If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Department of Psychiatry, Helsinki University and Helsinki University Hospital, P.O. Box 590, FI-00029, HUS, Helsinki, FinlandDepartment of Neuroscience and Biomedical Engineering, and Advanced Magnetic Imaging Center, Aalto NeuroImaging, Aalto University School of Science, P.O Box 13000, FI-00076, AALTO, Espoo, Finland
Department of Psychiatry, Helsinki University and Helsinki University Hospital, P.O. Box 590, FI-00029, HUS, Helsinki, FinlandDepartment of Neuroscience and Biomedical Engineering, and Advanced Magnetic Imaging Center, Aalto NeuroImaging, Aalto University School of Science, P.O Box 13000, FI-00076, AALTO, Espoo, Finland
Department of Neuroscience and Biomedical Engineering, and Advanced Magnetic Imaging Center, Aalto NeuroImaging, Aalto University School of Science, P.O Box 13000, FI-00076, AALTO, Espoo, FinlandA.I. Virtanen Institute for Molecular Sciences, University of Eastern Finland, Kuopio, Finland
Department of Neuroscience and Biomedical Engineering, and Advanced Magnetic Imaging Center, Aalto NeuroImaging, Aalto University School of Science, P.O Box 13000, FI-00076, AALTO, Espoo, Finland
Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, USADepartment of Radiology, Harvard Medical School, USA
Core network model (CNM) to functionally target TMS in depression.
•
Strong face validity for emotion regulation in depression.
•
Clinically feasible accuracy and repeatability at the individual level.
•
Association with depression and treatment response support validity.
Abstract
Background
Transcranial magnetic stimulation (TMS) of the dorsolateral prefrontal cortex (DLPFC) is an established treatment for major depressive disorder (MDD). Recent attempts to improve TMS efficacy by individually targeting DLPFC subregions that are functionally connected to the subgenual cingulate cortex (sgACC) appear promising. However, sgACC covers only a small subset of core MDD-related areas. Further, fMRI connectivity of sgACC is poorly repeatable within subjects.
Methods
Based on an fMRI database analysis, we first constructed a novel core network model (CNM), capturing voxelwise emotion regulation- and MDD-related DLPFC connectivity. Then, in a sample of 15 healthy subjects and 29 MDD patients, we assessed (i) within-subject repeatability of the DLPFC connectivity patterns computed from time segments of varying lengths of individual-level fMRI data and (ii) association of MDD severity with the individual DLPFC connectivity strengths. We extracted group-level connectivity strengths in CNM from individual DLPFC coordinates stimulated with neuronavigated TMS in a separate sample of 25 MDD patients. These connectivity strengths were then correlated with individual TMS efficacy.
Results
Compared with sgACC connectivity, CNM increased intraindividual repeatability 5-fold. DLPFC connectivity strength from CNM was associated with MDD severity and TMS efficacy. While the locations of CNM-based individual TMS targets remained constant within individuals, they varied considerably between individuals.
Conclusions
CNM increased repeatability of functional targeting to a clinically feasible level. The observed association of MDD severity and TMS efficacy with DLPFC connectivity supports the validity of the CNM. The interindividual differences in target locations motivate future individualized clinical trials leveraging the CNM.
Transcranial magnetic stimulation (TMS) is a promising treatment in several brain disorders, including major depressive disorder (MDD), which is among the leading causes of disability worldwide [
GBD Global, regional, and national incidence, prevalence, and years lived with disability for 354 diseases and injuries for 195 countries and territories, 1990-2017: a systematic analysis for the Global Burden of Disease Study 2017.
]. In MDD patients failing to respond to first-line treatments, TMS of the dorsolateral prefrontal cortex (DLPFC) is an established and well-tolerated treatment [
Response, remission and drop-out rates following high-frequency repetitive transcranial magnetic stimulation (rTMS) for treating major depression: a systematic review and meta-analysis of randomized, double-blind and sham-controlled trials.
Unilateral and bilateral repetitive transcranial magnetic stimulation for treatment-resistant depression: a meta-analysis of randomized controlled trials over 2 decades.
Anatomical and fMRI-network comparison of multiple DLPFC targeting strategies for repetitive transcranial magnetic stimulation treatment of depression.
Subgenual functional connectivity predicts antidepressant treatment response to transcranial magnetic stimulation: independent validation and evaluation of personalization.
]. The overall conclusion of these studies has been that the clinical outcomes correlate with functional connectivity strength between the individual TMS target and a seed region [
Subgenual functional connectivity predicts antidepressant treatment response to transcranial magnetic stimulation: independent validation and evaluation of personalization.
]. Typically, this seed (i.e., a region to compare fMRI time course with those from other locations in the brain) has been the subgenual cingulate cortex (sgACC), as the sgACC has been associated with outcomes of several MDD treatments [
]. The fMRI signal-to-noise ratio is, however, poor in the sgACC regions due to susceptibility artifacts, resulting in poor reproducibility of sgACC connectivity-based targets at the individual level [
]. In these studies, targeting was, however, combined with a very intense TMS protocol and not compared with scalp-based targeting. To support such comparisons, feasibility of techniques for functional individualized targeting needs to be ascertained, including increasing validity for functioning of the core networks underlying MDD (referred to as network validity) and enhancing repeatability of individualized target localization.
Many alternatives exist to map treatment targets. While data-driven approaches help to avoid some potential errors, results have been poorly reproducible with currently available samples sizes [
]. One option to restrict problems related to multiple comparisons is to focus analyses based on clinical features. For example, while MDD is associated with many symptoms, its two key features are low mood and inability to feel pleasure [
]; other symptoms can vary, but without one of these features MDD cannot be diagnosed. Mood is a prevalent emotional state that is coupled to more fleeting emotions, linking the two fundamental symptoms with emotion regulation. In line, accumulating evidence points to dysfunctional emotion regulation strategies in MDD [
]. At the neuronal level, the cingulo-opercular network, which is structurally and functionally altered in MDD and several other common psychiatric disorders [
], but according to current views, any single brain region or connection could cover only a small subset of the networks involved in emotion regulation in MDD [
]. Fox and coworkers suggested expanding the seed region to cover the entire brain, weighting each seed voxel in computation of connectivity with the voxel's connectivity with i) the sgACC or ii) effective TMS sites [
]. It remains, however, unknown whether such sgACC connectivity-based approach is optimal, and the field may benefit of consideration of network models considering other approaches as well.
Meanwhile, knowledge about the brain correlates of MDD and emotion regulation has increased, enabling construction of models with stronger network validity. MDD-related networks overlap widely with emotion regulation networks, including i) emotion-processing regions, such as the amygdala and the sgACC, and ii) regions that regulate the emotion-processing regions, such as the DLPFC and the cingulo-opercular network [
], which can be used to improve target network validity. Further, delivering stimuli that engage emotion regulation instead of resting-state during fMRI might increase validity for emotion regulation. Emotional stories are an engaging naturalistic stimulus [
], their validity for the targeted disorder and TMS efficacy remains unknown. Accurate and repeatable targeting methods with enhanced network model validity would be necessary not only to define target locations, but also to estimate whether target locations vary between individuals to an extent that TMS targeting needs to be individualized [
Antidepressant efficacy of prolonged intermittent theta burst stimulation monotherapy for recurrent depression and comparison of methods for coil positioning: a randomized, double-blind, sham-controlled study.
Here, we aimed to increase network validity by constructing a novel CNM to capture MDD- and emotion regulation-related DLPDC connectivity. We hypothesized that this approach increases within-subject repeatability of fMRI connectivity-based DLPFC targets to a clinically feasible level, and that the resulting DLPFC connectivity reflects MDD severity and TMS outcome. Moreover, we hypothesized that presenting emotionally engaging audio stories during fMRI, instead of recording resting-state fMRI data without stimuli, would enhance the model performance. Finally, as repeatability depends on scan time, we assessed the effect of scan length on repeatability, with the overall aim of developing a feasible individualized functional targeting method for clinical use.
2. Materials and methods
2.1 Participants
Data from three separate studies were used for evaluating the network model. The first of these (“MDDAD”) included 29 MDD patients randomized to receive 10 mg of escitalopram (n = 15) or placebo (n = 14) daily for one week prior to imaging to evaluate early medication-related brain functions [
]. This network model was constructed a priori, before testing it with the individual level data. To derive seed regions associated with MDD and emotion regulation, we constructed two probabilistic maps of voxels, the first one associated with the keywords “major depression” and the second associated with “emotion regulation”. Next, we multiplied these two maps, resulting in regions associated both with MDD and emotion regulation. As brain imaging studies on emotion regulation typically evoke emotions, the emotion regulation-related regions include both regulatory and regulated regions. As a regulatory region, the DLPFC has positive connectivity with other regulatory regions, such as the cingulo-opercular network, and negative connectivity with the regulated regions (especially in standardized connectivity maps), such as the amygdala and the sgACC. If connectivity with the DLPFC was summed from the regulatory and regulated regions, connectivities of these regions might cancel each other out. We thus differentiated the negatively and positively connected regions based on group-level functional connectivity with DLPFC in Neurosynth resting-state fMRI connectivity data of 1000 individuals. The DLPFC was defined here as the largest clusters of voxels in Neurosynth associated with the keyword “DLPFC”, multiplied with the Harvard-Oxford cortical (binary) mask. Both the left and the right DLPFC were included because both are used as TMS targets. We then selected as seeds the largest clusters of voxels associated with MDD and emotion regulation that showed positive or negative DLPFC connectivity (maximum correlation >0.1 or < –0.1 within the DLPFC mask) and agreed with the literature on emotion regulation and MDD [
]. The resulting clusters were located bilaterally within the amygdala and the cingulo-opercular network. In addition, even though sgACC, as defined by Fox et al. [
], did not meet the inclusion criteria for network nodes (existing association with “depression” but absent association with “emotion regulation”), this area was added post hoc to build a network that is more comprehensive for TMS targeting in MDD. Specifically, the rationale for including sgACC was several previous studies linking it to emotion regulation [
]. Due to potential inter-hemispheric differences in brain functioning, seeds crossing the midsagittal plane were separated to left and right hemisphere parts, resulting in a total of eight seeds in non-DLPFC regions (Fig. 1 and Table 1). Supplementary Figure 1A shows differences together with considerable overlap of MDD-related and emotion-regulation related regions in the Neurosynth data.
Fig. 1Title: Seed regions of CNM and their DLPFC connectivity. Caption: Panel A shows seed region connected positively (blue) and negatively (red) with the DLPFC. Green in panel B presents sgACC connectivity and blue CNM connectivity (overlap in turquoise). Crosshair points to a previously suggested connectivity-based group-level TMS target in the left DLPFC (x, y, z = −38, 44, 26) [
Z-scores indicate the association strengths with keywords “major depression” and “emotion regulation”, and r values the connectivity strengths of each seed with DLPFC (strongest connection is shown independently of laterality) in Neurosynth database. Coordinates refer to the voxel with the strongest association with the keywords or the strongest (negative or positive) connectivity with a seed region.
] to compare CNM connectivity with clinical outcomes. Eighteen patients received standard 10-Hz stimulation and 7 patients received 20-Hz stimulation. Eleven patients were treated with the Magstim® and 14 with the Neuronetics® TMS system, and the number of treatment sessions ranged from 20 to 32. MRIs were acquired before treatment and coil position was registered with the Brainsight® Neuronavigation system (see original publication for further details).
2.4 Imaging and stimulus
We had imaging data from the MDDAD and HC samples (but only stimulation coordinates for the MDDTMS sample). The functional blood-oxygenation-level-dependent (BOLD) and T1-weighted MRIs were acquired at 3T (Supplementary Material) in our earlier studies [
]. In MDDAD and HC each subject's fMRI data with “stories” consisted of time series of 1000 full volume acquisitions that had been recorded while presenting a total of 30 audio stories. Neutral stories and stories with negative and positive emotional content, ten of each three categories, and 45 s of duration each, were presented with the Presentation® software (Neurobehavioral Systems Inc., Albany, CA, USA). Valence of the stories was confirmed with subjective ratings in an earlier study [
]. The resting-state fMRI data were available from the MDDAD patients only (not from the HC dataset) and consisted of 250 full-volume acquisitions without stimuli. The data were acquired during the same MRI session before obtaining the data with “stories”. For further details, see Supplementary Material and references [
]. The MRI data from MDDAD and HC were also used to derive estimates connectivity strengths in coordinates actually stimulated in the separate MDDTMS sample [
] and SPM12 (https://www.fil.ion.ucl.ac.uk/spm/software/spm12/). We used movement correction and spatial normalization of fMRIs into the Montreal Neurologic Institute template based on structural T1 images with a fast diffeomorphic image registration algorithm (DARTEL) [
]. We did not use spatial smoothing, as it increases repeatability at the cost of spatial accuracy. We regressed out movement parameters and their exponentials, CSF and white matter signals, bandpass-filtered signal time series at 0.01–0.1 Hz, and removed 1 previous and 2 later volumes at time points with framewise displacement (FD_Power) > 0.5 mm to reduce movement artifacts.
2.6 CNM to define targets at the individual level
We computed connectivity of the seed regions defined in 2.2 (see also Fig. 1). as Pearson's correlation between the seed fMRI time series and those in other voxels in the whole brain, but focus on the connectivity of the seed regions with the DLPFC due to its established role in TMS treatment of MDD. In the model, the DLPFC and the cingulo-opercular regions relate to downregulation of the negatively connected regions (sgACC and amygdala) that are involved in eliciting and experiencing emotional responses [
]. In such a network, a regulatory region may activate together with other regulatory regions (reflected in positive connectivity between regulatory regions) to down regulate activity of a regulated region (reflected in negative connectivity between the regulatory regions and the regulated region). In this network, the connectivities of regulatory and regulated regions with a target region are thus opposite (i.e., have different sign). Combining connectivities with different signs would cancel each other out and result in biased targeting. Thus, our model combines sign-reversed negative connectivity of the sgACC and amygdalae and positive connectivity of the cingulo-opercular seeds. This was operationalized by multiplying the individual-level sgACC and the amygdala connectivity maps (images of voxelwise connectivity) by −1 (note that absolute values cannot be used here as this would mix positive and negative connectivity). Then the eight connectivity maps were averaged to one target map for each individual, reflecting voxelwise strength of emotion regulation-related DLPFC connectivity in the MDD-related networks. These maps were computed separately for resting state and story stimulus data. Supplementary Figure 1B shows both differences and overlap in connectivity of the MDD-related and emotion-regulation related regions.
2.7 Assessing within-subject repeatability of target maps
Individual target maps, within-subject spatial correlations between the target maps, and connectivity strengths were computed from the MDDAD and HC datasets in Python using standard libraries (e.g., pandas [
]. We split the fMRI time series to segments of equal length and compared target maps derived from the separate segments within individuals. To assess the dependency of repeatability on scan length, we used data segments of different lengths. Specifically, the 1000 volumes collected during stories were split into eight segments of 125 images, 4 segments of 250 images, and 2 segments of 500 images. The 250 volumes collected during rest were split to two segments of 125 images. Next, we computed the target maps from each segment separately for each participant. Then, we computed intraindividual spatial correlation (rspat) across all the individual maps computed from the same length of scan (e.g., an average of spatial correlation between all pairs of the maps computed from four fMRI data segments of 250 volumes each). We focused on the DLPFC target regions used in TMS treatment in MDD, computing rspat within the bilateral DLPFC mask. We used Fisher Z-transformation before two-tailed statistical tests on correlation values.
As the relationship between scan time and repeatability of connectivity follows an exponential function [
]. Here, e and a are constants, rspatmax is the maximum achievable rspat with the present data, and t is the scan time.
2.8 Comparing repeatability between models and stimulus type
For fMRI data recorded with stories in the MDDAD and HC datasets, we computed rspat for 7 min (250 volumes) and 14 min (500 volumes) using sgACC only vs. the entire CNM. We then compared rspat between the two models with equal scan durations (250 vs. 250 volumes and 500 vs. 500 volumes). For resting-state fMRI data, available in MDDAD, we computed rspat using the CNM for a data segment of 3.5 min (125 images as longer repeated segments were not available) and then compared it with rspat for equally long data segments recorded with stories.
2.9 Comparing target locations across individuals
Using the MDDAD and HC data, we selected 100 DLPFC voxels with the largest connectivity values in the CNM for each subject for each hemisphere (referred to as targets) computed individually (see section 2.6) from 1000 functional volumes during stories. The resulting maps were binarized, summed together, divided by the number of subjects, and multiplied by 100, resulting in a percentage map of common target voxels between individuals. We also computed the Euclidean distance between the most strongly connected voxels across individuals in the left DLPFC.
2.10 Comparing target connectivity strengths between stimuli and groups
To test the validity of the model in independent samples, the connectivity strengths of individual targets (computed as average connectivity across the 100 target voxels and 1000 fMRI volumes, referred to as target connectivity strengths) were compared across individuals between MDDAD and HC and correlated with severity of MDD in the MDDAD sample. Here, target connectivity values were pooled across the left and right hemispheres because they were strongly intercorrelated (r = 0.91, P = 3*10−17). MDD severity outcomes comprised of Montgomery-Åsberg Rating Scale (MADRS) [
]. These coordinates were defined as the nearest cortical location from the coil center during treatment. CNM connectivity values for the coordinates were derived for each individual in the MDDTMS group from the average across the individual CNM connectivity maps of the MDDAD patients. We then correlated the group level CNM connectivity strengths from individual stimulation locations with the TMS-related BDI % change across individual MDDTMS patients following Fox and coworkers [
]. These correlations were adjusted for device, stimulation protocol, and number of treatment sessions. For comparison, we correlated the TMS outcome with the sgACC, emotion-regulation, and MDD-related connectivity separately.
] did not differ significantly in sex, but differed in age. Thus, age was adjusted for when comparing these groups. In the MDDAD group, BDI score was 25 ± 8. In MDDTMS data [
], BDI score was 39 ± 9 before and 21 ± 13 after treatment.
3.2 Comparison of within-subject repeatability of the target maps between CNM and sgACC connectivity
These analyses focused on the DLPFC. We first compared repeatability using within-subject spatial correlation (rspat) of the target maps computed from juxtaposed 7-minute fMRI-data segments during stories. As the difference in rspat between HC vs MDDAD was non-significant (P > 0.22), the analysis of repeatability for stories (available for both groups) was conducted in pooled groups in addition to MDDAD group. Compared with sgACC connectivity alone (rspat = 0.27 ± 0.14 and 0.26 ± 0.14, for pooled and MDD groups, respectively), the CNM resulted in a clearly higher rspat (0.58 ± 0.14 and 0.59 ± 0.12, P = 6.1 × 10−14, and 8.2 × 10−12 for the pooled and MDDAD group, respectively, paired samples t-test). Variance of a target map constructed from one data segment explained by those constructed from other data segments (i.e., rspat2) was 4.5 and 5.1fold higher (for the pooled and MDDAD group, respectively) when computed between consequent story segments and 5.5-fold higher when computed between the story and the resting-state segments of equal length (MDDAD group). Examples of individual targets with the two models are presented in Fig. 2A and B, and Fig. 2C shows rspat dependence on scan time.
Fig. 2Title: Intraindividual repeatability of target maps for CNM and sgACC connectivity. Caption: Panels A and B show intraindividual repeatability of target maps for a representative individual for CNM and sgACC connectivity, respectively. Green shows target maps from one 14-minute fMRI data segment and blue from another 14-minute fMRI data segment in the same individual during stories. Panel C presents dependency of repeatability on scan time. Error bars present ±1 SD for CNM (upper bars) and for sgACC connectivity (lower bars) at 7 and 14 min of fMRI data acquisition during stories. An exponential function was fitted to the CNM data to evaluate repeatability at different scan lengths. CNM, core network model; sgACC, subgenual anterior cingulate cortex.
3.3 Within-subject repeatability of the target maps depending on scan time and stimulus type
Within the pooled sample of MDDAD and HC, with stories as stimuli, the spatial correlation of the CNM target maps increased from rspat = 0.58 ± 0.14 and 0.59 ± 0.12 at 7 min to rspat = 0.75 ± 0.12 and 0.76 ± 0.08 for the pooled and MDDAD group, respectively, at 14 min segments. For resting-state fMRI in MDDAD sample, the spatial correlation of target maps from 7 min story segments was 0.49 ± 0.11. For story stimulus, the best fit for the dependency of rspat on scan time was achieved with an exponential function with maximum achievable group average of rspat = 0.83 during stories.
The stories tended to result in higher rspat than rest (rspat = 0.36 ± 0.11 vs. 0.31 ± 0.12 for 3.5 min that were the longest available repeated segments for resting-state data, P = 0.06). Mean framewise displacement was similar during the stories (0.09 ± 0.04 mm) and rest (0.08 ± 0.03 mm; P = 0.52).
3.4 Interindividual differences between targets
Group average of the target maps from the CNM during stories in the MDDAD sample matched well with an earlier estimate of the most efficient stimulation site in the left hemisphere (x, y, z = −38, 44, 26) [
]. However, this group-level result did not represent the individual-level maxima well because this voxel was among the 100 strongest individual DLPFC connections in the CNM in only 28% of the patients. The most strongly connected voxels tended to cluster within an individual, and the locations and shapes of these clusters differed clearly between individuals (Fig. 3). The strongest overlap was located at x, y, z = 36, 45, 24, which was among the top 100 individual target voxels in 41% of the patients (Fig. 3). The between-subject Euclidean distance between the most strongly connected voxels was 18 ± 4 mm (mean ± SD) in the left hemisphere.
Fig. 3Title: Intraindividual vs interindividual variance of the CNM target locations. Caption: In panel A, red/yellow show target maps computed from two 14-minute segments of fMRI data in one individual (overlap in orange) and blue/green in another individual (overlap in turquoise). Panel B shows CNM targets in eight randomly selected subjects on a sagittal slice. Different colors represent different subjects computed from 28 min of individual-level fMRI data. Panel C shows percentual intersubject overlap of targets across all 29 MDDAD patients during stories. Individual targets consist of 100 voxels for each hemisphere. The crosshair points to the connectivity-based group-level target suggested by Fox and coworkers [
3.5 Comparison of CNM target connectivity strengths between stimuli and groups
Target connectivity strengths from the CNM in the MDDAD group were stronger while listening to the stories than during rest (r = 0.27 ± 0.04 vs. 0.23 ± 0.05, P = 0.002; paired samples t-test; note that connectivity strengths do not depend on scan time). Fig. 4A shows that the patients (MDDAD) had stronger target connectivity than the healthy controls (HC) (r = 0.27 ± 0.05 vs. 0.24 ± 0.05) during stories (P = 0.04). There was no difference between the medicated and unmedicated patients in connectivity strengths during stories or rest (P > 0.91).
Fig. 4Title: Relationship of the core network model target connectivity strengths with depression and effect of TMS. Caption: Panel A shows difference between patients (pt) and healthy control subjects (hc) in the target connectivity strengths during stories. Panels B and C present correlation of the connectivity strengths during rest with depression severity and TMS response, respectively. BDI, Beck's Depression Inventory; BDI change %, percentual decrease in BDI during TMS treatment. Error bars represent ±1 SD.
3.6 Correlation of target connectivity strengths with MDD severity
Fig. 4B shows that during rest the target connectivity strength correlated with depression severity (r = 0.40, P = 0.03) as assessed with BDI. No such correlation was observed during stories or with MADRS (P> 0.05). Subgenual ACC connectivity strength did not differ between patients and control subjects; it also did not correlate with symptom severity during stories or rest (p > 0.25).
3.7 Correlation of target connectivity strengths with clinical response to TMS
Fig. 4C shows that the MDDAD group's average connectivity strengths from CNM, extracted from the coordinates of the MDDTMS patients targeted during TMS treatment, correlated (r = 0.44, P = 0.03 during rest; r = 0.48, P = 0.02 during stories) with the treatment outcomes, as measured with BDI. These correlations survived adjusting for device, stimulation protocol, and number of treatment sessions (partial correlation P < 0.05). Correlations of TMS outcome with connectivity of emotion regulation and MDD-related regions, computed separately, were r = 0.32, P = 0.12; r = 0.37, P = 0.07 respectively, during stories. No correlation with TMS outcome was observed with sgACC connectivity alone (P > 0.3 during stories and rest).
4. Discussion
We constructed a novel network model with strong face validity for MDD-related core networks and showed the resulting targets to be repeatable within individuals and valid for MDD and the TMS treatment response in MDD in independent samples.
The finding that the cingulo-opercular regions and the DLPFC were negatively connected with the sgACC and amygdala is not new. The cingulo-opercular regions and the DLPFC are, however, large and functionally heterogeneous regions. The CNM seeds were localized based on a large database to the cingulo-opercular subregions most strongly associated with MDD and emotion regulation. Thus, connectivity of these subregions may help to advance localization of TMS-targets for MDD.
The CNM, based on association with the keywords “major depression” and “emotion regulation” as well as DLPFC connectivity in the Neurosynth database, agrees well with the previous literature. MDD relates to an imbalance between the cortical (regulatory) and the deep emotion-related (regulated) systems centered at the amygdala and the sgACC [
]. Furthermore, recent findings suggest that the most important cortical emotion regulation circuitries include the anterior insula and the dorsal anterior cingulate cortex [
]. These cingulo-opercular regions have direct structural connections with the DLPFC, amygdala, and sgACC, being well-positioned to regulate the deep emotion-related systems and to mediate regulatory signals from the DLPFC [
Emotion regulation encompasses explicit and implicit processes that identify the need to regulate emotions, select among available regulatory strategies, implement the selected strategy, and monitor the outcome [
]. The present model may benefit further studies aimed at testing the long-held hypothesis that TMS effects in MDD are mediated by modulation of the emotion regulation pathways [
] and linked in causal models to treatment-related changes in symptoms and brain measures.
Patients had stronger target connectivity in the CNM than control subjects, and the target connectivity strengths were positively correlated with the BDI scores. This agrees with the previous literature suggesting that during automatic emotion regulation situations (such as when presented with emotional stimuli without an explicit emotion regulation task) patients with MDD tend to engage DLPFC more strongly than healthy control subjects, which may reflect compensatory mechanisms [
Subgenual functional connectivity predicts antidepressant treatment response to transcranial magnetic stimulation: independent validation and evaluation of personalization.
] our results suggest that functionally defined TMS targets differ considerably between MDD patients. The observed 18-mm mean difference between targets across individuals was larger than the error of a few millimeters resulting from modern non-linear spatial normalization to a common brain template [
Evaluating the validity of volume-based and surface-based brain image registration for developmental cognitive neuroscience studies in children 4 to 11 years of age.
]. Meanwhile, intraindividual repeatability was high, especially considering that no spatial smoothing was used. While smoothing of fMRI data increases apparent repeatability [
], it reduces spatial accuracy. Importance of the spatial accuracy of the target maps depends on the strength and spatial distribution of the TMS-induced electric fields needed to trigger therapeutic mechanisms, although such measures remain inadequately known [
How does repetitive transcranial magnetic stimulation influence the brain in depressive disorders?: a review of neuroimaging magnetic resonance imaging studies.
]. However, induction of individual finger movements in humans as well as single-cell recordings in non-human primates suggest that TMS-induced action potentials can be restricted to a few square millimeters of the cortex [
]. If the presently observed spatial accuracy with non-smoothed data was unnecessary for some applications, smoothing of the data would increase repeatability considerably beyond the present results [
The present results also suggest that employing emotional story stimuli during fMRI, as opposed to resting-state paradigms, may increase connectivity of the target maps and reduce scan time demands. On the other hand, the relatively high spatial correlation between story and resting-state target maps implies that resting-state data may be sufficient, provided that the scan time is adequate. As auditory stimuli such as stories can be easily presented in the MRI-scanner, we recommend further studies to consider the potential advantagef of using auditory narratives.
Limitations of this study include relatively small samples, limited amounts of resting-state data and the lack of individual-level brain imaging data of MDDTMS patients. This may explain the mixed findings on correlation between the CNM target connectivity strengths and the measures of MDD severity, as well as the trend-level difference in repeatability between the rest and story stimulus. While repeatability of connectivity from 3.5 min fMRI segments is generally poor, this depends on the analysis. The fitted function in Fig. 4C suggests that repeatability of the CNM-based targets is relatively good already at 3.5 min fMRI data. Notably, we used these short data segments only to compare repeatability and connectivity strengths of the CNM-based targets between rest and story stimulus, while other analyses used longer data segments.
Better understanding of the subprocesses of emotion regulation would enhance development of these lines of research. Even if the fundamental symptoms of MDD relate to emotion regulation [
], it is possible that at least some patient groups may benefit from separate targeting for different symptoms or symptom clusters such as anxiety and somatic symptoms [
]. Although we focused on the DLPFC, future research should consider potential targets in other regions as well. An alternative way to map seeds or targets related to MDD and emotion regulation would be to directly assess regions of altered emotion regulation in MDD patients. We selected the present way as a large amount of imaging data is considered necessary for reliable localization, and the available amount of data on MDD and emotion regulation was considerable larger than that on emotion regulation in MDD patients. Even if brain correlates of psychiatric disorders may be better explained as functions of networks than of a single region, future clinical trials are needed to show the difference in treatment outcome between different single and multi-seed models. Prospective randomized controlled trials are also needed to exclude the possibility that such targeting models would relate to treatment resistance rather than to clinically useful localization.
The main strength of the study is its innovative use of individualized neural network connectivity patterns to identify targets for treatment of MDD. To our knowledge, there are no previous studies presenting a network targeting model with strong face validity for the fundamental emotion regulation-related core symptoms of depression, supported by association between the target connectivity and both the MDD symptoms scores and TMS outcome. Furthermore, our findings add to the literature showing that such a model associates with good repeatability and spatial accuracy, which may be further enhanced using a naturalistic stimulus during imaging.In conclusion, these findings suggest that repeatable and accurate individual targets can be computed based on connectivity of the emotion regulation-related core neuronal networks of MDD. The target connectivity strengths differed between MDD patients and healthy control subjects and correlated with MDD severity and TMS outcome in MDD in separate samples, supporting model validity. The distance between repeatable individual targets was considerable. Such findings build the foundation for future clinical trials to test whether and to what extent individual functional targeting enhances outcomes.
Funding sources
This work was financially supported by the Academy of Finland (grant number 250935 to JE, 348631 to DBA, and 315861 to TTR) and the U.S. National Institutes of Health (NIH) R01-NS126337 (TR). The Academy of Finland or NIH had no further role in the study.
CRediT authorship contribution statement
Tuukka T. Raij: Conceptualization, Methodology, Formal analysis, Writing – original draft. Emma Komulainen: Resources, Writing – review & editing. Dogu Baran Aydogan: Investigation, Writing – review & editing. Siina Pamilo: Software, Writing – review & editing. Erkki Isometsä: Resources, Writing – review & editing. Tommi Raij: Supervision, Writing – review & editing.
Declaration of competing interest
TThe authors declare the following financial interests/personal relationships which may be considered as potential competing interests: TTR: patent application related to the present methods, lecture fee from Lundbeck inc. Other authors report no conflict of interest.
Acknowledgments
We thank Ana Triana Hoyos, Matti Holi, Risto Ilmoniemi, Matti Stenroos, Marita Kattelus, and Juho Ylä-Jääski for expert help and advice, and Dr. Jesper Ekelund for providing the fMRI data collected for a different study.
Appendix A. Supplementary data
The following is the Supplementary data to this article.
Global, regional, and national incidence, prevalence, and years lived with disability for 354 diseases and injuries for 195 countries and territories, 1990-2017: a systematic analysis for the Global Burden of Disease Study 2017.
Response, remission and drop-out rates following high-frequency repetitive transcranial magnetic stimulation (rTMS) for treating major depression: a systematic review and meta-analysis of randomized, double-blind and sham-controlled trials.
Unilateral and bilateral repetitive transcranial magnetic stimulation for treatment-resistant depression: a meta-analysis of randomized controlled trials over 2 decades.
Anatomical and fMRI-network comparison of multiple DLPFC targeting strategies for repetitive transcranial magnetic stimulation treatment of depression.
Subgenual functional connectivity predicts antidepressant treatment response to transcranial magnetic stimulation: independent validation and evaluation of personalization.
Antidepressant efficacy of prolonged intermittent theta burst stimulation monotherapy for recurrent depression and comparison of methods for coil positioning: a randomized, double-blind, sham-controlled study.
Evaluating the validity of volume-based and surface-based brain image registration for developmental cognitive neuroscience studies in children 4 to 11 years of age.
How does repetitive transcranial magnetic stimulation influence the brain in depressive disorders?: a review of neuroimaging magnetic resonance imaging studies.
01705-9&id=gr1.jpg){kind=link}
01705-9&id=gr2.jpg){kind=link}
01705-9&id=gr3.jpg){kind=link}
01705-9&id=gr4.jpg){kind=link}