A comparative study of the dorsolateral prefrontal cortex targeting approaches for transcranial magnetic stimulation treatment: Insights from the healthy control data

Repetitive transcranial magnetic stimulation (rTMS) to the left dorsolateral prefrontal cortex (DLPFC) is an established treatment for medication-resistant depression. Several targeting methods for the left DLPFC have been proposed including identification with resting-state functional magnetic resonance imaging (rs-fMRI) neuronavigation, stimulus coordinates based on structural MRI, or electroencephalography (EEG) F3 site by Beam F3 method. To date, neuroanatomical and neurofunctional differences among those approaches have not been investigated on healthy subjects, which are structurally and functionally unaffected by psychiatric disorders. This study aimed to compare the mean location, its dispersion, and its functional connectivity with the subgenual cingulate cortex (SGC), which is known to be associated with the therapeutic outcome in depression, of various approaches to target the DLPFC in healthy subjects. Fifty-seven healthy subjects underwent MRI scans to identify the stimulation site based on their resting-state functional connectivity and were measured their head size for targeting with Beam F3 method. In addition, we included two fixed stimulus coordinates over the DLPFC in the analysis, as recommended in previous studies. From the results, the rs-fMRI method had, as expected, more dispersed target sites across subjects and the greatest anticorrelation with the SGC, reflecting the known fact that personalized neuronavigation yields the greatest antidepressant effect. In contrast, the targets located by the other methods were relatively close together with less dispersion, and did not differ in anticorrelation with the SGC, implying their limitation of the therapeutic efficacy and possible interchangeability of them.


Introduction
Repetitive transcranial magnetic stimulation (rTMS) is a noninvasive technique that modulates cortical activity by short, rapidly changing magnetic field pulses that induce electrical currents in the cerebral cortex (Fox, Buckner, White, Greicius, & Pascual-Leone, 2012;Hallett, 2007).It has gradually been proven to be effective for some disorders or states, such as major depressive disorder (MDD) including treatment-resistant depression, bipolar disorder, obsessive-compulsive disorder, and chronic pain (Rizvi and Khan, 2019;Nguyen et al., 2021;Di Ponzio et al., 2022;Goudra et al., 2017;Noda et al., 2023).For many psychiatric disorders such as the ones we have mentioned, the suggested stimulation site is a part of the brain called dorsolateral prefrontal cortex (DLPFC) which is known to have the role in cognitive control, and is dysregulated in patients with those disorders.(Weissman et al., 2008;Janicak and Dokucu, 2015;Townsend et al., 2010;Piras et al., 2015).
To date, research has suggested various approaches for determining the stimulation target.Evidence has gradually emerged in recent years that targeting the stimulation site based on the altered functional connectivity in patients with MDD would yield the better result (Klooster et al., 2023).The anticorrelation of functional connectivity between the DLPFC stimulation site and subgenual cingulate cortex (SGC) determined by resting-state functional magnetic resonance imaging (rs-fMRI) is considered to be associated with better therapeutic outcome of rTMS for MDD, providing more personally optimized stimulation method for the DLPFC targeting (Klooster et al., 2023;Weigand et al., 2018;Fox et al., 2012).Nonetheless, neuronavigation is not applied in many clinics, because of the burden of implementation (Najafpour et al., 2021).Alternative approaches of targeting the DLPFC have been studied, including EEG F3 electrode location, and Beam F3 method, which account for individual scalp shape and size (Beam, Borckardt, Reeves, & George, 2009;Herwig, Satrapi, & Schönfeldt-Lecuona, 2003).Furthermore, some studies have reported the possibly suitable sites for the DLPFC stimulation (Fitzgerald et al., 2009a;Fox et al., 2012).Fitzgerald et al. have identified the coordinate of ideal stimulating location at the junction of Brodmann area 9 and 46, and Fox et al. has reported the theoretical stimulation site by assessing previous studies with different coordinates and efficacy (Fitzgerald et al., 2009b;Fox et al., 2012).
The targeting approaches of rs-fMRI and the other conventional methods are very different, as the former is based on functional connectivity and the latter is based on the brain surface data or structural MRI.It brings up a question of how much neuroanatomical difference they have.There are some previous studies that compared several DLPFC targeting methods (Cardenas et al., 2022;Caulfield et al., 2022).However, to our knowledge, most of those studies have targeted patients with MDD, and only one study has compared targeting methods using healthy control subjects (Trapp et al., 2020).It is also important to evaluate differences in preexisting DLPFC targeting methods using data from healthy subjects because the use of healthy subject data has the advantage of excluding the effects of changes in brain structure and function due to psychiatric disorder itself, and of understanding the relationship between target location, head size, and functional connectivity based on normal neurofunctional anatomy.Furthermore, since the previous study using healthy control subjects has compared the Beam F3 method and 5.5 cm rule method in terms of inter-and intra-rater variabilities, it did not evaluate the difference in functional connectivity (Trapp et al., 2020).The anticorrelation of the functional connectivity between the DLPFC and SGC is known to be the shared neural network in both patients with MDD and healthy controls (Vink et al., 2018).Thus, there are neuroscientific implications in assessing functional connectivity associated with depressive symptoms using MRI data from healthy subjects whose neurological function is intact and who are not affected by psychiatric disorders.
Here, the present study aimed to compare the neuroanatomical and neurofunctional differences among the rs-fMRI neuronavigation method, Beam F3 method, and the two target coordinates from the literature by using healthy sample data.Specifically, we examined the distance between each pair of methods, data dispersion, and anticorrelation with the SGC of methods other than rs-fMRI neuronavigation.We did not include data of fixed distance targeting methods because there is growing evidence that anatomical differences result in variability in targeted regions and networks (Cardenas et al., 2022;Trapp et al., 2020;Johnson et al., 2013).Exploring the features of the four targeting methods which were not affected by the pathophysiology of the neuropsychiatric disorders, this study potentially leads to the acceleration of clinical use of rTMS by helping clinicians to choose a targeting approach which is the better or best for individual patients.

Participants
This study acquired data in 57 healthy individuals (25 females; mean ± standard deviation: 30.5 ± 5.5 years old; all Japanese as Asians) who met the following criteria: (1) between ages 18 and 80 at the time of obtaining consent; (2) no history of neuropsychiatric disorders at screening assessed by board-certified psychiatrists; (3) normal cognitive function assessed by a Mini-Mental State Examination score of at least 27 points; (4) no substance related-disorders in the 6 months prior to participation in the study; (5) no contraindications to MRI such as magnetic metal implants, pacemakers, or claustrophobia; (6) no serious or unstable physical diseases; (7) no history of seizures and epilepsy; and (8) not receiving any prescriptions for central nervous system agonists, including psychotropic medications.The experiment was conducted in accordance with the Declaration of Helsinki and was reviewed and approved by the Ethics Committee of Keio University School of Medicine (20170152) and The University of .All subjects provided informed consents before participating in the study.
The subjects underwent structural T1 and T2 MRI and rs-fMRI.Their head sizes, which are length from tragus to tragus, length from nasion to inion, and head circumference, were also measured using a tape measure.The head circumference was measured through Fpz, T3, Oz, and T4.Demographics, averages of each length, and the comparison of them between gender groups using t-test are summarized in Table 1.

Structural and functional MRI
All images were acquired by a 3 T siemens MAGNETOM Prisma MRI scanner (Siemens AG, Erlangen, Germany) with a 32-channel head coil.T1-weighted images were acquired with the following parameters: magnetization prepared rapid acquisition gradient-echo (MPR), echo time (TE) = 2.22 ms; repetition time (TR) = 2400 ms; inversion time (TI) = 1000 ms; flip angle = 8 • ; field of view (FOV) = 240 x 240 mm; slice thickness = 0.8 mm (Koike et al., 2021).We also acquired T2weighted images with the following parameters: sampling perfection with application optimized contrasts using different flip angle evolutions (SPC), TE = 563 ms; TR = 3200 ms; FOV = 240 x 240 mm; slice thickness = 0.8 mm.Two runs of the resting state functional MRI was acquired with the following parameters: TE = 37 ms, TR = 800 ms, flip angle = 52 • , FOV = 208 x 208 mm; slice thickness = 2.0 mm; multiband acceleration factor = 8; 420 volumes for a 5 min 46 sec run in a different phase-encoding direction (AP or PA).The spin echo fieldmap images were acquired with opposite phase encoding directions (Glasser et al., 2016).Each participant was instructed to focus on a fixation cross during the rs-fMRI scan.Immediately after the scan, each participant rated their sleepiness during the scan using a 7-point Likert scale (1 = not at all, 7 = more than ever).No physiological parameters were measured.

Resting-state fMRI analysis
The rs-fMRI data was preprocessed using the HCP pipeline, which involves fieldmap corrections, head movement correction, intensity normalization, single spline resampling of EPI frames into 2 mm isotropic Montreal Neurological Institute (MNI) space, and removal of temporal artifacts using ICA + FIX approach (Glasser et al., 2016;Glasser et al., 2013).In addition, global signal regression (GSR) was implemented prior to band pass temporal filtering, since most previous studies about rTMS antidepressant effect and DLPFC-SGC functional connectivity have implemented GSR (Cash et al., 2019;Cash, Cocchi, Lv, Fitzgerald, & Zalesky, 2021;Fox et al., 2012;Weigand et al., 2018).The DLPFC region of interest was defined following the method of previous studies, which centered 20 mm radius sphere along the left hemisphere at the center of Brodmann area (BA) 9 (MNI: − 36, 39, 43), the center of BA 46 (MNI: − 44, 40, 29), the average stimulation site using the standard 5 cm targeting technique (MNI: − 41, 16, 54) (Fox, Liu, & Pascual-Leone, 2013), and the average stimulation site using the Beam F3 targeting method (MNI: − 37, 26, 49) (Cash et al., 2019).The SGC was also defined based on previous reports, and a 10 mm sphere centered at the SGC was masked (Fox et al., 2012;Hadas et al., 2019).Then the functional connectivity between the SGC sphere and a 10 mm sphere centered at each voxel within the left DLPFC was derived, by averaging the BOLD time courses from voxels within the SGC sphere and correlating it with the average BOLD time course of the sphere at each voxel.Finally, the most anticorrelated voxel was defined as the DLPFC target.

Data of the targets identified by Beam F3 method
We input each of the head size data collected from each subject (head width: length from tragus to tragus; head length: length from nasion to inion; and head circumference) into the F3 Measurement System created by Beam et al. to identify the distance along the left head circumference from the midline at the frontal area (X) and the distance on the line segment through X from the vertex (Y adjusted) and determine target location (Beam et al., 2009;Mir-Moghtadaei et al., 2015).We used the adjusted y-value to best approximate the optimal target as recommended by Mir-Moghtadaei et al. (Mir-Moghtadaei et al., 2015).Then each subject's T1-image was loaded into Brainsight image-guidance software (Rogue Research, Montreal Canada) to reconstruct individual three-dimensional scalp images.Using the ruler function of Brainsight, the obtained X and Y adjusted values were traced virtually on the scalp images to find the MNI coordinates of the targets gained by the Beam F3 system.Secondly, to calculate the functional connectivity between a target identified by the Beam F3 method and SGC, the nearest fMRI voxel from the target was determined for each subject by using an algorithm based on the Euclidean distance metric.A Python script was created which iteratively calculates the distance between the reference point and each voxel using the square root of the sum of squared coordinate differences.The voxel with the minimum calculated distance was identified as the nearest point.The correlation between the voxel and SGC was used as the approximation of the Beam F3 target's correlation value with SGC.

Data of the targets with fixed coordinates
The targets reported by Fitzgerald 2009 (MNI: − 45, 45, 35) and Fox 2012(MNI: − 38, 44, 26) were included in the analysis (Fitzgerald et al., 2009a;Fox et al., 2012).For each coordinate, the nearest fMRI voxel was determined, and the approximate correlation between the coordinate and SGC was obtained, in the same way as for 2.2.2.Beam F3 method.

Comparative analysis
The coordinates for each subject were plotted on the ICBM 152 average brain on Brainsight to visualize the distance and dispersion.The distances between each pair of targets were calculated.The dispersion of the targets located by rs-fMRI and Beam F3 methods were assessed for each by calculating the average of coefficients of variation of the x, y, and z coordinates.
The correlation coefficients of functional connectivity between the DLPFC and the SGC obtained for each subject were averaged, and first analyzed using one-way analysis of variance (ANOVA) to find whether there was a significant difference in those values, where p < 0.05 was considered to be statistically significant.Then the difference between every possible pair of values were analyzed using Tukey's Honestly Significant Difference (HSD) post hoc tests, and statistical significance was defined as adjusted p-value less than 0.05.

Distance and dispersion of the DLPFC targets
Fig. 1A shows the average target coordinates for each method plotted on the ICBM 152 average brain.The small yellow spheres represent the MNI coordinates of each subject determined by rs-fMRI neuronavigation, while the small cyan spheres represent those located by the Beam F3 method.The large yellow and cyan spheres are the average coordinates of each method.The large red sphere is the target reported by Fitzgerald 2009, and the blue one is the target reported by Fox 2012.The distances between each pair of targets were calculated and shown in Table 2.
Targets determined by rs-fMRI appear to have greater dispersion than the Beam F3 method; the average coefficient of variation of the three coordinates determined by rs-fMRI was 0.429, while that of Beam F3 was 0.133.

Comparison of functional connectivity with the SGC across the DLPFC targeting methods
Fig. 1B shows the correlation coefficients with the SGC for each targeting method.The one-way ANOVA indicated significant differences of correlation among the four methods (F(4, 52) = 51.973,p < 0.001, partial η 2 = 0.821).The results of subsequent Tukey's HSD post hoc tests are shown in Table 3. Rs-fMRI targeting method had statistically significant differences with each of the other methods, with the average correlation coefficient of − 0.491 ± 0.088 (mean ± SD).On the other hand, no significant differences were found among the results of Beam F3, Fitzgerald 2009, and Fox 2012 targeting methods, suggesting similar correlation coefficients with the SGC among these groups.When limiting to male or female participants, one-way ANOVA and subsequent Tukey's HSD post hoc tests also resulted in significantly greater average correlation coefficient of rs-fMRI and no significant differences of functional connectivity among the other three methods.

Discussion
This study demonstrated the differences regarding (1) anatomical location, (2) anatomical dispersion (if data for each subject was applicable), and (3) functional connectivity with SGC among the DLPFC targets determined by four methods: 1) rs-fMRI neuronavigation, 2) Beam F3, 3) the target on the structural MRI reported by Fitzgerald 2009, and 4) the one reported by Fox 2012.The target coordinates for all subjects were averaged for each method, and those were plotted on the ICBM 152 average brain to compare visually.From the result shown in Table 2, the targets located by the three methods other than rs-fMRI neuronavigation were relatively closed together, while all of them had great distances to the rs-fMRI target.Fig. 1A also revealed that the targets were more distributed in rs-fMRI neuronavigation.Furthermore, the correlation coefficients with the SGC were compared among the methods, and our analysis indicated rs-fMRI targeting method had a significantly greater anticorrelation compared to the other three methods, and there were no significant differences in the anticorrelation among the three methods.In addition, these results remained the same when limiting to male or female participants that had statistically different head sizes.
One of our main findings was that targets located by rs-fMRI had more dispersion compared to the Beam F3 method, as shown in Fig. 1A.As mentioned above, rs-fMRI targeting approach is known to yield better therapeutic outcome of rTMS for patients with MDD (Klooster et al., 2023;Weigand et al., 2018;Fox et al., 2012), possibly because it takes account of not only the subjects' head size but also the functional connectivity within each subjects' brain.Our results of disperse targets by rs-fMRI reflects the known fact, and could indicate rs-fMRI is the more tailored targeting method for each subject.In contrary, the targets located by Beam F3 method did not have much dispersion, suggesting its limitation of demonstrating different neural activity for each subject.In addition, Table 2 shows that the three conventional approaches other than rs-fMRI resulted in targets relatively close together.This also suggests that conventional targeting methods do not approximate personalized neuronavigation targeting methods, although in some ways this is to be expected.This study used data from healthy subjects, but when it comes to patients with MDD, it is possible that the individual variability in neural network would be greater as it is affected by the pathophysiology of the disorder.If so, there would be less chance for the conventional targeting approach to approximate the optimal stimulation site for the individual.
Furthermore, the personalized rs-fMRI approach targeted the stimulation site with the greatest anticorrelation with SGC, whereas the anticorrelations with SGC were significantly smaller when targeted by the other three methods.These results confirmed that rs-fMRI was the most precise targeting method, given the previous reports of greater treatment outcome by targeting sites which had more anticorrelation with the SGC (Cash et al., 2019).On the other hand, the correlation did not differ among the other three targeting methods, which suggests the interchangeability of these three non-personalized approaches in terms of potential therapeutic effects by rTMS.However, we acknowledge that there is a limitation that the present study is based on data from healthy subjects.Although the anticorrelation of the functional connectivity between the DLPFC and SGC is present in both patients with MDD and healthy controls (Vink et al. 2018), it is difficult for this study to directly discuss the anticorrelation between the DLPFC stimulation sites and SGC in the context of the therapeutic effects of rTMS in patients with depression.Nonetheless, we believe that it is still important to examine the difference of the data which is purely based on the head size and functional connectivity on healthy subjects, not affected by mental illness itself, to compare the different methods of targeting the DLPFC.
Another finding was that our results possibly indicated the optimal targeting methods for a specific symptom of MDD.Recently there are some studies reporting that the optimal stimulation site of rTMS varies within the DLPFC depending on the targeted symptoms (Fitzgerald, 2021).Siddiqi et al. has retrospectively analyzed the relationships between rTMS stimulation sites and the improved symptoms, and the results indicated that dysphoric symptoms such as sadness and anhedonia responded best to the stimulation at the intersection of BA 9, 10, and 46, which is the left anterior site of DLPFC, while anxiety and somatic symptoms responded best by stimulating BA 8, which is the posterior site of DLPFC (Siddiqi et al., 2020).Furthermore, in the same study, Siddiqi et al compared some targeting methods on patients with MDD and found that targets identified by the rs-fMRI neuronavigation and Beam F3 methods were both at the left anterior site of DLPFC which was close to the peak dysphoric target, and the target determined by the 5 cm rule was at the posterior site of DLPFC which was close to the peak anxiosomatic target (Siddiqi et al., 2020).Although the rs-fMRI targeting method of our study resulted in the posterior site of DLPFC, which is because the degree and localization of anticorrelation of functional connectivity may differ between healthy subjects and patients with MDD, Beam F3 targeting method of our study also resulted in the left anterior site of DLPFC, supporting the previous study which suggested that Beam F3 targeting method was optimal for dysphoric symptoms such as sadness, anhedonia, and suicidal thoughts.Furthermore, though the previous study did not include the targets of Fox 2012 andFitzgerald 2009, our results suggest that Fox 2012 targeting method may be more effective for dysphoric symptoms rather than anxiosomatic symptoms because it was located in the left anterior part of the DLPFC, while it was not clear about Fitzgerald 2009 targeting method because the target was at the middle of the left anterior and posterior parts of the DLPFC.
Building upon the findings of this study, it is expected to compare the actual therapeutic outcomes among the different targeting methods, examining patients with MDD prospectively.Moreover, it is important to examine the contrasts between the improvement of various symptoms, assessing the reproducibility of previous studies about the optimal targeting sites for different symptoms and hopefully developing the idea of customized targeting by the main symptom (Siddiqi et al., 2020).
To date, there are several studies that compared the different targeting methods for the rTMS stimulation site.Trapp et al. have compared the targeted locations and reliability of some targeting methods using healthy samples, though it did not include the functional connectivity method for their comparison as this study did (Trapp et al., 2020).Cardenas et al. have compared the fixed distance targeting rules, Beam F3 method, and the stimulation sites reported as optimal in previous studies using data of patients with MDD in terms of cortical regions of the stimulation sites and their functional connectivity with the SGC, while Fox et al. used the data of rs-fMRI of healthy samples to assess the difference in functional connectivity with the SGC among different MNI coordinates that were reported in previous literatures (Cardenas et al., 2022;Fox et al., 2012).However, these previous studies did not include the rs-fMRI neuronavigation method for their comparison, whereas our study compared the average targeted location of different targeting methods including the rs-fMRI neuronavigation method as well.Last but not least, the samples of this study were limited to Japanese, Asians, which would have different scalp structure compared with the other human races (Matsumura et al., 2022), making the targeting methods using scalp measurement differ from the other previous studies that have focused on largely Caucasian populations.
The present study has several limitations.First, although the discussion above is based on the fact that personalized neuronavigated stimulation is associated with greater antidepressant effect, this method is only used in research at this time and has not yet been applied in clinical practice, because of the burden of implementation of the technique.Moreover, Hebel et al. have reported no significant difference in therapeutic effect between the neuronavigated method and the conventional method (Hebel et al., 2021).Nonetheless, in the study, stimulation target was not defined by individual functional connectivity but by the fixed coordinate on neuroanatomical neuroimages, which is different from the rs-fMRI targeting methods that are proven to be therapeutically more effective (Klooster et al., 2023;Weigand et al., 2018;Fox et al., 2012).Second, though we did not include the fixed distance targeting method in our study since it is known to have significant inter-rater variability and less anticorrelation with the SGC (Cardenas et al., 2022;Trapp et al., 2020;Johnson et al., 2013), one study has reported similar antidepressant effects between the Beam F3 method and 5.5 cm method (Trapp et al., 2023).Third, some data values in this study may not be exactly precise.For instance, the correlation coefficients with the SGC of the targets located by the methods other than the rs-fMRI were approximated using the voxel coordinate of rs-fMRI, which was the nearest from each actual coordinate.
In conclusion, our results provided support for the optimal selection among the methods of targeting rTMS stimulation sites.Locating the DLPFC by rs-fMRI resulted in the more dispersed targets depending on the subjects, reinforcing the known fact that the method yields better therapeutic outcome with personalization for individual neural networks.While targets located by rs-fMRI had significantly greater  Tukey's HSD post-hoc test was performed to correct for multiple comparisons.CI: confidence interval.
anticorrelation with the SGC, Beam F3 method and the targets reported by Fitzgerald 2009 andFox 2012 did not differ in terms of the anticorrelation with the SGC, which could imply the possible convertibility of them regarding the therapeutic efficacy of TMS treatment.

Role of funding source
This research received no external funding.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Fig. 1 .
Fig. 1.The stimulation sites over the brain image and functional connectivity with the subgenual cingulate cortex (SGC) determined by each targeting method.1A: The three-dimensional brain image with plots were displayed from three directions: left frontal oblique (the top), left lateral (the bottom left), and frontal (the bottom right).The large spheres indicate the average target coordinates by rs-fMRI and the Beam F3, or the targets defined by Fitzgerald et al. and Fox et al. rs-fMRI: yellow; Beam F3: cyan; Fitzgerald et al.: red; Fox et al.: blue.1B: Correlation coefficients with the SGC were evaluated for each targeting method on each participant.The data was plotted as black dots, and the wider section of the graph represents a higher density of the dots.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) authorship contribution statement Megumi Kinjo: Writingoriginal draft, Methodology, Investigation, Formal analysis.Shiori Honda: Writingreview & editing, Methodology, Investigation, Data curation.Masataka Wada: Writingreview & editing, Investigation.Shinichiro Nakajima: Writingreview & editing, Investigation.Shinsuke Koike: Writingreview & editing, Investigation.Yoshihiro Noda: Writingreview & editing, Writingoriginal draft, Supervision, Project administration, Methodology, Investigation, Conceptualization.

Table 1
Demographics and head sizes of the subjects; comparison of the demographics between gender groups.
Mean ± standard deviation is given for each variable, cm: centimeter.M.Kinjo et al.

Table 2
Distance between targets.

Table 3
Comparison of correlation coefficients with the SGC.