Pregenual or subgenual anterior cingulate cortex as potential effective region for brain stimulation of depression.

Abstract Background The dorsolateral prefrontal cortex (DLPFC) is the standard stimulation target for the repetitive transcranial magnetic stimulation (rTMS) treatment of major depression disorder (MDD). A retrospective study by Fox and colleagues found that a more negative resting‐state functional magnetic resonance imaging (RS‐fMRI) functional connectivity (FC) between left DLPFC and the subgenual anterior cingulate cortex (sgACC) in a large group of healthy participants is associated with a better curative effects of rTMS in MDD, suggesting that the sgACC may be an effective region. However, a recent meta‐analysis on RS‐fMRI studies found that the pregenual ACC (pgACC), rather than the sgACC, of MDD patients showed increased local activity. Methods We used the stimulation coordinates in the left DLPFC analyzed by Fox et al. to perform RS‐fMRI FC between the stimulation targets obtained from previous rTMS MDD studies and the potential effective regions (sgACC and pgACC, respectively) on the RS‐fMRI data from 88 heathy participants. Results (a) Both the pgACC and the sgACC were negatively connected to the left DLPFC; (b) both FCs of sgACC‐DLPFC and pgACC‐DLPFC were more negative in responders than in nonresponders; and (c) the associations between DLPFC‐sgACC functional connectivity and clinical efficacy were clustered around the midline sgACC. Conclusions Both the pgACC and the sgACC may be potential effective regions for rTMS on the left DLPFC for treatment of MDD. However, individualized ACC‐DLPFC FC‐based rTMS on depression should be performed in the future to test the pgACC or the sgACC as effective regions.


| INTRODUC TI ON
Repetitive transcranial magnetic stimulation (rTMS) has been approved by the FDA (food and drug administration) for depression treatment. In past years, meta-analyses showed that application of high-frequency rTMS over the left dorsolateral prefrontal cortex (DLPFC) had antidepressant effect (Berlim, Frederique, & Daskalakis, 2013;Burt, Lisanby, & Sackeim, 2002;Kedzior & Reitz, 2014;Lesenskyj, Samples, Farmer, & Maxwell, 2018;Slotema, Blom, Hoek, & Sommer, 2010). The left DLPFC was selected as a stimulation target in light of a previous neuroimaging study, which showed decreased glucose metabolism in the left DLPFC in patients with MDD (Baxter et al., 1989). The general method, namely "5-cm rule," of locating DLPFC was established in 1995 (George et al., 1995). It consists of locating the hotspot in the left primary motor cortex first, then moving 5 cm anteriorly in the parasagittal plane, presumably targeting the left DLPFC. Although it is a convenient way to locate the DLPFC stimulation target, the "5-cm rule" does not account for individual variability of brain size and morphology, potentially resulting in studies finding no significant stimulation effects (Herbsman et al., 2009;Herwig, Padberg, Unger, Spitzer, & Schönfeldt-Lecuona, 2001).
Moreover, the locations of the hand motor hotspot are varied largely in population (Ahdab, Ayache, Brugieres, Farhat, & Lefaucheur, 2016), and it makes the target location defined by "5-cm rule" more heterogeneous. Investigators started to notice about the importance of precisely localizing of stimulation target (Battelli, Grossman, & Plow, 2017;Eldaief, Halko, Buckner, & Pascual-Leone, 2011;Wang et al., 2014). Fox, Buckner, White, Greicius, and Pascual-Leone (2012) proposed that the rTMS stimulation on the left DLPFC was related to a deep brain region named subgenual anterior cingulate cortex (sgACC). To test their hypothesis, they performed functional connectivity (FC) of the sgACC on a dataset of resting-state functional magnetic resonance imaging (RS-fMRI) from 98 healthy young adults. The authors found that the left DLPFC targets which were reported in previous studies (Fitzgerald et al., 2009;Herbsman et al., 2009;Paillere Martinot et al., 2010) with stronger negative FC of sgACC-DLPFC were associated with better efficacy (Fox et al., 2012). Based on these evidences, they concluded that the stimulation on the DLPFC may take antidepression effects through the sgACC-DLPFC network. Such remote effect on deep brain region via stimulating on superficial cortex has also been reported in rTMS studies (Arfeller et al., 2013;Lazzaro et al., 2011;Nahas et al., 2001;Solomon-Harris, Rafique, & Steeves, 2016). For simplicity, we hereafter called the superficial cortex target as "stimulation target," and correspondingly called the deep brain region as "effective region." Fox et al. (2012) listed a few evidences for taking the sgACC as an effective region for rTMS treatment of MDD. One evidence was that the regional cerebral blood flow (rCBF) in the sgACC decreased after TMS treatment (Kito, Fujita, & Koga, 2008;Kito, Hasegawa, & Koga, 2011). Another evidence showed that the sgACC was a successful target for deep brain stimulation (DBS) for depression treatment (Drevets, Savitz, & Trimble, 2008). However, some studies showed that the pregenual ACC (pgACC) is also a pivotal brain region for MDD (Ken-Ichi & Graybiel, 2012;Mannie et al., 2008). The pgACC region consistently showed elevated rCBF during the episode of major depressive disorder (Drevets, 2000), and significant increases of glucose and lactate are associated with depression severity (Ernst et al., 2016;Sacher et al., 2012). Apart from that, there are ample evidences elucidated that the increased pretreatment pgACC activity (the theta activity of electroencephalogram signal, the rCBF, the activation of fMRI signal during simple task and so forth) predicts better antidepressant response after kinds of treatment (TMS, medicine, sleep deprivation and so forth) (Pizzagalli, 2010). Another magnetic resonance spectroscopy (MRS) study found decreased glutamate and glutamine ratio in the pgACC in MDD patients (Horn et al., 2010).
More recently, a coordinate-based meta-analysis study on RS-fMRI found that the pgACC of patients with MDD had increased amplitude of low frequency fluctuation (ALFF) (Zhou et al., 2017). Therefore, based on the above multi-modal imaging studies, the pgACC might also be a potential effective region for treatment of MDD. To this end, the current study hypothesized that the FC values in the pgACC also have certain association with clinical improvement. We test this hypothesis by investigating the anticorrelation between the pgACC and the DLPFC in a large group of healthy participants and related the pgACC-DLPFC anticorrelation to the reported clinical efficacy of rTMS, similarly as did by Fox et al. (2012). The results would help us to understand the brain mechanism of rTMS treatment on MDD.

| Data composition
There were two datasets in the current study. Dataset 1 was RS-fMRI from 88 young healthy adults (Elaborated on below). Dataset 2 was some data from published studies, including: (a) the coordinates Note: To be noticed, the coordinates of more effective target which recorded by Fitzgerald (Fitzgerald et al., 2009) were located out of the brain cortical area, so we projected this coordinate to the nearest cortex.

| Participants
Eighty-eight healthy young adults (43 female, age = 23.2 ± 2.9) with no history of neurological or psychiatric disorders were recruited.
The present research was approved by the Ethics Committee of the Center for Cognition and Brain Disorders (CCBD) at Hangzhou Normal University. Written informed consent was signed by each subject before the experiment.

| Data preprocessing
RS-fMRI data were preprocessed using the Statistical Parametric

| Whole-brain voxel-wise FC of the sgACC and the pgACC
The seed ROIs were placed at the pgACC and the sgACC separately.
The sgACC coordinate (x = 6, y = 16, and z = −10) was selected from the study by Fox et al. (2012). The pgACC coordinate (x = 0, y = 42, and z = 6) was obtained from an RS-fMRI meta-analysis in which increased ALFF in the pgACC was reported in depressive patients (Zhou et al., 2017) (Table 1). Two kinds of radius (5 and 10 mm) for the ROIs were used in the current study. We used the 10-mm radius ROI in order to keep consistent with Fox et al. (2012). Considering that the sgACC and the pgACC are not far from each other anatomically, a 10-mm radius ROI could probably increase the similarity of their time courses.
We therefore also used a 5-mm radius ROI. A gray matter probability threshold of 0.25 was used on the Harverd/Oxford cortical template The average time course was extracted from each ROI. To generate FC map, Pearson's correlation coefficients were computed for each of the four ACC seed ROIs (the sgACC of 5-and 10-mm, and the pgACC of 5-and 10-mm, respectively) in a voxel-wise way through the whole brain. Fisher's r-to-z transformation  was applied for each correlation coefficient to fit the normal distribution.
After the calculation of FC, one sample t tests were performed on z-FC maps to reveal the FC pattern of ACC ROIs in the whole brain. The FDR (false discovery rate) correction (q < 0.001, cluster size > 100 voxels) was used to produce robust statistical maps. Fox et al. (2012) selected two previous studies of rTMS on MDD, each study reporting two DLPFC stimulation targets, that is, responder's target versus nonresponder's target (Herbsman et al., 2009) and more effective target versus less effective target (Fitzgerald et al., 2009), respectively, from previous MDD TMS treatment studies (  (2017) proposed a method to project the scalp TMS target to cortical surface and successfully applied by Wang et al. (2019)

| Correlation between ACC-DLPFC FC of 27 stimulation targets and clinical efficacy of 27 patients
Paillere We replicated the above process as did by Fox et al. in our data of 88 healthy participants, but we included both the sgACC and the pgACC ROIs with two kinds of radius (5 and 10 mm). As did by Fox et al., we centered the 10-mm DLPFC ROIs on 27 stimulation targets in Table 2. Using the Harverd/Oxford cortical template with an intensity of 0.7, the voxels lying outside of gray matter were eliminated (Fox et al., 2012). After that, the mean time course was extracted from each of the 27 10-mm DLPFC ROIs and the four ACCs (sgACC and pgACC, both with 5-and 10-mm radius).
Then, we computed ROI-wise FC between DLPFC targets and ACCs in our 88 healthy participants. All the FC values were averaged across the 88 participants, so every DLPFC target had one

| Correlation between ACC-DLPFC FC of nine DLPFC sites and corresponding predicted clinical efficacy
In Fox's study, the authors selected nine DLPFC sites from seven studies (Table 3) ). In addition, for the ACC, we utilized four ACC ROIs (sgACC and pgACC, both with 5-mm and with 10-mm radius).

| The voxel-wise FC between ACCs and whole brain
Both the pregenual and subgenual ACCs were negatively connected with the DLPFC (threshold q < 0.001, cluster size > 100 voxels) ( Figure 2).

| ACC-DLPFC FC comparisons
For the stimulation targets in Herbsman' study (Herbsman et al., 2009), ANOVA results revealed a significant main effect of target factor (F 1,87 = 22.496, p = 8.0 × 10 -6 ). After pairwise comparisons, we found both the sgACC and the pgACC showed significantly higher negative ACC-DLPFC FC for the responders' DLPFC target than the nonresponders' DLFPC target (Table S2 and Figure 3). Please see the Supporting Information for the details of interaction effect and following simple effect analyses ( Figure S2).

F I G U R E 2
The FC patterns of the sgACC and pgACC with different size of seeds (a, b) (FDR correction, q < 0.001, cluster size > 100 voxels, two-tailed  (Table S2). The follow-up pairwise comparison results showed that only the 10-mm radius sgACC-DLPFC FC had significant difference between more effective target and less effective target (F 1,87 = 5.032, p = .027) (Figure 4). The results of interaction effect and following simple effect analyses can be seen in Supporting Information (Figure S3, S4).

F I G U R E 3
The differences of ACC-DLPFC FCs between the responders and nonresponders. The DLPFC stimulation targets were from Herbsman et al. (2009). The colored regions in the brain indicate the ROIs of DLPFC (a). The definition of DLPFC was centered in the mean coordinate of responders and nonresponders with a 20-mm radius, respectively. ACC, anterior cingulate cortex; DLPFC, dorsal lateral prefrontal cortex; FC, functional connectivity; pgACC, pregenual ACC; SE, standard error; sgACC, subgenual ACC. *p < .05; **p < .01

| Correlation between ACC-DLPFC FC of nine DLPFC coordinates and corresponding estimated clinical efficacy
There were significant anticorrelations between sgACC-DLPFC FC (both in 5-mm and in 10-mm radius) and the corresponding esti-

| New analyses for sgACC-DLFPC FC
The laterality of sgACC-DLPFC FC seems to be a paradox: the sgACC coordinates (x = 6, y = 16, and z = −10) are in the right sgACC; however, the rTMS targets are in the left DLPFC. Furthermore, the radius of the sgACC is also a concern because, for Fitzgerald's pair of targets, the 10-mm radius FC showed significant differences between more and less rTMS efficacy, but the 5-mm did not (as shown in the Section 3.2). We supposed that the sgACC should be in the left side and that a 10-mm radius of the seed ROI may cover heterogeneous functional areas. We thus performed new analyses as follows.

F I G U R E 4
The differences of ACC-DLPFC FCs between the more effective target and less effective target. The DLPFC stimulation targets were from Fitzgerald et al. (2009) andFox et al. (2012). The colored regions in the brain indicate the ROIs of DLPFC (a). The definition of DLPFC was centered in the coordinate of more effective target and less effective target with a 20-mm radius, respectively. ACC, anterior cingulate cortex; DLPFC, dorsal lateral prefrontal cortex; eff., effective; FC, functional connectivity; pgACC, pregenual ACC; SE, standard error; sgACC, subgenual ACC. *p < .05   Table 3 As shown in Figure 7d, the overlapped voxels of the three statistical maps after thresholded at uncorrected p < .05 did not contain the right sgACC (the original sgACC coordinate: x = 6, y = 16, and z = −10 in Fox's study).

| D ISCUSS I ON
We systematically investigated the resting-state functional connectivity between the ACCs (both pgACC and sgACC) with the left DLFPC, that is, the stimulation target area for the rTMS treatment of MDD. While the analyses and general results were similar as those in the study by Fox et al. (2012), two new findings were found in the current study as discussed below.

| The pgACC may also be an effective region of rTMS
Albeit the results of many imaging studies support the sgACC as a critical region of MDD (Downey et al., 2016;Ho et al., 2014;Jaworska et al., 2014;Liu et al., 2015),

F I G U R E 7
The voxel-wise FC analyses between DLPFC and sgACC in a 20-mm radius medial sgACC ROI (x = 0, y = 16, and z = −10). The crosshair located in the original sgACC coordinate (x = 6, y = 16, and z = −10) from Fox et al. (2012). The colored regions in a and b represent the differences between two DLPFC TMS targets (better efficacy target vs. less efficacy target). The cold color indicates that better clinical efficacy target showed more negative FC (a and b). The cold color region in c represents negative correlation between DLPFC-sgACC FC and estimated clinical efficacy scores in nine DLPFC coordinates. All the statistical maps were thresholded at uncorrected p < .05. The red color in d represents overlapping brain region of a, b, and c. DLPFC, dorsal lateral prefrontal cortex; FC, functional connectivity; L, left; R, right; ROI, region of interest; sgACC, subgenual anterior cingulate cortex; TMS, transcranial magnetic stimulation Blumberger, Daskalakis, & Vila-Rodriguez, 2020;Liston et al., 2014;Weigand et al., 2018), there have a lot of evidences supporting the pgACC as a critical region of MDD (Ball, Stein, & Paulus, 2014;Boes et al., 2018;Pizzagalli, 2010;Silverstein et al., 2015;Zhou et al., 2017), and pgACC has been reported to show higher FC with the left lateral parietal cortex (IPL) at baseline in better clinical response group (Ge et al., 2020). In view of the above FC-based researches, the current study investigated the ACC-DLPFC FC as well as its association with rTMS efficacy. Similar as the sgACC, the pgACC also exhibited anticorrelation with the left DLFPC ( Figure 2). Furthermore, for Herbsman's pair of targets, the negative FC of ACC-DLPFC was stronger for responders' target than the nonresponders' (Figure 3).
These results indicate that, similar as the sgACC, the pgACC may be a potential effective region of rTMS, that is, the rTMS stimulation on the left DLPFC probably takes effect on the pgACC via DLPFC-pgACC FC.

| The midline sgACC, instead of the right sgACC: potential effective region of rTMS on the left DLFPC for MDD treatment
For the sgACC, although we generally replicated the results of Fox et al. (2012), there is an apparent paradox for the sgACC-DLPFC FC laterality: the sgACC was in the right side (x = 6, y = 16, and z = −10, the crosshair in Figure 7d) and the rTMS target was on the left DLPFC. As shown in the Section 3.5, we added a voxel-wise analysis of association of sgACC-DLPFC in the medial sgACC ROI centered at the coordinates (x = 0, y = 16, and z = −10, 20 mm radius) instead of (x = 6, y = 16, and z = −10). Results demonstrated that significant associations between FC and clinical efficacy were clustered around the midline sgACC ( Figure 7). However, the voxel in the right sgACC (the original sgACC coordinate: x = 6, y = 16, and z = −10 in Fox's study) fell outside of the overlapping area. It means that, although the mean time course of the 10-mm ROI centered at the right sgACC showed significant results (two t tests and one correlation analysis), the voxel per se did not. Instead, the voxels along the midline sgACC showed significant association between FC and clinical efficacy.
We did not repeat Section 2.2.6 in the voxel-wise way. The first reason was that neither the 10-mm radius ROI nor the 5-mm radius Martinot's study (Paillere Martinot et al., 2010). The correlation results became nonsignificant (r = −.210, p = .304) ( Figure S1). It means that the predictive function on clinical efficacy of the ACC-DLPFC FC in Fox's study might be largely driven by an outlier. Another reason was that unlike the nine DLPFC sites of which each value of coordinates was from averaged results of each study group, the 27 targets were from 27 individual patients. Hence, the 27 individuals may show larger variability.

| The implications for rTMS treatment on MDD
Although some studies support only sgACC or only pgACC as critical node of MDD, a few studies indicated that both the sgACC and the pgACC could be critical nodes for MDD (Drevets, 2000;Pizzagalli, 2010). The current results indicate that both the mid-

| CON CLUS ION
Either the midline sgACC (rather than the right sgACC) or the pgACC could be taken as effective region of rTMS on the left DLPFC for MDD treatment. The ACC-DLPFC resting-state functional connectivity can be considered to guide individualized precise localization of rTMS stimulation target on the left DLPFC in depression treatment.

| LI M ITATI O N S
One limitation is that the current study only analyzed the RS-fMRI data of the healthy subjects. It would be more reliable and more helpful on data from MDD patients, ideally, on the RS-fMRI data before and after rTMS treatment. Another limitation is that there has been lack of strong evidence for either the pgACC or the sgACC as a critical node. Future functional neuroimaging studies should focus on this topic and reveal individual abnormality in either the pgACC or sgACC.

ACK N OWLED G M ENTS
We thank all colleagues for their help on data collection. This research was supported by the Key Project of the Department of Science and Technology of Zhejiang Province (2015C03037) and the National Natural Science Foundation of China (Nos. 81701776, 31471084, and 81520108016).

CO N FLI C T O F I NTE R E S T S
All authors declare that they have no conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.