Influence of prefrontal target region on the efficacy of repetitive transcranial magnetic stimulation in patients with medication-resistant depression: a [¹⁸F]-fluorodeoxyglucose PET and MRI study Free

Abstract

It is currently unknown whether the antidepressant effect of repetitive transcranial magnetic stimulation (rTMS) depends on specific characteristics of the stimulated frontal area, such as metabolic changes. We investigated the effect of high-frequency rTMS, administered over the most hypometabolic prefrontal area in depressed patients in a two-site, double-blind, randomized placebo-controlled add-on study. Forty-eight patients with medication-resistant major depression underwent magnetic resonance imaging and [¹⁸F]-fluorodeoxyglucose positron emission tomography (PET) in order to determine a target area for rTMS. After randomization to PET-guided (n=16), standard (n=18), or sham rTMS (n=14) conditions, the patients received 10 sessions of 10-Hz rTMS (1600 pulses/session) at 90% motor threshold. Change from baseline in Montgomery–Åsberg Depression Rating Scale (MADRS) scores did not differ between PET-guided, standard and sham groups at 2-wk end-point. Exploratory comparison of left PET-guided (n=9), right PET-guided, standard, and sham rTMS revealed significant effects. The highest improvement in MADRS scores was observed with left PET-guided (60±31%), significantly superior to sham (30±37%, p=0.01) and right-guided (31±33%, p=0.02) stimulation. Comparison between left PET-guided and standard rTMS (49±28%) was not significant (p=0.12). Comparison between stimulation over dorsolateral prefrontal cortex (BA 9-46), stimulation of other areas, and sham rTMS was statistically significant. Stimulation over BA 9-46 region (n=15) was superior to sham rTMS (p=0.02). The results do not support the general hypothesis of increased antidepressant effects of high-frequency rTMS with prefrontal hypometabolism-related PET guidance. Nonetheless, whether metabolism and anatomy characteristics of left frontal area underneath the coil might account for an increase or speeding up of rTMS effects needs further investigation.

Introduction

Over the past decade, repetitive transcranial magnetic stimulation (rTMS) has emerged as a potential new therapy for medication-resistant depression (George et al.1995; Lam et al.2008; Pascual-Leone et al.1996). However, despite an increasing number of studies, the optimal stimulation site has not been unequivocally determined (Daskalakis et al.2008; Fitzgerald et al.2006a). Several reviews and meta-analyses of rTMS studies in patients with resistant depression have demonstrated, at best, a moderate clinical effect of rTMS (Burt et al.2002; Gershon et al.2003; Loo & Mitchell, 2005; Martin et al.2002), and other studies have reported no difference between active and sham rTMS (Couturier, 2005; Mogg et al.2008). The high variability in response across studies has been related to differences in parameters such as the patients' clinical characteristics and the intensity, frequency, and location of the stimulation (Daskalakis et al.2008; George et al.2008; Herrmann & Ebmeier, 2006). To increase the effects, some authors have proposed bilateral stimulation with high-left and low-right frequencies (Fitzgerald et al.2006b), or suggested that longer durations of treatment (at least 6 wk) and/or higher stimulation intensity might result in better effectiveness (Anderson et al.2007; Avery et al.2008; O'Reardon et al.2007); however, these strategies increase the treatment complexity and cost (Knapp et al.2008).

The neurophysiology behind the potential antidepressant effect of rTMS is not clearly understood. The effects of rTMS on mood have been related to its ability to modulate brain functions such as cortical excitability (Pascual-Leone et al.1994) or blood flow. High-frequency rTMS (⩾5 Hz) has been shown to increase brain activity both locally and in distant regions while low-frequency TMS (⩽1 Hz) may decrease brain activity in depressed patients (Speer et al.2000).

Based on empirical observations of mood-related effects of prefrontal stimulation and reports of left prefrontal hypometabolism in depressed patients in previous neuroimaging studies (Baxter et al.1989; Bench et al.1992; Martinot et al.1990), George and co-workers (1995) suggested that the most effective antidepressant effect might be obtained by applying rTMS to the left prefrontal cortex. Thereafter, Pascual-Leone et al. (1996) specified that the targeted area should be Brodmann area (BA) 46 and, more recently, functional studies have suggested that BA 9 would be the optimal site (Fitzgerald et al. 2006 a). In addition, high-frequency rTMS was found to be more effective over the left dorsolateral prefrontal cortex (DLPFC) than over the right DLPFC (Pascual-Leone et al.1996). Thus, most high-frequency rTMS studies have aimed at stimulating the left DLPFC (particularly BA 46 or an area encompassing left BA 9 and BA 46) either by using a ‘standard’ procedure that targets an area 5 cm anterior to the hand motor cortical representation (George et al.1997; Paus & Barrett, 2004) or by using neuronavigation devices (Herwig et al.2003). However, the choice of this target region remains speculative (Fitzgerald et al. 2006 a), and the clinical effects have been only moderate, even when using neuronavigation (Herwig et al.2003). Indeed, functional patterns most often result from group analyses and do not take into account individual variability. Imaging studies conducted in depressed patients have actually revealed various metabolic patterns, with considerable inter-subject variability regarding the hypometabolic frontal subregions (Fitzgerald et al. 2006 a; Videbech et al.2002). Moreover, although BA 9 and BA 46 are connected to regions implicated in the regulation of mood, such as the anterior cingulate and the caudate (Barrett et al.2004), it is not clear whether the prefrontal hypometabolic clusters that should be targeted are within those areas or within other subregions in the DLPFC. One study has suggested that the application of high-frequency rTMS over hypometabolic prefrontal areas would yield better clinical effects in patients (Kimbrell et al.1999), which is in line with earlier reported increased left prefrontal metabolism in patients recovering from depression (Baxter et al.1989; Kennedy et al.2001; Martinot et al.1990). Furthermore, the stimulation of the DLPFC according to the underlying cortical dysfunction has been proposed to be superior to stimulation that does not take the metabolic state into account (Herwig et al.2003; Garcia-Toro et al.2006; Mottaghy et al.2002).

In the present study, we investigated whether focusing high-frequency rTMS on the most hypofunctional prefrontal cluster determined at the individual level would be more effective than stimulation of the ‘standard’ location in the treatment of depressive symptoms. Additionally, we sought to determine whether stimulation side or stimulation over the BA 9 and/or BA 46 regions in the DLPFC would influence the therapeutic effects of rTMS.

We designed a two-site, randomized controlled study in which we compared the effects of positron emission tomography (PET)-guided rTMS, left prefrontal standard rTMS, and sham rTMS in patients with pharmaco-resistant depression. The PET-guided rTMS target was the prefrontal cluster with the maximal hypometabolic voxel (i.e. peak voxel), which was determined individually using [¹⁸F]-fluorodeoxyglucose PET (FDG PET).

Method

The investigation was performed in accordance with the Declaration of Helsinki. The study was approved by the ethics committee Ile de France 6, Paris. Written informed consent was obtained from all subjects after full description of the study.

Participants

Fifty-eight patients with a DSM-IV-R diagnosis of major depressive disorder were recruited by senior psychiatrists from consecutive admissions at five university psychiatry departments. Diagnoses were established on the basis of clinical interviews and administration of the Mini-International Neuropsychiatric Interview (MINI; Sheehan et al.1998). The patients were screened for resistance to at least two trials of antidepressants of different classes given at adequate doses (>150 mg/d in an equivalent dose of imipramine) and duration (at least 4 wk for each drug). No incentives were offered.

Exclusion criteria included age >65 yr, alcohol or substance dependence in the past 6 months, electroconvulsive therapy (ECT) treatment in the past 6 months, any present medical condition, history of epileptic seizures, history of neurological disorders or substantial brain damage, and contraindication to magnetic fields, according to established safety criteria (Wassermann, 1998).

After initial screening, the patients underwent brain imaging. Four patients were then excluded due to silent neuropathological abnormalities. Of the remaining 54 patients, two patients dramatically improved in the 3-d gap between brain imaging and the start of the TMS protocol. One patient who scored 18 on the Montgomery–Åsberg Depression Rating Scale (MADRS; Montgomery & Åsberg, 1979) and 16 on the Hamilton 21-item Depression Scale (HAMD₂₁; Hamilton, 1960) at baseline was excluded from the study as she did not fulfil the scoring requirements for inclusion in clinical trials (Mottram et al.2000). Finally, one patient was excluded from the TMS protocol as his motor threshold was higher than the maximal capacity of the rTMS device used. Thus, 50 patients (32 females) entered the TMS study.

Comparison group

Each patient was compared for FDG PET with 25 healthy volunteers (values given are mean±s.d.) (age 37.04±10.1 yr, 13 males) who were recruited from the community by word of mouth and had no personal or family history of psychiatric disorder, as assessed by a medical examination. Due to the necessity to constitute the comparison group before starting the study, there was a significant age difference between the patients (age 47.14±8.20 yr) and the comparison subjects (t test=−4.65, d.f.=73, p<0.0001). The comparison subjects did not significantly differ from patients with respect to gender (five males, Fisher's exact test, two-tailed=0.57) and education level (years of education after primary school, patients 7.5±4.14 yr; comparison subjects 9.24±4.38, t test=−1.69, d.f.=72, p=0.10).

Clinical assessment

Baseline assessment was performed on the day before brain imaging, and the last clinical assessment was performed after 10 sessions, on the last treatment day. Clinical evaluations were performed by senior psychiatrists or psychologists trained together on ratings, using the MADRS, HAMD₂₁, and the Clinical Global Impression of Illness – Severity (CGI-S).

Treatment allocation

Stratified randomization was performed in blocks including 11 subjects (4+4+3), with each treatment at least once in the first four, the second four, and the last three patients. Randomization was stratified on the stimulation site and two allocation lists were generated by the Biostatistics Department. Allocation concealment was performed using closed envelopes that indicated the treatment modality for each patient and were kept in each stimulation site and opened by the investigator performing the treatment immediately before the first treatment session.

Hence, the patients were randomly assigned either to PET-guided TMS (n=17), standard TMS (n=19), or sham standard TMS (n=14). Patients and symptom raters were blind to the treatment modality.

Comorbidities

Twelve patients reported symptoms of a comorbid diagnosis of anxiety disorder (panic disorder with or without agoraphobia, or generalized anxiety disorder); four were in the PET-guided, three in the standard, and five in the sham-treated groups (χ²=1.52, p=0.47). None of them had anxiety symptoms during the scanning protocol. Sixteen patients had resistant bipolar depression (seven in the PET-guided group, four in the standard group, and five in the sham group, χ²=1.86, p=0.39).

Medication

rTMS was administered as add-on therapy (Table 1). During the study, the patients were treated with minimal and stable doses of their previous treatment for at least 2 wk. Low-dose hypnotics prescribed in a naturalistic manner were allowed in case of severe insomnia only.

Current and past treatments, tricyclic antidepressants, selective serotonin reuptake inhibitors (SSRIs), mood stabilizers, or ECT prescriptions, did not differ between groups (Table 1).

Brain imaging

All participants were investigated at rest. Magnetic resonance imaging (MRI) high-resolution T1-weighted images were acquired with a 3D MRI sequence (124 T1-weighted images; field of view 24 cm; 256×256×128 matrix; voxel size 0.94×0.94×1.3 mm³) on a 1.5 T GE Signa scanner (General Electrics Medical Systems, USA).

FDG PET 3D-images were obtained following a transmission scan for attenuation correction from a Siemens ECAT-EXACT-HR+ tomograph (Siemens Medical Solutions, USA), which collects 63 simultaneous slices (intrinsic in-plane resolution 4.3 mm; voxel size 2.42×2.42×2.43 mm³). Two 3D time-frames (10 min each) were collected 30 min after injection. Attenuation- and decay-corrected PET images were summed. All participants were injected with 111–222 MBq of FDG. The mean (±s.d.) injected radioactivity was 156.43±10.55 MBq in the patients and 157.92±24.47 MBq in healthy subjects (unpaired t test=0.36, d.f.=72, p=0.72).

Image processing

Image transformation and voxel-based analysis were performed using Statistical Parametric Mapping software (SPM99, Wellcome Department of Cognitive Neurology, University College, London, UK; http://www.fil.ion.ucl.ac.uk/spm), implemented on Matlab^®.

Each FDG PET image was co-registered with the corresponding T1 MRI image. As SPM provides no FDG template, we first created a FDG template to improve spatial normalization. The healthy subjects' MRI images were spatially normalized using the FIL T1 template provided within the SPM99 software package; then, the obtained transformation matrix was used to normalize the FDG PET realigned images. The mean sum of the comparison subjects' FDG PET images was computed, providing a mean image that was then smoothed with an 8 mm full-width half-maximum isotropic Gaussian kernel and was used as the FDG template to normalize all subjects' images. Following spatial normalization, all images were smoothed by 8 mm.

rTMS prefrontal target determination

In order to determine a TMS target based on prefrontal hypometabolism in each patient, the individual FDG PET baseline images were compared with the healthy comparison group using the single-subject conditions and covariates procedure in SPM99, with age and gender as confounding covariates [a method similar to that previously validated with SPM96 (Signorini et al.1999)]. These ‘healthy comparison group minus single-patient’ contrasts were examined using an exploratory height threshold at p<0.05, uncorrected, and an extent threshold at k=10 voxels (80 mm³), in a whole-brain analysis, which allowed determination of hypofunctional prefrontal clusters at the individual level. Then, as the usual target for treating depression with rTMS is the prefrontal cortex (including superior, middle and inferior frontal gyri), the most significant peak voxel (i.e. voxel with the highest t value) was chosen in the prefrontal clusters within TMS coil-cortex accessible distance range. The coordinates in MNI space (Montreal Neurological Institute) for the peak voxel within the chosen cluster were obtained using SPM99. This peak voxel chosen as a target was then projected onto the head surface (MRI head mesh) using a procedure analogous to that described elsewhere for fMRI (Andoh et al.2006, 2008; Andoh et al. in press). Briefly, MRI T1-weighted native images were segmented using Brainvisa (http://brainvisa.info) image processing pipeline to obtain 3D individual head meshes (Mangin et al.1998). Then, the FDG PET ‘healthy comparison group minus patient’ SPM map was automatically projected onto the patient's head mesh using Anatomist (http://brainvisa.info) and the projection of the previously chosen peak voxel on the SPM map was selected as a target on the head mesh. This procedure allowed checking of the depth of the projected FDG PET peak voxel. A depth <25 mm was required to ensure that the magnetic field peak delivered by rTMS could reach the cortex. Along with the peak voxel projection, anatomical landmarks including the nasion, vertex, and left and right auricular tragi were selected. Coordinates of the landmarks computed with Anatomist were then entered into an algorithm embedded in Brainvisa that allowed computation of geodesic distances (i.e. tangents to the head surface) between the target and the landmarks. The three distances between target and nasion, target and left auricular tragus, and target and right auricular tragus were provided in millimetres by Anatomist software and were then used to position the TMS coil manually over the subject's head (Andoh et al. in press) (Fig. 1).

In addition, hand motor cortex targets were computed on each patient's head mesh using the projection of the hand knob that was previously determined on the subject's MRI. In order to estimate the cortex-under-the-coil coordinates in each patient that had been allocated to standard rTMS, standard location was virtually re-constructed after treatment completion, 5 cm ahead of the motor target represented on the head mesh, using the Anatomist distance scale centred by the motor target. Then, a sphere centred by the target was defined using the Anatomist distance scale; variation in the distance scale radius allowed selection of the cortical region closest to the scalp on the brain mesh. MNI coordinates of those standard locations at the brain surface were then computed in Anatomist using each subject MRI image normalized in SPM2. Those coordinates were then translated into Talairach coordinates (Talairach & Tournoux, 1988) using Pick Atlas toolbox (Maldjian et al.2003) in SPM2 and further translated into Brodmann areas. Finally, in order to estimate coil–cortex distance, scalp–cortex distance was measured a posteriori in each patient using the head mesh and brain mask (i.e. voxel value is 1 in the brain and 0 elsewhere) fusion procedure in Anatomist at the target or at reconstructed standard locations. All of these assessments were performed blind to clinical ratings.

Stimulation procedure

rTMS treatment was administered in two stimulation sites that were equipped with the same TMS devices and by investigators who were not involved in clinical ratings or PET target processing.

TMS protocol

We used Magstim super-rapid^® devices with active and sham air-cooled figure-of-eight coils (Magstim Co., UK). Both coils had the same appearance and made a similar noise. The patients were blind to the treatment modality and had never previously been treated with TMS. They had all been told that the treatment side could be related to their brain-imaging data. The investigators administering the treatment were blind to the clinical ratings but not to the status of the coil.

Motor threshold (MT) was determined according to a standardized procedure (Rothwell et al.1999) either on the left thumb, if the stimulation target was on the right hemisphere, or the right thumb, for the left hemisphere.

Twenty trains of 8 s with 60-s inter-train intervals were administered with stimulus frequency at 10 Hz and intensity at 90% of MT, resulting in a total of 1600 pulses over 20 min. rTMS was administered on 10 consecutive working days, providing a total of 16 000 impulses. Sham rTMS was performed using the same procedure. All patients wore earplugs.

Coil positioning

The patients wore a stretchable swimming cap on which the target position was drawn according to the computed geodesic distances of the target in the guided group, or according to the motor cortex location in the other groups. PET-guided rTMS was applied on the target determined for each patient, either on the left or on the right hemisphere.

Standard stimulation was left prefrontal, 5 cm anterior to the motor hot-spot of the hand (abductor pollicis brevis muscle).

Statistical analysis

Sample size determination was based on an estimation of the effects variability size, as reported in the literature at the beginning of the study (7-point decrease in HAMD₂₁ score, with a similar standard deviation) (George et al.1995). At least 15 patients in the placebo-treated group and 16 patients in each actively treated group were required to control for an α-level error of 5% and a β-level error of 20%, based on unilateral assumptions.

The clinical data were analysed using JMP 6 software from SAS (USA). Intent-to-treat efficacy analysis was performed on all patients who had a baseline measure and at least one post-baseline observation available for analysis, as previously done in other studies (O'Reardon et al.2007). The main efficacy criterion was change from baseline in MADRS score. Secondary criteria included change from baseline in HAMD₂₁ and CGI-S scores.

The primary analysis compared baseline scores and change from baseline across treatment subgroups using factorial ANOVA with treatment subgroup (guided, standard, sham) and stimulation site (1, 2) as independent variables. Post-hoc analyses were performed using least square (LS) means differences Student's t tests.

Second, as the FDG PET targets were located in both brain hemispheres and the effects of rTMS depend on the stimulated hemisphere (Pascual-Leone et al.1996), the PET-guided group was divided into left-side and right-side treatment subgroups. A secondary analysis was performed using factorial ANOVA with treatment subgroup (left PET-guided, right PET-guided, standard, sham) and stimulation site (1, 2) as independent variables.

In addition, an exploratory analysis of the effects of the active treatment over BA 9-46 compared with active rTMS over other areas and with sham rTMS was performed. Responders were defined as having a MADRS score change from baseline ⩾50%.

The Kruskal–Wallis test was used to compare age, years of education, duration of illness, duration of episode, and scalp–cortex distance in the three treatment groups. Age and years of education between patients and controls of the brain-imaging protocol were compared using unpaired t tests. Gender differences were compared using χ² test.

Significance was set at p<0.05, two-tailed. Moreover, due to the small sample sizes of the subgroups in the secondary analysis, indices of effect size (Cohen's d; Cohen, 1988) and Number needed to treat (NNT; Cook & Sackett, 1995) were computed.

Results

Due to marked anxiety, two patients dropped out at the beginning of the study (after one or two rTMS sessions) (Fig. 2); one of them was in the left PET-guided group (a 49-yr-old female, MADRS score: 38, HAMD₂₁: 30) and the other in the standard group (a 38-yr-old male, MADRS: 37, HAMD₂₁: 25). These patients were excluded from further analyses. No other serious adverse event was observed. Overall, the stimulation was well tolerated.

Brain-imaging analysis allowed computation of functional SPM map-driven targets in all of the patients except for two who did not display any prefrontal hypometabolic cluster. Those two patients were both randomly allocated to the standard group. In all the patients randomly assigned to the PET-guided treatment, a stimulation target could be determined from functional SPM maps (Table 2, Fig. 3). Thirty out of 48 patients (10 sham, 11 standard, and 9 PET-guided treated, i.e. 62.5% of the patients) displayed the most hypometabolic peak voxel in the left hemisphere, and 16 patients in the right hemisphere (4 sham, 5 standard, and 7 PET-guided treated, i.e. 33.3% of the patients). Those hypometabolic peak voxels corresponded to BA 9 or BA 46 MNI coordinates in 37.5% of the patients (Tables 2, 3).

Scalp–cortex distance measured with Anatomist was 19.6±6.4 mm in the PET-guided group and 18.4±4.9 mm in the standard group (t test=0.63, d.f.=32, p=0.53). No difference was found in scalp–cortex distance between standard, left and right target subgroups (Kruskal–Wallis test=0.40, d.f.=2, p=0.82).

Eight patients in the PET-guided group and seven patients in the standard group were actually stimulated over BA 9-46 (Table 2, Fig. 3).

The response rate was 55% in the standard group, 50% for the whole PET-guided group, and 21% in the sham group (3/14 patients, χ²=0.13). In the patients treated over BA 9/46, 47% were responders, while in the patients treated over other areas, 58% were responders (Fisher'exact test=0.73, two-tailed) (Table 2).

Primary analyses

There was no between-group difference in patients' baseline characteristics or in changes from baseline in MADRS, HAMD₂₁, or CGI scores (Table 1). No significant effect of stimulation site or significant interaction between treatment group and stimulation site was observed.

Secondary analyses

The comparison between left PET-guided, right PET-guided, standard rTMS, and sham rTMS groups revealed a significant between-group difference in MADRS score changes from baseline (Table 4). No significant effect of stimulation site or significant interaction between treatment group and stimulation site was observed. Post-hoc analysis showed that left-guided rTMS was significantly more effective than sham and right-guided conditions, with large effect sizes. No difference was found between left-guided and standard, although the effect size was in the medium range favouring the left-guided group. Standard and sham, standard and right-guided, and right-guided and sham conditions did not differ from each other (Table 4).

The comparison of the 15 patients stimulated over BA 9-46 with the 19 patients treated over other areas (BA 6, BA 8, BA 10, BA 45, BA 47) and with the 14 sham-treated patients showed a significant difference in treatment effects (F=3.34, d.f.=2, p=0.04) favouring the BA 9-46 stimulation (MADRS mean change −20.4±10.6) over sham stimulation [MADRS LS mean difference 10.5, t=2.48, d.f.=42, 95% CI 2–19, p=0.02; Cohen's d=0.86; NNT=3.96] and, at a trend level, over stimulation of other BAs (MADRS mean change −12.1±10.3, LS mean difference 7.54, t=1.93, d.f.=42, 95% CI 0–15, p=0.06; Cohen's d=0.79; NNT=8.92). Sham stimulation was similar to ‘other BA’ stimulation [MADRS LS mean difference 2.9, t=0.74, d.f.=42, 95% CI −5 to 11, p=0.46; Cohen's d=0.14; NNT=2.74].

Discussion

We investigated whether individual prefrontal metabolic rate underneath the stimulation site could influence the effects of rTMS. We used a PET-guided method to treat pharmaco-resistant depressed patients with high-frequency rTMS over the most hypometabolic prefrontal area accessible to the magnetic field. After a 2-wk treatment, improvement in depression scores did not differ between patients treated with standard rTMS, sham rTMS, or PET-guided TMS. However, the improvement of patients with the left-side PET-guided condition surpassed that of right-side PET-guided and that of sham-treated patients. Moreover, the results showed an advantage in stimulating over BA 9-46.

These findings confirm the previously reported lack of antidepressant effect of high-frequency rTMS over right prefrontal cortex, even when taking into account the regional hypometabolism to guide stimulation. This is in contrast with the study by Herwig and co-workers (2003), who found similar positive effects of both right and left rTMS when stimulation occurred over the DLPFC. Our findings also contradict the hypothesis that stimulating hypometabolic areas with high-frequency rTMS, irrespective of the stimulation side, should improve depressive symptoms (Kimbrell et al.1999), as right PET-guided rTMS effects were similar to sham stimulation effects. As in a recent study, using higher intensity stimulation and a larger number of patients (O'Reardon et al.2007), no difference was seen between standard and sham modalities at 2-wk end-point. In that larger and longer trial, significant effects of rTMS appeared at the 4-wk end-point. In contrast, in the present study, an early significant effect of left PET-guided over sham rTMS could be detected at 2 wk. However, no difference was found between left PET-guided and standard stimulation.

Despite the small number of subjects, our results indicate a strong effect size of the left PET-guided over the sham modality, with a NNT of ∼2 in order to achieve treatment response. Thus, it could be suggested that targeting left hypometabolic prefrontal areas might increase – or speed up – the effect of the treatment compared with standard location alone.

Two previous studies with rather small numbers of patients attempted to locate the stimulation coil over hypofunctional areas using either SPECT imaging (Garcia-Toro et al.2006) or neuronavigation and PET imaging (Herwig et al.2003). Those trials were not able to show any difference between the effects of standard stimulation and stimulation of the hypofunctional side. However, in both studies the stimulation was applied over the neuronavigated DLPFC as a whole region of interest (when hypofunctional) (Herwig et al.2003), or over hypofunctional prefrontal or temporoparietal cortices (Garcia-Toro et al.2006). Furthermore, in those studies the stimulation was not focused over the prefrontal hypometabolism peak voxel as the functional data were not co-registered on individual MRIs.

In line with previous hypotheses (Fitzgerald et al. 2006 a; Pascual-Leone et al.1996), the present study shows that stimulation of BA 9-46 appeared to be more effective than sham rTMS. However, due to the small sample size, it was difficult to disentangle the effect related to BA 9-46 stimulation from the effect of stimulating a specific hypofunctional cluster or the effect of a combination of both, as 6/15 patients were treated with the left-guided modality, while, conversely, most of the left PET-guided patients (6/9) were stimulated over BA 9-46. Moreover, the number of responders in patients treated over BA 9-46 did not differ from that treated over other areas, with 8/15 patients treated over BA 9-46 being non-responders.

In addition to the small size of the sample in each subgroup of PET-guided treatment, other limitations and possible confounding factors need to be addressed. First, the comparison group was recruited before the patients in order to compare the first included patients' FDG PET images to those of healthy subjects and, thus, determine a target for stimulation. Although we aimed to constitute a healthy subject group with a wide age range, there was a significant age difference between the patient and comparison groups as the comparison group had to be the same for all comparisons. We cannot rule out the possibility that the hypofunctional clusters detected in older patients would not have been observed in a comparison with age-matched control subjects. However, this is unlikely as the age ranges were similar across patient subgroups, and as age was included as a confounding covariate in modelling the statistical comparisons to minimize this effect.

Second, as in other studies of single-subjects vs. a group (Ohta et al.2008), the statistical threshold used in the single-patient vs. healthy group analysis (p<0.05, uncorrected) was low compared with the p<0.001 threshold often used in group analyses, as we aimed to display enough hypometabolic clusters to choose a peak voxel within the range of the magnetic field. Nevertheless, two patients did not display hypofrontality, even at this threshold. We cannot preclude that choosing a more stringent threshold would have allowed selection of patients with more severe hypofrontality, which perhaps would have yielded stronger effects of PET-guided rTMS.

Third, the parameters used in this study (90% MT, 10 sessions) might be viewed as being in the low range. Indeed, some studies have underlined the better efficacy of higher stimulation intensities and prolonged treatment over several weeks (O'Reardon et al.2007). However, although low parameters have been found to yield inconsistent effects (Boutros et al.2002), the use of higher intensities or treatment durations has not always provided better results (Garcia-Toro et al.2001). Moreover, in the present study, the number of pulses per session (1600) was high (Gershon et al.2003) and, although the treatment duration was short (2 wk), the response rate in all the actively treated patients (59%) was higher than in many other studies (Ebmeier & Herrmann, 2008).

Fourth, parallel with rTMS, the patients were receiving various medications such as benzodiazepines, or antipsychotics, whose influence on cortical excitability might have blurred the effects of the rTMS treatment. More patients were treated with benzodiazepines in the PET-guided group (10/16, 62%) than in the standard group (7/18, 38%), which could have diminished the effects in the PET-guided group, and conversely, more antipsychotics were prescribed in the standard group (22% vs. 0%), which could have diminished the efficacy of stimulation in the standard group. However, the randomization groups only differed from each other at the trend level with regard to those medications, and given the high response rate in this study (59%), this medication effects seem unlikely. Conversely, it is also unlikely that the findings might result from an ‘add-on’ effect of concomitant medications, since such effects are controversial (Burt et al.2002; Bretlau et al.2008; Herwig et al.2007), and should be the same in all groups. It is also unlikely that a comorbid diagnosis of bipolar or anxiety disorders interfered with response to rTMS as these disorders were equally prevalent across subgroups.

Another limitation arises from the operator-dependent target determination method, in which the choice of the hypometabolism peak voxel must take into account the hypometabolism depth and the hemispheric level (Talairach z coordinate) in order to stimulate areas that are accessible to the magnetic field. However, this procedure was the same for the right- and left-sided targets, and performed by the same investigator blind to patient randomization. The measure of coil–cortex distance or the reconstruction of standard locations from previously determined motor cortex targets are also operator-dependent, although easily reproducible with Anatomist. Of note, the reconstructed standard locations corresponded to BA 8 and BA 6 in 61% of the patients, which is close to the 68% BA 8-6 in Herwig et al.'s study (2001), in which they used neuronavigation to reconstruct standard locations.

It is also possible that the accuracy of the TMS guidance and focality might have blurred the effects. However, this guidance method has an accuracy of 10 mm comparable to that of frameless stereotactic neuronavigating systems (Andoh et al. in press). The focality of the TMS figure-of-eight coil is theoretically 2–3 cm² at 110% intensity of MT, and may depend on stimulation intensity, coil–cortex distance, and hemisphere radius (Thielscher & Kammer, 2004). The size of the cortical area stimulated at 90% MT intensity, as in the present study, should be very small, <2.5 cm² according to Thielscher & Kammer (2004). Yet in a few patients stimulation of contiguous areas other than those targeted might have confounded the expected effects.

Finally, it is unlikely that the absence of a right-sided sham-treated group might have biased the results because the right-sided stimulated patients showed no significant improvement.

Conclusion

This study does not support the hypothesis of increased effectiveness of high-frequency rTMS when applied over hypometabolic prefrontal regions, irrespective of the side of stimulation. Stimulation of the DLPFC (BA 9 and/or 46) was more effective than sham, but failed to be an effective target area for stimulation in half of the patients. Finally, whether brain-imaging guidance over the left prefrontal cortex might speed up or increase the effects of rTMS in the treatment of resistant depression warrants further investigation.

Acknowledgements

The study was supported by a grant to Dr M.-L. Paillère-Martinot, from the ‘Programme Hospitalier de Recherche Clinique, Délégation à la Recherche Clinique de l'Assistance Publique-Hôpitaux de Paris’ (AP-HP) and the Health Ministry (PHRC/AOM-98099), a grant from the French Institute for Health and Medical Research (INSERM-PROGRES A99013LS), and an AP-HP/INSERM interface grant. Dr D. Ringuenet was supported by the ‘Fondation pour la Recherche Médicale’ (FRM) and the Atomic Energy Commission (CEA). The authors are grateful to Professor André Syrota, Frédéric Dollé, Dr Bernard Guéguen, Dr Jani Penttilä, Dr François Pinabel, Xavier Neveu and Edouard Duchesnay for their support.

Statement of Interest

None.

References