 109 (pre), 109 (post); PCC, n = 249 (pre), 249 (post); anterior-dorsal insula, n = 100 (pre), 100 (post); dACC, n = 162 (pre), 162 (post); control side: vmFC, n = 109 (pre), 109 (post); PCC, n = 249 (pre), 249 (post); anterior-dorsal insula, n = 100 (pre), 100 (post); dACC, n = 162 (pre), 162 (post). Functional connectivity values were compared using three-way ANOVA. Statistical comparisons assessed differences between pre- and post-DBS conditions within each hemisphere (stimulated versus control) and target ROI. All tests were two-sided. Tukey–Kramer post hoc methods corrected for multiple comparisons. There was a significant interaction between PrePost and side (F_1,20279 = 186.48, P = 2.87 × 10⁻⁴², η_p² = 0.0091). A significant PrePost difference was observed in the vmFC (stimulation side, P < 1.00 × 10⁻³⁰⁰; control side, P < 1.00 × 10⁻³⁰⁰), PCC (stimulation side, P = 8.22 × 10⁻¹²; control side, P = 4.03 × 10⁻⁷), anterior-dorsal insula (stimulation side, P = 4.42 × 10⁻⁸; control side, P = 0.0293) and dACC (stimulation side, P = 4.75 × 10⁻⁴; control side, P < 1.00 × 10⁻³⁰⁰). *P < 0.05, **P < 0.01, ***P < 0.001. c, Intrinsic brain network-level changes in functional connectivity after 6 weeks of SCC-DBS. Bars represent the mean value of functional connectivity between the stimulated SCC (cyan region) and six brain network hubs (hubs of the DMN, LIM, SAN, CEN, SMN and VIS). Each data point represents voxel-wise measurements in each brain network hub ROIs, pooled across two monkeys. Symbols on the plots represent individual monkeys’ mean normalized functional connectivity values (circle/triangle, monkey N/T, respectively). Error bars show s.e.m. Colored bars are the functional connectivity between stimulated side SCC and network hubs in the stimulated hemisphere (red for DMN hubs, blue for LIM hubs, green for CEN hubs, yellow for SAN hubs, purple for SMN hubs and pink for VIS hubs). The gray boxes show the mean functional connectivity with stimulated side SCC in the control hemisphere. Lighter color or lighter gray bars indicate pre-DBS, and the darker color or darker gray bars represent post-DBS scores. Number of voxel-wise measurements (n): stimulation side: DMN, n = 857 (pre), 857 (post); LIM, n = 821 (pre), 821 (post); SAN, n = 262 (pre), 262 (post); CEN, n = 1102 (pre), 1,102 (post); SMN, n = 1,239 (pre), 1,239 (post); VIS, n = 1,217 (pre), 1,217 (post); control side: DMN, n = 857 (pre), 857 (post); LIM, n = 821 (pre), 821 (post); SAN, n = 262 (pre), 262 (post); CEN, n = 1,102 (pre), 1,102 (post); SMN, n = 1,239 (pre), 1,239 (post); VIS, n = 1,217 (pre), 1,217 (post). Functional connectivity values were compared using three-way ANOVA. Statistical comparisons assessed differences between pre- and post-DBS conditions within each hemisphere (stimulated versus control) and brain network hub ROIs. All tests were two-sided. Tukey–Kramer post hoc methods corrected for multiple comparisons. There was a significant interaction between PrePost and side (F_1,21967 = 51.09, P = 9.10 × 10⁻¹3, η_p² = 0.0023). A significant PrePost difference was observed in the DMN (stimulation side, P < 1.00 × 10⁻³⁰⁰; control side, P < 1.00 × 10⁻³⁰⁰), LIM (stimulation side, P < 1.00 × 10⁻³⁰⁰; control side, P < 1.00 × 10⁻³⁰⁰), CEN (stimulation side, P < 1.00 × 10⁻³⁰⁰; control side, P < 1.00 × 10⁻³⁰⁰), SMN (stimulation side, P = 2.83 × 10⁻²⁰; control side, P = 1.21 × 10⁻¹⁴) and control side SAN (stimulation side, P = 1.00, NS; control side, P = 6.20 × ¹⁰⁻⁷). *P < 0.05, **P < 0.01, ***P < 0.001.

Source data

Extended Data Fig. 7. Functional connectivity with SCC is altered consistently across individual subjects by SCC-DBS.

The left and middle columns show functional connectivity maps with the stimulated SCC for each monkey at the pre-DBS (top row) and post-DBS (second row) time points (left column: monkey N; middle column: monkey T). The third row presents the difference maps (post-DBS functional connectivity – pre-DBS functional connectivity) for each monkey. The sagittal slices show the stimulated hemisphere. For axial images of monkey N, the left and right sides were flipped to display the stimulated hemisphere on the right side of the image. Green circles indicate the location of the stimulated contact of the SCC-DBS lead. The right column displays averaged functional connectivity changes with right SCC in non-surgical control group, shown in sagittal (left, right hemisphere is shown) and axial (right) views. Images are thresholded at P < 0.01 with a minimum cluster size of >20 voxels (cluster-level corrected p < 0.01, two-sided). Regions outlined in black indicate statistically significant clusters.

Following 6 weeks of SCC-DBS, there was a marked reduction in functional connectivity between the SCC and the PCC as well as the vmFC. Reductions were also apparent between the SCC and HIP, AMY and DLPFC (Fig. 5a, third row; Fig. 5b; Extended Data Fig. 7, bottom row in left and middle columns; Extended Data Fig. 8, first to third rows; and Extended Data Fig. 9; three-way ANOVA, F_1,20279 = 186.48, P < 0.001; post hoc Tukey–Kramer tests, vmFC, PCC, STG, HIP, AMY, DLPFC, dACC: P < 0.001). These reductions in functional connectivity from SCC were consistent across both monkeys (Extended Data Figs. 7 and 8). Notably, such reductions in connectivity were absent in the nonsurgical control group, indicating that these changes are specific to SCC-DBS (Fig. 5a, bottom row; Extended Data Fig. 7, right column; and Extended Data Fig. 8, bottom row). DBS-induced changes in functional connectivity were, however, not limited to decreases. The insula and somatosensory cortex exhibited increases in functional connectivity with the SCC following DBS (Fig. 5a,b and Extended Data Fig. 9; post hoc Tukey–Kramer tests, anterior-dorsal insula, somatosensory cortex: P < 0.001). Thus, DBS caused both increases and decreases in functional connectivity with SCC.

Extended Data Fig. 8. Functional connectivity with SCC is altered bilaterally by SCC-DBS stimulation.

The mean functional connectivity changes with stimulated SCC are shown for the hemispheres on stimulation side and control side. Only functional connectivity results that survived thresholding of P < 0.05 and clustering of > 20 voxels are shown (cluster-level corrected p < 0.01, two-sided). All images are the subtraction of pre-DBS SCC functional connectivity from post-DBS SCC functional connectivity. The left panel shows the stimulated hemisphere and the right panel shows the control hemisphere. The first row is the average of 2 monkeys’ functional connectivity with stimulated SCC changes, the second row is the result of monkey N, and the third row is the result of monkey T. Green circles indicate the location of the stimulated contact of the SCC-DBS lead. The bottom row is the result of non-surgical control group.

Extended Data Fig. 9. Functional connectivity with stimulated SCC changes induced by SCC-DBS in brain network hubs.

A) The functional connectivity between stimulated SCC and 17 nodes per hemisphere included in the six brain networks. Colored bars are the mean value of functional connectivity normalized z-score between stimulated SCC and each brain network ROI hub in the stimulated hemisphere (red for the DMN hubs, cyan for SCC (area 25 and area 32), blue for the limbic network (LIM) hubs, yellow for the salience network (SAN) hubs, green for the central executive network (CEN) hubs, purple for the sensorimotor network (SMN) hubs and pink for the visual network (VIS) hubs. SCC, dACC and PCC were counted as hubs for the LIM (blue hubs) as well). These colors are matched with the ROI color in the centered brain image. The gray bars show the mean functional connectivity with SCC in the control hemisphere. Lighter color or lighter gray bars indicate pre-DBS functional connectivity z-score and the darker color or darker gray bars represent post-DBS scores. The gray circle represents monkey N and the gray triangle represents monkey T. Error bars show SEM. Bars represent the mean value of functional connectivity computed from voxel-wise measurements in each target ROI (brain network node ROI), pooled across two monkeys. Number of voxel-wise measurements (n) are shown in Supplementary Table 8. FC values were compared using 3-way ANOVA. Statistical comparisons assessed differences between pre- and post-DBS conditions within each hemisphere (stimulated or control) and target ROI. All tests were two-sided. Tukey–Kramer post-hoc methods to corrected for multiple comparisons. There was a significant interaction between PrePost and side (F_1,20279 = 186.48, P = 2.87e-42, ηp² = 0.0091). P values for the post hoc comparisons are provided in Supplementary Table 9. * P < 0.05, ** P < 0.01, *** P < 0.001. B) Summary table of functional connectivity between the stimulated SCC and brain network hubs, corresponding to the data shown in panel A. Statistical significance was assessed using the same 3-way ANOVA and Tukey–Kramer post hoc tests as described in panel A. All tests were two-sided, and Tukey–Kramer post hoc tests were used to correct for multiple comparisons. Arrows indicate significant DBS-related effects (pre-DBS vs post-DBS within each hemisphere and region, P < 0.05). Red arrows indicate significant increases in functional connectivity with stimulated SCC and blue arrows indicate significant decreases. Exact P values are identical to those reported in Supplementary Table 9.

Source data

Despite DBS being delivered unilaterally and white matter remodeling being confined to the stimulated hemisphere (Figs. 1 and 2), functional connectivity changes were also apparent in the unstimulated control hemisphere (Fig. 5b,c and Extended Data Figs. 8 and 9). This matches findings that functional connectivity is not solely defined by structural connectivity²⁰ and that unilateral SCC-DBS in humans bilaterally impacts neural activity²¹. Importantly, bilateral functional connectivity changes were not observed in the nonsurgical control group (Fig. 5a, bottom row; Extended Data Fig. 7, right column; and Extended Data Fig. 8, bottom row).

Next, we investigated network-level functional connectivity changes using a hub-based approach (Methods). Here we calculated functional connectivity between SCC and each of six intrinsic functional networks: the default mode network (DMN), salience network (SAN), limbic network (LIM), central executive network (CEN), sensorimotor network (SMN) and visual network (VIS). After chronic stimulation, the DMN, LIM and CEN showed significantly decreased functional connectivity with SCC (post hoc Tukey–Kramer tests, DMN: P < 0.001, LIM: P < 0.001, CEN: P < 0.001) (Fig. 5c). These networks and areas within them receive connections through the CB. For example, the PCC and vmFC, both key areas in the default mode network, receive CB inputs¹⁶ and showed reduced connectivity induced by DBS (Fig. 5b and Extended Data Fig. 9). In comparison, there was no change in relation to the VIS or areas within the VIS (Fig. 5c and Extended Data Fig. 9). Such a pattern of changes fits with the idea that SCC-DBS preferentially influences activity in areas that connect via the CB, which include the DMN, CEN and LIM^16,22. There were, however, also changes in networks with limited direct CB connections, such as the SMN (post hoc Tukey–Kramer tests, SMN: P < 0.001) (Fig. 5c).

The SAN, which has also been implicated in depression²³, did not as a whole exhibit consistent changes after SCC-DBS (Fig. 5c, yellow network). When looking more closely at the individual areas within this network, however, a more varied pattern emerges across areas. Specifically, the anterior-dorsal insula exhibited increased connectivity with the SCC, whereas the dACC showed decreased connectivity following 6 weeks of DBS (post hoc Tukey–Kramer tests, anterior-dorsal insula: P < 0.001, dACC: P < 0.001) (Fig. 5b and Extended Data Fig. 9). Consequently, although both regions showed significant changes in their connectivity with the SCC, because these nodes changed in opposite directions, the net effect of DBS on the SAN was effectively zero. This highlights that SCC-DBS did not have a uniform effect on all nodes within an intrinsic network, and in the SAN it appears to modulate the balance of connectivity with the SCC.

Taken together, SCC-DBS was associated with brain-wide functional modulation characterized by a selective shift of functional connectivity between the SCC and the rest of the brain. Notably, this also included areas that are not directly anatomically linked to the SCC through the CB. This pattern of effects is thus consistent with prior neuroimaging investigations of depression that implicate altered connectivity between the SCC and the DMN and SAN^11,23–26.

The impact of SCC-DBS lead insertion on functional and structural brain networks

The preceding sections suggest that SCC-DBS induced both white matter remodeling and selective alterations in large-scale brain networks. However, previous research has shown that even small brain lesions, similar to that which occurs as a consequence of DBS lead insertion, can result in structural and functional changes in the brain²⁷. To dissociate the possible effects of the DBS lead insertion from the SCC-DBS, we conducted a nonstimulation control experiment. In a single animal, we implanted a DBS lead using a protocol identical to the one used in the SCC-DBS group (Fig. 1), but no stimulation was delivered. In addition to behavioral assessments, diffusion-weighted and functional MRI scans were obtained with the same experimental timeline as the SCC-DBS group. Finally, microstructural white matter changes in the mid-cingulate portion of the CB were assessed histologically.

Across all modalities, including behavior, diffusion imaging, histology and functional MRI, the nonstimulation control animal exhibited patterns distinct from those observed in the SCC-DBS group. First, there was no change in the amount of movement or foraging in naturalistic behavior (Extended Data Fig. 1, monkey D). Second, in direct contrast to the SCC-DBS animals, we saw a significant decrease in FA in the CB of the implanted hemisphere and no changes in any of our other target tracts (three-way ANOVA, F_1,1238 = 6.02, P = 0.0025; post hoc Tukey–Kramer test, P < 0.001; Fig. 6a and Extended Data Fig. 10a). This decrease in FA is more anterior and the opposite direction of that observed in the stimulation group, although diffuse reductions were seen across the CB (Fig. 6b). Third, histological analysis confirmed the absence of lead-induced microstructural changes: the ratio of myelinated oligodendrocytes (CC1/DAPI) in the MCC-CB did not differ between the implanted and control hemispheres (paired t-test, t₂₇ = −0.99, P = 0.33, s.d. = 9.95; Fig. 6 and Extended Data Fig. 10b). Together, these findings indicate that in the absence of stimulation, the surgical implantation of the DBS lead does not result in increased myelination within the cingulum and, if anything, is associated with a decrease in FA.

Fig. 6. Structural and functional changes in a nonstimulation control animal implanted with an SCC-DBS electrode but without stimulation.

a, FA changes in the three white matter tracts following DBS lead insertion in a single nonstimulation control animal (DBS lead implanted, but no stimulation applied, monkey D). Top: subtraction of postimplantation FA (6 weeks after DBS lead implantation) from preimplantation FA values (before implantation) in the CB (left), UF (middle) and FM (right). Inset color scale bar indicates positive (red) to negative (blue) change in FA. The sagittal images show the sham hemisphere where a DBS lead was implanted. Bottom: corresponding box plots represent the IQR (25th–75th percentile) of FA values computed from voxel-wise measurements in each tract in each hemisphere from one monkey. Center black line is the median. Whiskers extend to 1.5 × IQR. Outliers are not displayed. Colored boxes indicate the mean FA value in the tract mask from the sham hemisphere (DBS-lead-implanted hemisphere, yellow for CB, blue for UF, red for FM), and the gray boxes are those from the control hemisphere. Lighter color/gray boxes represent pre FA and darker bars demonstrate post FA. Number of voxel-wise measurements (n): sham side: CB, n = 162 (pre), 162 (post); UF, n = 120 (pre), 120 (post); FM, n = 64 (pre), 64 (post); control side: CB, n = 148 (pre), 148 (post); UF, n = 74 (pre), 74 (post); FM, n = 57 (pre), 57 (post). FA values were compared using three-way ANOVA. Statistical comparisons assessed differences between pre- and post-implantation within each hemisphere (sham or control) and tract. All tests were two-sided. Tukey–Kramer post hoc methods corrected for multiple comparisons. There was a significant interaction between PrePost and side and tract (F_1,1238 = 6.02, P = 0.0025, η_p² = 0.0096). A significant PrePost difference was observed in the CB on the sham hemisphere (P = 5.07 × 10⁻⁹). ***P < 0.001. b, FA value changes in subparts of the CB in a nonstimulation control animal. Mean (±s.e.m) FA value in four subparts (SCC, dACC, MCC and PCC portion) of the CB before and after DBS lead implantation in sham (left graph) and control hemisphere (right graph) were plotted. Each FA value computed from voxel-wise measurements within each CB subregion, collected from one monkey. Number of voxel-wise measurements (n): sham side: SCC, n = 44 (pre), 44 (post); dACC, n = 58 (pre), 58 (post); MCC, n = 23 (pre), 23 (post); PCC, n = 16 (pre), 16 (post); control side: SCC, n = 28 (pre), 28 (post); dACC, n = 53 (pre), 53 (post); MCC, n = 13 (pre), 13 (post); PCC, n = 19 (pre), 19 (post). FA values were compared using three-way ANOVA. Statistical comparisons assessed differences between pre- and post-implantation within each hemisphere (sham or control) and CB subregion. All tests were two-sided. Tukey–Kramer post hoc methods corrected for multiple comparisons. There was a significant interaction between PrePost and side (F_1,492 = 24.55, P = 9.97 × 10⁻⁷, η_p² = 0.048). A significant PrePost difference was observed in the dACC-CB on the sham hemisphere (P = 1.30 × 10⁻⁸). *P < 0.05, ***P < 0.001. c, Oligodendrocyte changes in MCC-CB in a nonstimulation control animal. Box plot of the ratio of CC1- and DAPI-positive cells (%). The box represents the IQR (25th–75th percentile). Green line is the median. Whiskers extend to 1.5 × IQR. Outliers are not displayed. Each data point represents measurements from individual fields (seven fields per hemisphere per section) across four coronal sections, collected from one monkey. The number of field-wise measurements (n) used for statistical analysis was as follows: sham hemisphere, n = 28 (7 fields × 4 sections × 1 subject); control hemisphere, n = 28. CC1/DAPI ratios were compared between hemispheres using two-sided paired t-test. Statistical comparisons assessed differences between sham side and control side. No significant change was observed (paired t-test, t₂₇ = −0.9888, P = 0.3316, s.d. = 9.9474, effect size (Cohen’s dz) = −0.19). d, Functional connectivity changes with SCC in the hemisphere implanted with a DBS lead in a nonstimulation control animal, shown on sagittal (left, the sham hemisphere) and axial (right) views. Images are thresholded for functional connectivity, with P < 0.01 and cluster size >20 (cluster-level corrected P < 0.01, two-sided). Regions outlined in black indicate statistically significant clusters. The green circle represents the location of the DBS lead. e, Intrinsic brain network-level changes in functional connectivity in a nonstimulation control animal. Changes in functional connectivity between the sham-side SCC (cyan region) and six brain network hubs (hubs of the DMN, LIM, SAN, CEN, SMN and VIS) are shown. Box plots represent the IQR (25th–75th percentile) of functional connectivity values computed from voxel-wise measurements from one monkey. Center black line is the median. Whiskers extend to 1.5 × IQR. Outliers are not displayed. Colored boxes are the functional connectivity between sham-side SCC and network hub ROIs in the sham hemisphere (red for the DMN hubs, blue for the LIM hubs, green for the CEN hubs, yellow for the SAN hubs, purple for the SMN hubs and pink for the VIS hubs). The gray boxes show the mean functional connectivity with sham-side SCC in the control hemisphere. Lighter color or lighter gray bars indicate pre-, and darker color or darker gray bars represent post-implantation scores. Number of voxel-wise measurements (n): sham side: DMN, n = 453 (pre), 453 (post); LIM, n = 438 (pre), 438 (post); SAN, n = 137 (pre), 137 (post); CEN, n = 552 (pre), 552 (post); SMN, n = 637 (pre), 637 (post); VIS, n = 602 (pre), 602 (post); control side: DMN, n = 404 (pre), 404 (post); LIM, n = 383 (pre), 383 (post); SAN, n = 125 (pre), 125 (post); CEN, n = 550 (pre), 550 (post); SMN, n = 602 (pre), 602 (post); VIS, n = 615 (pre), 615 (post). FC values were compared using three-way ANOVA. Statistical comparisons assessed differences between pre- and post-implantation conditions within each hemisphere (sham or control) and network hub ROIs. All tests were two-sided. Tukey–Kramer post hoc methods corrected for multiple comparisons. There was a significant interaction between PrePost and side (F_1,10972 = 7.36, P = 0.0067, η_p² = 0.00067). A significant PrePost difference was observed in the control-side DMN (P = 8.20 × 10⁻⁸), control-side CEN (P = 0.0088), sham-side SMN (P = 1.38 × 10⁻⁴) and control-side SAN (P = 1.06 × 10⁻⁴). *P < 0.05, **P < 0.01, ***P < 0.001.

Source data

Extended Data Fig. 10. Structural changes and functional changes in a non-stimulation control animal.

A) Fractional anisotropy (FA) in the three white matter tracts before and 6 weeks after DBS lead implantation in a non-stimulation control animal (DBS implanted, but no stimulation applied, monkey D). Extracted FA values are shown from the cingulum bundle (CB, top), the uncinate fasciculus (UF, middle), and the forceps minor (FM, bottom). The left column shows pre-implantation FA value, the middle column is post-implantation FA value, and the right column is the subtraction image of post-implantation FA and pre-implantation FA values. The color indicates FA value. The sagittal images show the sham hemisphere where a DBS lead was implanted. B) Representative immunofluorescence images showing myelinated oligodendrocytes (CC1, marker for myelinated oligodendrocytes, red) and nuclei (DAPI, marker for nuclei, blue) in the mid-cingulate portion of the cingulum bundle (MCC-CB) from a non-stimulation control animal. The top panel shows the control hemisphere, and the bottom panel shows the sham hemisphere (DBS lead implanted hemisphere). Yellow scale bars indicate 100μm. C) Brain-wide functional connectivity in cortical and subcortical areas based on CHARM and SARM atlas (level 4). Confusion matrices represent the functional connectome in pre-implantation scans (left column), post-implantation scans (middle column), and the difference between pre and post scans (right column), respectively. Color represents z-transformed correlation coefficients. The right graph is averaged functional connectivity. Box plots represent averaged functional connectivity (z score) for each ROI of the CHARM and SARM atlas. The box represents the interquartile range (IQR; 25th–75th percentile). Central black line is the median. Whiskers extend to 1.5 × IQR. Outliers are not displayed. Each data point represents measurements from individual ROI (pre, n = 122; post, n = 122), collected from one monkey. Colors represent significant negative (blue) and positive changes (orange) of functional connectivity from zero in each ROI (P < 0.01 with Bonferroni correction, rank-sum test). Statistical comparisons assessed differences between pre and post conditions. All tests were two-sided. Exact P values are shown in Supplementary Table 10. Effect sizes (rank-biserial correlation, r_rb) are shown in Supplementary Table 11. The dotted lines divide cortical (left) and subcortical (right) ROIs. D) Functional connectivity changes with SCC in the hemisphere implanted with a DBS lead in a non-stimulation control animal, shown on sagittal (left, the sham hemisphere) and axial (right) views. The top row is pre-implantation of a DBS lead, the middle is 10 weeks after implantation without stimulation, and the bottom row is the subtracted results between pre and post implantation. Images are thresholded for functional connectivity with P < 0.01 and cluster size >20 (cluster-level corrected p < 0.01, two-sided). Regions outlined in black indicate statistically significant clusters. The right graphs are functional connectivity changes with SCC on the sham side induced by DBS lead implantation in the default mode and salience network hubs. Colored bars are the mean value of functional connectivity normalized z-score between stimulated SCC and each brain network ROI hub in the sham hemisphere (red for the DMN hubs, and yellow for the SAN hubs). The gray bars show the mean functional connectivity with SCC in the control hemisphere. Lighter color or lighter gray bars indicate pre-implantation functional connectivity z-score with stimulated SCC and the darker color or darker gray bars represent post-implantation scores. Box plots represent the interquartile range (IQR; 25th–75th percentile) of functional connectivity computed from voxel-wise measurements from one monkey. Center black line is the median. Whiskers extend to 1.5 × IQR. Outliers are not displayed. Number of voxel-wise measurements (n): sham side: vmFC, n = 66 (pre), 66 (post); PCC, n = 138 (pre), 138 (post); Anterior-dorsal insula, n = 49 (pre), 49 (post); dACC, n = 88 (pre), 88 (post); control side: vmFC, n = 43 (pre), 43 (post); PCC, n = 111 (pre), 111 (post); Anterior-dorsal insula, n = 51 (pre), 51 (post); dACC, n = 74 (pre), 74 (post). FC values were compared using 3-way ANOVA. Statistical comparisons assessed differences between pre- and post-implantation within each hemisphere (sham or control) and target ROI. All tests were two-sided. Tukey–Kramer post-hoc methods to corrected for multiple comparisons. There was a significant interaction between PrePost and side (F_{1, 10106} = 4.58, P = 0.0323, ηp² = 0.00045). A significant pre–post difference was observed in the sham side vmFC (P = 7.83e-04) and control side anterior-dorsal insula (P = 1.37e-12). * P < 0.05, ** P < 0.01, *** P < 0.001.

Source data

Finally, we found some changes in functional connectivity following lead implantation. When we looked at whole-brain functional connectivity (Extended Data Fig. 10c), we saw less coherent changes between ROIs compared with the group that received stimulation (Fig. 4). Cortical-to-cortical connections showed more random connectivity changes, and subcortical-to-subcortical connections showed a consistent negative change, which is the opposite direction of what we saw after SCC-DBS (Fig. 4). Additionally, there was very little alteration in the functional connectivity associated with the SCC, except for a localized increase in connectivity at the site of implantation (Fig. 6d and Extended Data Fig. 10d). Notably, the widespread reductions in connectivity that we saw in the DMN, LIM and CEN after SCC-DBS were not observed (compare Fig. 5c and Fig. 6e). Taken together, this experiment demonstrates that the structural and functional changes observed following SCC-DBS are not attributable to the effects of DBS lead insertion alone and thus underscores the specific role of chronic stimulation in driving structural and functional changes in the brain.

Discussion

The present results indicate that SCC-DBS is associated with focal white matter structural remodeling in the MCC-CB (Figs. 2 and 3) as well as a brain-wide shift in the functional connectivity patterns across multiple networks (Figs. 4 and 5). These stimulation-induced changes could not be explained by the alterations associated with implantation of the DBS lead alone (Fig. 6). Our results highlight the potential role of white matter plasticity in the efficacy of DBS therapies for psychiatric and neurological disorders.

DBS induced changes in white matter macro- and microstructure

Prior work has shown that white matter remodeling can be induced by electrical, optogenetic or chemogenomic stimulation of neurons in gray matter^28–30. Our data reveal that DBS targeted to white matter induces both macro- and micro-structual changes in white matter in the CB distal to the site of stimulation. This pattern of results complements prior studies that have pointed to a role for white matter dysfunction as contributing to the pathophysiology of depression^31–33. Notably, abnormal white matter integrity in the CB was recently found to be positively correlated with symptom severity in patients with treatment-resistant depression¹³. In this context, our observation that SCC-DBS selectively induces white matter remodeling in the CB provides additional evidence that inducing structural plasticity in this tract may be a key mechanism through which DBS for depression mediates its therapeutic effects and results in sustained recovery over years with ongoing stimulation⁸.

Surprisingly, only the CB exhibited a change in white matter integrity following SCC-DBS, despite chronic stimulation of the three targeted tracts, as used with clinical DBS, all of which have been shown to be necessary for clinical response⁷. Several mechanisms could account for this selective effect. First, unlike the FM or the UF, the CB uniquely contains long-range dopaminergic and serotonergic projection fibers projecting to the dorsal and medial aspects of the brain¹⁶. Given the anatomical proximity of the DBS electrode to the CB, stimulation may preferentially modulate these monoaminergic pathways, contributing to structural changes. Second, despite the volume of activated tissue encompassing the CB, UF and FM, the extent of stimulation across the three tracts may not have been equal. Although the DBS target was located at the anatomical confluence of the CB, FM and UF, the actual number of fibers stimulated within each tract may have varied due to subtle anatomical differences and electrode position. Third, it is possible that structural remodeling occurs on different timescales across tracts. Previous clinical reports suggest that symptom improvement following SCC-DBS can continue for up to 2 years⁸. Changes in the FM and UF may emerge only after longer durations of stimulation and/or be secondary to plastic changes in the CB. Finally, it is critical to acknowledge that the animals in the current study were healthy, whereas the studies of patients with depression all involved individuals in pathological states. On this view, stimulation of pathological brains and specific tracts may have a different time course of response to DBS. Taken together, these factors highlight the complexity of SCC-DBS effects on distributed white matter circuits and underscore the need for future studies using alternative approaches to fully characterize tract-specific responses.

In addition, although DBS caused an increase in white matter integrity in the MCC-CB (Figs. 2 and 3), implantation of the DBS lead without stimulation was associated with a slight decrease in FA (Fig. 6b), and there was no discernible change in myelinating oligodendrocytes (Fig. 6c). This suggests that lead implantation may induce myelin and axonal damage to white matter without enhancing oligodendrocyte proliferation. Importantly, glial responses, particularly astrocytic reactions, to electrode implantation have gained increasing attention not merely as passive physiological responses but as potential effectors of neural remodeling³⁴. In the peripheral nervous system, astrocytic scars following injury promote axon recovery from damage to the spinal cord³⁵. Moreover, interactions between oligodendrocytes and astrocytes are critical for myelin remodeling³⁶. Thus, it is possible that the combination of damage caused by lead implantation and DBS promote remodeling of the white matter, especially in the CB.

Effect of SCC-DBS on functional networks

A host of prior research has found that there are differences in resting-state network connectivity between healthy individuals and people with depression³⁷. Moreover, therapeutic interventions for depression, such as antidepressants and electroconvulsive therapy, produce reproducible changes in functional connectivity^11,38,39. In our data, SCC-DBS–induced changes in functional networks were far more widespread (Figs. 4 and 5) in contrast to the localized white matter remodeling observed in the mid-cingulum. Such a localized change in white matter supporting larger-scale changes in functional networks is consistent with prior work showing that even a small number of indirect structural connections can support and maintain large-scale functional connectivity across the brain⁴⁰.

SCC-DBS was associated with a broad decrease in cortico–cortical connectivity and an increase in subcortico–subcortical connectivity (Fig. 4). Specifically, we found a decrease in connectivity between the SCC and the DMN, CEN and LIM (Fig. 5). Altered connectivity within the DMN is one of the most consistent findings in people with depression³⁷ and is hypothesized to be related to pathological rumination^26,37. Notably, patients with depression exhibit greater coupling between the SCC and the DMN relative to healthy controls^25,26, and SCC-DBS reduces metabolic activity in the SCC and increases activity in the DMN hubs in association with clinical recovery^4,11. Our findings add further credence to the idea that altering connectivity within the DMN and specifically between the SCC and PCC may be essential for restoring normal resting-state network activity, which is necessary for facilitating the recovery from depression.

Functional network connectivity was, however, not uniformly decreased. We observed an increase in connectivity between the SCC and the SMN, but a rebalancing of connectivity within the SAN (Fig. 5b). Recently, a central role of the SAN in depression has garnered substantial attention²³. The changes that we observed between the SCC, insula and dorsal ACC indicate that SCC-DBS has a rebalancing effect on SAN functional interactions. This contrasts with other brain networks, where SCC connectivity tended to change in a consistent direction. These findings underscore that the differential response both within and across functional networks may be a key mechanism through which stimulation alters pathologically altered networks in depression.

As noted above, there was an increase in connectivity between the SCC and the SMN (Fig. 5c), and at the same time we also found that SCC-DBS was associated with increased in-cage behavior (Extended Data Fig. 1). This would seem to align with behavioral changes seen in humans receiving SCC-DBS, where improvements in psychomotor slowness and increases in motivated activity are correlated with clinical improvement⁴. In the context of the present experiment, it is important not to overinterpret the relationship between changes in motor network connectivity and behavior (as well as alterations in white matter structure). Notably, we found that there were no apparent changes in connectivity between the stimulated SCC and motor-related regions, including the primary motor cortex or supplementary motor area. Thus, we interpret this as indicating that other network-level mechanisms, such as those involving motivational circuits or shifts in the balance among large-scale brain networks, may underlie the increase in in-cage behavior.

Interaction between functional and structural effects of DBS

Based on the progressive response observed in depressed patients receiving therapeutic bilateral SCC-DBS^4,7,8, it is likely that the previously discussed structural and functional changes dynamically interact. Brain networks are known to dynamically rebalance to optimize input and outputs between brain areas⁴¹, and SCC-DBS causes rapid changes in brain activity, especially in the DMN^11,12. On the other hand, remodeling white matter structures can take weeks⁴², with its effects on network function likely being more sustained^11,12. In this context, SCC-DBS may be associated with a rapid shift in the balance of functional brain networks and behavior followed by more protracted white matter structural remodeling to strengthen modified functional connections or potentially to compensate for them⁴³. This is consistent with the concept of activity-dependent plasticity^18,19,29,44, highlighting the reciprocal and dynamic interaction between brain structure, function and behavior.

We found that SCC-DBS induced white matter remodeling specifically in the mid-cingulate portion of the CB (Fig. 2). One possibility for why this portion near the MCC was modified is that the change in white matter may be related to the alterations in functional brain networks. The mid-cingulate cortex has been previously described as a central hub based on its functional and anatomical connectivity⁴⁵. Hub regions tend to exhibit greater structural and/or functional connectivity than other regions, and it is hypothesized that this is associated with a role in integrating information within or across networks^45,46. Several different white matter projections course through the mid-cingulate portion of the CB near to the MCC, connecting to ventromedial PFC, ACC, DLPFC, PCC, AMY and thalamus¹⁶. These microstructural and indirect connections in the MCC support a diverse range of functions and are difficult to detect using diffusion-weighted imaging alone⁴⁷. This may partly explain the absence of detectable functional connectivity changes in the MCC. Regardless, because the mid-cingulate cortex sits at the confluence of the DMN, SAL and SMN, the DBS-induced shift in brain-wide functional connectivity could be a key factor in driving white matter structural remodeling.

One seemingly puzzling result that we reported here is that increased white matter integrity in the CB (Figs. 2 and 3) was accompanied by reduced functional connectivity between the SCC and PCC (Fig. 5). Mismatches between structural and functional connectivity have been previously discussed in several studies²⁰. Additionally, several mechanisms may explain this apparent discrepancy. First, functional connectivity is not strictly constrained by structural connectivity⁴⁸. Second, 130-Hz stimulation has been shown to block neural communication by interfering with intrinsic neural activity³. Third, it is possible that increased structural connectivity represents a compensatory adaptation to reduced functional connectivity between the SCC and PCC. In addition, the direction of functional changes may depend on whether the enhanced projections are excitatory or inhibitory. Although the projections from the SCC to the PCC are likely glutamatergic, local circuit dynamics within the PCC mean that SCC-PCC connections may cause local inhibition as they connect to inhibitory interneurons. Under this framework, increased activity in this pathway may promote myelination while simultaneously disrupting synchrony or increasing inhibitory tone between the SCC and PCC, leading to a reduction in functional connectivity.

Interpretational limitations

Given the inherent constraints of nonhuman primate research, the current study used a within-subjects approach with a small sample size and was designed to focus on stimulation delivered through the contact located nearest to the confluence of the three white matter tracts (Fig. 1a). Consequently, we show both group and individual-level results to demonstrate consistency and our findings are supported by multiple complementary approaches (diffusion MRI, histology and functional imaging), emphasizing the robustness of the effects.

One key consideration for interpreting the effects of DBS on brain-wide functional connectivity relates to our experimental design. To ensure consistent and reproducible results free from motion artifacts, all scanning was conducted under anesthesia, which is known to influence neural activity and resting-state connectivity patterns. Although the use of anesthesia may have attenuated and thus underestimated the measurable impact of the DBS effects on brains, we applied a low-dose isoflurane approach to maximize preservation of resting-state networks, as has been demonstrated in previous studies⁴⁹. To obtain reliable diffusion and fMRI signals from near to the SCC, DBS leads were removed before scanning. One limitation of this approach is the potential for a rebound effect in functional connectivity following the cessation of stimulation. We note, however, that leaving the lead in place would have disrupted MRI signals in midline areas, including the SCC; this would have severely compromised MRI analyses between the SCC and the rest of the brain, which was our focus here.

Our focus on clinically relevant stimulation meant that we did not explore the effects of stimulation on different lead contacts or of individual bundles. We note, however, that the available evidence in humans indicates that stimulation outside of the confluence of the CB, UF and FM would lead to the engagement of a set of circuits distinct from those highlighted here and is not associated with clinical response^12,21. We also used stimulation parameters that are routinely applied in patients receiving DBS for depression. Again, this was because our focus was on how DBS used to treat depression changes anatomy and function of the brain. It remains an open question as to how stimulation delivered at different frequencies and amplitudes would impact our findings, as it is known that different frequencies of DBS for Parkinson’s disease elicit distinct patterns of symptom recovery⁵⁰.

Summary and conclusion

Our study demonstrates that SCC-DBS induces white matter remodeling in parallel with brain-wide alterations in functional network interactions in nonhuman primates. Specifically, we show that localized increases in myelination within the CB, together with widespread suppression of cortico–cortical functional connectivity and rebalancing across multiple brain networks, most notably DMN and SANs. Our findings provide insight into the mechanisms engaged by DBS and may offer opportunities to further optimize the efficacy of DBS as well as deepen our understanding of the pathophysiology of depression. Further investigation into the relationship between structural and functional changes induced by chronic stimulation of white matter in animal models could also inform the development of white-matter-targeted DBS interventions for a broader range of neurologic and psychiatric disorders.

Methods

Subjects

Three male adult rhesus macaques (Macaca mulatta, monkeys N, T and D), aged 7–9 years, were used. Monkeys N and T received SCC-DBS stimulation, and monkey D served as a nonstimulation control (SCC-DBS implanted but not stimulated). Subjects had ad libitum access to food and water and were maintained on a 12-hour light/dark cycle. All three animals underwent SCC-DBS implantation, MRI scans (T1 weighted image (T1w), rs-fMRI and DWI), behavioral assessments and histological experiments. For nonsurgical control comparisons, rs-fMRI data were obtained from three additional unoperated adult male rhesus macaques (Macaca mulatta, monkeys B, W and H), aged 7–8 years. All procedures were approved by the Institutional Animal Care and Use Committee of Icahn School of Medicine at Mount Sinai.

Acquisition and processing of neuroimaging data

Two MRI sessions were conducted in each subject (monkeys N, T and D) to acquire T1w, DWI and rs-fMRI before SCC-DBS implantation as baseline scans (pre-DBS MRI) and immediately after the 6 weeks of chronic stimulation (post-DBS MRI) (Fig. 1c). In monkey D, no stimulation was delivered, and the post-DBS MRI was acquired after 6 weeks of lead implantation without stimulation. Approximately 20 min before the post-DBS MRI, the DBS system, including the implanted DBS lead, was removed to avoid dropout and artifacts in the MRI images as well as histological damage caused by DBS lead and extension heating. For the nonsurgical controls (monkeys B, W and H), rs-fMRI and T1w scans were acquired 6–7 weeks apart under identical imaging conditions, enabling assessment of time-dependent changes in functional connectivity independent of surgical intervention.

All scans were acquired on a 3 Tesla Siemens Skyra scanner (Siemens Healthineers) and were conducted under general anesthesia to minimize motion artifacts. Animals were sedated using ketamine (5 mg kg⁻1) and dexmedetomidine (0.0125 mg kg⁻¹) and were then intubated and maintained on isoflurane (0.7–3% to effect). Throughout the scan acquisitions, the plane of anesthesia was continuously monitored by certified veterinarian staff monitoring a suite of vital signs (pulse, oxygen saturation, end-tidal CO₂, capnograph, blood pressure, body temperature).

Using a human 32ch head coil (Siemens Healthineers), DWI (b = 1,000 s mm⁻², repetition time (TR) 6,900 ms, echo time (TE) 96.0 ms, 72 directions, 1.5 mm isotropic voxels, eight b₀ volumes per run, multiband acceleration = 3, phase-encoding directions = anterior-to-posterior (AP) and posterior-to-anterior (PA)) were acquired with the animal in the supine position. We collected six DWI runs in total, three with AP phase encoding and three with reversed phase encoding (PA). Two to three T1w images (0.5 mm isotropic, TR/TE 2,500/2.81 ms, flip angle 8°) were also collected. For both DWI and structural scans, the level of isoflurane was kept around 1–1.5% to ensure the stability.

rs-fMRI scans were performed as previously described⁵¹, using a custom-built four-channel phased array transmit/receive coil (Windmiller-Kolster Scientific). Anesthesia was strictly maintained at low-dose isoflurane (0.7–0.9%). Although anesthesia can alter functional connectivity compared to the awake state, prior studies have shown that low-dose isoflurane preserves the integrity of resting-state networks^49–55. To reduce stimulation, the animal’s head was not restrained but supported with towels and bandages to promote stability with minimal discomfort. A session-specific three-dimensional T1w image (0.5 mm isotropic, TR/TE 2,500/2.81 ms, flip angle 8°) was acquired. Following intravenous injection of monocrystalline iron oxide nanoparticle to improve the contrast-to-noise ratio of the functional data^56,57, six rs-fMRI runs were obtained (echo planar images (EPI): 1.6 mm isotropic, TR/TE 2,120/16 ms, flip angle 45°, 300 volumes per run). At the completion of all scans, subjects were extubated and continuously monitored until they were able to sit unaided.

Identification of white matter tracts for DBS targeting

In the current study, we targeted DBS to the confluence of three white matter tracts adjacent to the SCC. To identify this target, we reconstructed the CB, UF and FM separately using probabilistic tractography (FSL’s ProbtrackX)⁵⁸ applied to preprocessed pre-DBS DWIs⁵⁹ (see Supplementary Methods 1 for DWI preprocessing). Probabilistic tract reconstruction was conducted in each subject’s native diffusion space (maximum steps per sample = 3,200, number of samples = 10,000, step size = 0.1 mm, curvature threshold = 0.2). Seed, waypoint and exclusion masks were made with session-specific principal eigenvectors. These masks were created based on previous anatomical, tract-tracing and diffusion tractography studies^16,59–65 (Supplementary Methods 2). Each tract was log-transformed to account for the exponential decrease of visitation probability with distance and normalized by dividing each voxel’s value by the maximum value across the tract, removing the potential bias of differences in numbers of streamlines between tracts. Normalized tracts were subsequently thresholded with minimum and maximum values equal to 0.7 and 1, respectively, matching previous methods⁶⁰.

The confluence of the three tracts was determined in each animal as the location where the independently reconstructed tracts overlapped in an 8-voxel volume (3 mm × 3 mm × 3 mm). To confirm whether the confluence seed represented the optimal SCC-DBS target, we compared the tracts delineated from the confluence seed with those generated from nearby seeds of the same size. This comparison was performed using probabilistic tractography without exclusion or waypoint masks to identify candidate pathways potentially affected by DBS. The final target (3 mm × 3 mm × 3 mm) was selected as the seed that showed (1) ipsilateral CB, (2) ipsilateral UF and (3) bilateral FM, including the posterior fibers interconnecting left and right area 25 and the anterior fibers projecting to the left and right frontal poles.

To counterbalance stimulation laterality and ensure robust overlap of the three tracts at the stimulation site, the confluence in the left hemisphere was selected in monkey N and monkey D and the right side in monkey T.

SCC-DBS implantation surgery

DBS leads were implanted in monkeys N, T and D. A preoperative MRI scan (T1w and DWI) was used in each animal to target the confluence of three white matter tracts (the CB, UF and FM; see ‘Identification of white matter tracts for DBS targeting’). As described above, the optimal DBS target was determined in the pre-DBS DWI space and aligned to the pre-DBS T1w space. A DBS lead (four-contact mini-DBS lead, NuMed Inc.) was implanted stereotaxically in one hemisphere per subject (left in monkeys N and D; right in monkey T) under isoflurane (2–3%) general anesthesia. The DBS lead was connected to a DBS generator (Activa SC for Monkey N, Percept PC for monkey T, Medtronic) via a human DBS extension (cat. no. 37086, Medtronic). In monkey D, only a DBS lead was implanted through a small craniotomy without a DBS generator. The contralateral, nonimplanted hemisphere served as a within-subject control in monkeys N and T and as a mechanically unaffected control in monkey D. Detailed surgical procedures are provided in Supplementary Methods 3.

Lead placement was confirmed by high-resolution CT (0.5 mm isotropic CT, Force CT, Siemens Healthineers) 2 weeks after implantation. CT images were then aligned to the pre-DBS T1w scans for each subject. The CT-defined lead location was then compared with the predefined optimal DBS target (Fig. 1a, bottom).

SCC-DBS chronic stimulation

Based on the CT scan, the lead contact closest to the predefined optimal SCC-DBS target was selected for stimulation. After a 4-week postsurgical recovery period, monkeys N and T received 6 weeks of chronic stimulation (130 Hz, 90 μzsec, 5 mA). No stimulation was delivered in monkey D. We chose this stimulation regime as it has been shown to be effective parameters for human depression^4,7,66.

FA analysis

FA was calculated from pre-DBS and post-DBS DWI data. For each session, the FA map was first registered from DWI space to the corresponding session-specific T1w space (pre-DBS FA → pre-DBS T1w; post-DBS FA → post-DBS T1w) using FSL’s epi_reg (boundary-based registration)⁶⁷. The post-DBS T1 image was then nonlinearly registered to the pre-DBS T1w image using ANTs^68,69. The resulting transforms were converted to FSL-compatible formats (via Convert3D) and applied to the post-DBS FA map to bring both FA maps into a common space (pre-DBS T1w space) for within-subject whole-brain FA subtraction. FA values were extracted voxel-wise (1.5 mm isotropic) within each white matter mask (CB, UF and FM). Voxels overlapping with DBS lead tracts were excluded from the analysis. Pre-DBS FA and post-DBS FA were compared for each tract (CB, UF and FM) using a three-way ANOVA (DBS (pre-/post-) × side (stimulation/control) × tract (CB/UF/FM)), followed by a post hoc Tukey–Kramer test, revealing the white matter tracts that were most affected by SCC-DBS (MATLAB R2023a, Mathworks). Monkeys were included in the model as a fixed effect to account for between-subject variability but were not considered in the interpretation of results.

To localize FA changes within the CB, we divided the CB mask into four subregions based on an anatomical atlas (D99 atlas⁷⁰); SCC mask (CB adjacent to area 25 and area 32), dACC mask (CB adjacent to area 24), MCC mask (CB adjacent to area 24′) and PCC mask (CB adjacent to area 23). Similar to the above analysis, the pre-DBS FA and post-DBS FA in each CB subregion were compared using three-way ANOVA (DBS × side × subregion (SCC part/dACC part/MCC part/PCC part)), followed by a Tukey–Kramer post hoc test. Monkeys were modeled as a fixed effect to control for between-subject variability but were not incorporated into the interpretation of the results.

To visualize the detailed FA changes in each white matter tract, a per-slice mean FA was computed by averaging FA values across all voxels within a tract mask on each MRI slice. The CB mask was divided into two parts—an SCC part (axial slices) and an dACC, MCC and PCC part (coronal slices)—because of its C shape. The UF mask was also divided into two parts—the temporal part (axial slices) and the frontal-insula part (coronal slices)—because of its curved trajectory. The FM mask was divided into left and right hemispheres.

EPI data analysis

Functional imaging data were preprocessed with a custom AFNI/SUMA pipeline^51,71,72. Raw EPI images were converted to Neuroimaging Informatics Technology Initiative (NIfTI) and organized in Brain Imaging Data Structure (BIDS) format⁷³. T1w images were skull-stripped⁷⁴ and aligned to NIMH Macaque Template (NMT)-2.0 space^70,75. EPI data were slice-time corrected, the first three TRs were removed, and motion correction was applied. EPIs were aligned to the within session T1w image and warped to NMT space. EPIs were blurred with a full width at half maximum of 3 mm and converted to percent signal change. Finally, the motion derivatives from each scan along with cerebrospinal fluid and white matter signal were regressed out. The residuals from this analysis were then used to compute functional connectivity.

The full connectome analyses were computed using ANFI’s 3dNetCorr function^51,71,76. The correlations between regions of interest (ROIs) were computed at the fourth level of the cortical (Cortical Hierarchy Atlas of the Rhesus Macaque (CHARM atlas)⁷²) and subcortical (Subcortical Atlas of the Rhesus Macaque (SARM atlas)⁷⁷) hierarchical atlas. For each run, the functional connectomes were Fisher’s Z-transformed and then averaged within condition to create mean connectomes. We then calculated the difference between pre-DBS and post-DBS connectomes. Finally, the average difference connectome was computed across subjects. Functional connectivity changes (post − pre, Z-score) from each ROI to the other brain regions were averaged to investigate the regional effects caused by SCC-DBS or nonsurgical control. A rank-sum test was conducted to determine statistical significance between pre and post conditions (P < 0.01 with Bonferroni correction). Functional connectivity changes were further visualized as circular plots using the circular Graph toolbox run in MATLAB.

For targeted seed-based analyses, the seed ROI was defined as the SCC (area 25 in the D99 atlas⁷⁰) in the stimulated hemisphere. Target ROIs were defined from the D99 atlas⁷⁰, CHARM atlas⁷² and SARM atlas⁷⁷ (described in Supplementary Methods 4). We calculated the Fisher’s Z-transformed Pearson’s correlation between the average time series of the seed ROI and the time series of every voxel in the whole brain (AFNI’s 3dTcorr1D). Because the pre-DBS and post-DBS EPIs are different sessions, the Z-scores were normalized across session by dividing each voxel’s Z-score with the maximum Z-score for each session. To statistically evaluate the effect of SCC-DBS on stimulated area 25, the normalized Z-scores for pre-DBS and post-DBS in each target ROI were submitted to three-way ANOVA (DBS × side × target ROI) followed by post hoc analysis (Tukey–Kramer test). Monkeys were included in the model as a fixed effect to address between-subject variability but were not considered in the interpretation of results. The functional connectivity changes were calculated by subtracting pre-DBS Z-scores from post-DBS Z-scores. To examine the brain network effects of SCC-DBS, hub brain region target ROIs were combined for each network respectively: PCC, vmFC and STG for the DMN; anterior-dorsal insula and dACC for the SAN; DLPFC and PPC for the CEN; area 32, dACC, MCC, PCC, HIP and AMY for the LIM (dACC was overlapped with the SAN and the LIM; PCC was double-counted in the DMN and the LIM); primary motor cortex, somatosensory cortex and SMA for the SMN; and V1 (primary visual cortex) and MT (middle temporal area) for the VIS. To assess the statistical effect of SCC-DBS on brain network hubs, three-way ANOVA (DBS × side × brain network hubs) followed by post hoc Tukey–Kramer tests were conducted. Monkeys were modeled as a fixed effect to account for between-subject variability but were not considered in the interpretation of results. The statistical significance was determined when the P value was under 0.05.

Behavior analysis

Behavioral data were collected in monkeys N, T and D while subjects were alone in their home cage for 10 min without human intervention. To quantify overall activity (that is, the amount of movement), each video was divided into 60 bins (10 s per bin) and binarized such that bins containing movement (walking, climbing, foraging, and so on) were recorded as 1 and bins without movement as 0. The same approach was applied to foraging behavior. Behavioral responses to food pieces presented in the home cage, primarily reaching behavior, assessed potential neural deficits and asymmetries, including rotational behavior⁷⁸ and hand preference. These analyses were performed manually. We compared pre-DBS behavior (before stimulation start) with post-DBS behavior (after 6 weeks of stimulation) to establish the effect of SCC-DBS on natural home cage behavior. The binarized data was analyzed by one-way ANOVA to assess the effect of stimulation. Monkeys were included in the model as a fixed effect to account for between-subject variability but was not included in the interpretation of results. The statistical significance was set at P value < 0.05.

Immunofluorescence of myelinating oligodendrocytes

Details of animal perfusion and tissue preparation are provided in Supplementary Methods 5. Free-floating immunohistochemistry was performed on four serial coronal sections per animal, spaced 750 μm apart for monkey N and 790 μm apart for monkeys T and D, collected from the mid-cingulate level, the dorsal cingulate level or the region where the FM and UF overlap.

After three washes in phosphate-buffered saline (PBS) (10 min each), sections were immersed in target retrieval solution at 60 °C for 30 mins (target retrieval solution, citrate pH 6.1, cat. no. S1699, Agilent Dako, adjusted to pH 6.0 before use). Sections were then washed twice in PBS and incubated for 24 h at 4 °C in blocking solution (5% normal goat serum (cat. no. PI31873, Invitrogen) prepared in PBS containing 0.3% Triton X-100 (TX-100)). Sections were incubated for 48 h at 4 °C with anti-CC1 primary antibody (anti-APC[CC-1] mouse monoclonal antibody, clone CC-1, 1:5,000, cat. no. ab16794, RRID:AB_443473, Abcam), a marker of myelinating oligodendrocytes, diluted in the blocking solution. After washing, sections were incubated for 2 h at room temperature with goat anti-mouse secondary antibody (IgG (H+L) highly cross-adsorbed goat anti-mouse, Alexa Fluor Plus 555, 1:500, cat. no. A32727, RRID:AB_2633276, Thermo Fisher Scientific) diluted in PBS containing 0.3% TX-100. After washing, sections were stained with DAPI (DAPI nucleic acid stain molecular probes, cat. no. D1306, Invitrogen, 0.0625Μg/ml diluted with PBS) for 10 min at room temperature followed by PBS washes. Sections were dehydrated in ethanol (50% for 5 min, then 70% for 5 min) and treated with autofluorescence eliminator reagent (cat. no. 2160, Millipore Sigma) for 10 min at room temperature to reduce lipofuscin autofluorescence. After washing with ethanol (70% for 3 min, then 50% for 1 min), sections were mounted on glass slides, air-dried and coverslipped with prolong gold mountant (cat. no. P36930, Invitrogen).

Images were acquired on a LSM780 confocal microscope (Zeiss). Seven nonoverlapping fields at ×20 magnification were acquired per region (MCC-CB and dACC-CB: 28 images per hemisphere per animal, 7 fields × 4 sections; overlapped region of UF and FM: 20 images per hemisphere per animal, 5 fields × 4 sections). CC1-positive and DAPI-positive cells were counted in FIJI software⁷⁹ (version 1.54) using section-specific intensity thresholds for each channel (CC1 and DAPI). Automated counts were validated against manual counts. In addition to CC1-positive and DAPI-positive cell counting, the ratio of CC1/DAPI was calculated to normalize the number of oligodendrocytes to the total cell number.

The number of CC1-positive cells, the number of DAPI-positive cells and the CC1/DAPI ratio were compared between the stimulated and control hemispheres using one-way ANOVA for monkeys N and T. Monkeys were modeled as a fixed effect to account for between-subject variability but were not incorporated into the interpretation of the results. For monkey D (nonstimulation control), a paired t-test was used to compare between hemispheres. Statistical significance was set at P value < 0.05.

Electron microscopy analysis of myelin thickness

After 2–3 months of fixation in 2% paraformaldehyde and 2.5% glutaraldehyde in 0.1-M sodium cacodylate buffer, the mid-cingulate portion of the CB was dissected as tissue blocks from the coronal sections of monkeys N and T, adjacent to those used for immunofluorescence, and prepared for electron microscopy (Supplementary Methods 6). Using adjacent sections ensured that myelin measurements were performed in CB regions corresponding to those assessed for CC1 immunofluorescence.

To evaluate the myelin thickness, 30–80 nonoverlapping images were taken per hemisphere for each animal at ×4,000 magnification using an HT7500 transmission electron microscope (Hitachi High-Technologies). For each myelinated axon, four measurements were obtained: the longest and shortest axon diameters and the longest and shortest myelin widths. The g-ratio was calculated as the mean axon diameter divided by the mean fiber diameter (mean axon diameter + 2 × mean myelin width)^80,81. Regions with fixation artifacts or noncompaction were excluded. Obliquely sectioned axons were further excluded if the ratio of the longest to shortest axon diameter exceeded 2 (aspect ratio > 2), thereby focusing on the CB fibers running in the anterior–posterior direction and connecting SCC-DBS and PCC. On average, 175 myelinated axons were analyzed per hemisphere for each animal. g-ratio measurements were independently verified by a second rater blinded to condition.

The g-ratio in the stimulation side CB and the control-side CB were compared using one-way ANOVA. Monkeys were modeled as a fixed effect to control for between-subject variability, but this factor was not considered in the interpretation of the results. Statistical significance was set at P < 0.05.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41593-026-02301-4.

Author contributions

S.H.F., B.E.R., H.S.M. and P.H.R. designed the study. S.H.F., A.F. and P.H.R. conducted the DBS implantation surgeries. S.H.F. collected and analyzed the behavioral data. S.H.F., A.F., C.E., G.V., L.F., K.S.C. and B.E.R. established the MRI acquisition protocol. S.H.F., D.F., K.S.C. and B.E.R. made MRI preprocessing pipelines. S.H.F. and A.F. analyzed the imaging data. S.H.F., W.G.M.J. and P.H.R. extracted and processed the brains. S.H.F., E.A., A.S. and K.H. conducted and analyzed immunohistology assessments. S.H.F. and A.S. analyzed electron microscopy analysis. Funds were acquired by B.E.R., H.S.M. and P.H.R. This project was supervised by B.E.R., H.S.M. and P.H.R. S.H.F., B.E.R. and P.H.R. wrote the original draft. All authors edited the paper.

Peer review

Peer review information

Nature Neuroscience thanks Suzanne Haber and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Data availability

The functional MRI data generated in this study have been deposited in the DANDI database (10.48324/dandi.001773/0.260416.1939). Openly available resources used in this study, such as the NMT atlas, are accessible through the Primate Data and Resource Exchange (PRIME-DRE; https://prime-re.github.io/). Source data are provided with this paper.

Code availability

The custom MATLAB scripts used to analyze the functional MRI data are available in the DANDI database (10.48324/dandi.001773/0.260416.1939).

Competing interests

H.S.M. and K.S.C. receive consulting fees from Abbott Neuromodulation. S.H.F., A.F., C.E., A.S., E.A., K.H., G.V., W.G.M.J., L.F., D.F., B.E.R. and P.H.R. declare no competing interests. S.H.F. is also affiliated with the Department of Neurosurgery, Institute of Science Tokyo, Tokyo, Japan.

Extended data

is available for this paper at 10.1038/s41593-026-02301-4.

Supplementary information

The online version contains supplementary material available at 10.1038/s41593-026-02301-4.