Using Brain Imaging to Improve Spatial Targeting of Transcranial Magnetic Stimulation for Depression

Major depressive disorder affects approximately 1 in 5 individuals and 350 million people worldwide and is the leading cause of years lived with disability (). As few as 30% of patients achieve remission with first-line therapies (). Individuals who fail multiple treatments are classified as having treatment-resistant depression, and these patients are unlikely to respond to further medication trials (). Repetitive transcranial magnetic stimulation (rTMS) is an established, safe, and efficacious treatment approved by the U.S. Food and Drug Administration that can alleviate symptoms in patients with treatment-resistant depression (). It involves focal magnetic stimulation applied externally to the scalp, typically at the dorsolateral prefrontal cortex (DLPFC), which induces electrical stimulation in underlying cortical tissue. The most commonly used clinical TMS coils are figure eight coils, which have a relatively focal stimulation field and will be the focus of this review. However, less focal TMS coils exist, an important topic that is addressed in the Supplement.

While rTMS is effective for some individuals, many others with similar clinical profiles receive little benefit. Typical response rates, defined as >50% reduction in depression score, and remission rates, defined as a post-TMS depression score below the level that qualifies for depression, are typically between 29% and 46% (response) and 18% and 31% (remission) (,). Although therapeutic response is clinically meaningful, particularly in a treatment-refractory population, remission should be considered the ultimate goal. Research efforts over the past 2 decades have aimed to improve the efficacy and consistency of treatment effects across individuals. While optimal stimulation parameters and dosing are important, one major source of interindividual heterogeneity in treatment outcomes arises from variation in the stimulated site across the spatial extent of the DLPFC (, , , , , , , , , , , , ).

Therapeutic targets for the treatment of depression using rTMS have been informed and guided by neuroimaging since its clinical inception in the mid-1990s (Figure 1). Early studies of patients with stroke and tumors suggested that the risk of depression increased with left prefrontal lesions [for further discussion see (,,)]. Functional neuroimaging studies in primary depression reported hypometabolism in the left prefrontal cortex that improved with successful antidepressant treatment (). As such, early rTMS studies targeted the left DLPFC commensurate with our neuroanatomical knowledge of depression (,).

Over time, it became clear that the left DLPFC is highly heterogeneous, and response rates may depend on exactly where in the DLPFC one administers rTMS (, , ,,). Lesion and metabolic neuroimaging studies failed to replicate the simple association between the left DLPFC and depression (,). More importantly, psychiatric disorders began to be conceptualized as disorders of brain networks, not individual brain regions (, , , , ). Similarly, rTMS began to be conceptualized as a network therapy—although stimulation is commonly applied to a single brain region, its effects are mediated via distributed networks (, , , ). Advances in mapping brain networks and brain connectivity now allow us to identify these networks and potentially refine our therapeutic targets for depression (, , , ,,,). This review describes the evolution of therapeutic targeting strategies in depression and new developments in this area.

The first clinical investigations of rTMS for depression identified the DLPFC target site as being 5 cm anterior to the motor cortical hotspot (,), overlying Brodmann area (BA) 46 and BA 9 in the Talairach atlas (,). This 5-cm method was subsequently used in larger clinical trials that led to Food and Drug Administration approval (, , ). However, this approach does not account for differences in head dimensions or anatomy, leading to stimulation of the premotor cortex or frontal eye fields in a large percentage of patients (,,,). Many clinical centers therefore adopted 5.5 or 6 cm to move the average stimulation site more anterior and lateral (,,,). Recent work estimates that the intersection of BA 9 and BA 46 is actually 6.9 cm anterior to the motor hotspot, but this has not been adopted clinically (). It is also perhaps worth noting that the areal delineation of BA 9 and BA 46 has been redefined several times (, , ) since originally parcellated by Brodmann in 1909 () and is highly variable across individuals (). The 5- to 6-cm approach has been the most commonly employed targeting method, accounting for 84% of randomized clinical trials as of 2016 (5 cm: 75%; 6 cm: 9%) [computed from supplemental information in ()].

A newer targeting approach based on the 10-20 electroencephalography system has been proposed to account for variation in individual skull dimensions. Coregistration of electroencephalography electrode positions with anatomical magnetic resonance imaging (MRI) suggests that the electrode position 1 cm anterolateral to F3 (), between F3 and F5 (), or between AF3 and F3 would reliably target the DLPFC (Figure 2) (). Software has since been developed to estimate the F3 electrode position based on only a few scalp measurements (Beam F3) (,). This approach has seen widespread clinical implementation, is located with high reliability (), and is endorsed by the Clinical TMS Society (). However, this targeting approach has not yet been validated in double-blinded randomized trials as the 5- to 6-cm method has. Further, although this technique ensures that the DLPFC is more consistently targeted (,,), and the more anterolateral sites of the DLPFC selected by this approach are understood to be more effective (Figure S2) (), clear gains in antidepressant efficacy remain to be demonstrated [e.g., (,)]. Another point of interest is the only marginal extent of spatial overlap between targets derived from the 5.5-cm and Beam F3 targeting methods (Figure 2). Given this spatial difference, future work should directly compare these targeting methods to determine if there are differences in antidepressant response or side effects.

Targeting TMS based on scalp-based measures is reasonable if the target is rather broadly defined as the left DLPFC. However, as our knowledge regarding the neuroanatomy of depression increases, more accurate targeting of TMS is warranted. In one of the first examples of this approach, Fitzgerald et al. () conducted a meta-analysis of functional imaging studies of depression to identify a precise coordinate in the left DLPFC that was most consistently abnormal. The authors then used neuronavigated TMS to target this coordinate in patients with depression. Neuronavigation enables the TMS coil to be positioned to target specific anatomical sites based on an individual subject’s structural brain images. Both the neuronavigated and the conventional 5-cm approaches yielded reductions in depression severity; however, there was no significant interaction between targeting approach and clinical trajectory across the two cohorts, nor was there clear decrease in response variability across individuals (). Nonetheless, the neuronavigated approach appeared to be more effective at a trend level, resulting in an overall reduction in symptom severity of approximately 49% compared with the 27% observed with the 5-cm approach [recomputed from the absolute values of clinical scores presented in ()]. Other studies targeting this anatomical location found it to be better than sham conditions () [although see ()], but likewise failed to demonstrate superiority relative to conventional scalp-based targeting (,). Interestingly, adjusting stimulus intensity to account for individual scalp-to-cortex distance did not enhance the effects of anatomical targeting (). This might be because while TMS effects are often dependent on intensity, cortical thresholds or reactivity may differ or even be unrelated across brain regions (, , ), presenting a potential caveat for current clinical practice. In any case, these studies do not provide clear support for the superiority of TMS targeting this anatomical location for depression, possibly because these studies were underpowered, the optimal anatomical target differs from that employed above (), or anatomical targeting does not address individual variation in DLPFC functional organization (,, , ).

Other groups have tried to improve TMS for depression by targeting individual differences in metabolism [for review of metabolic abnormalities see (,,)]. In one such study, the site of maximum hypometabolism in the DLPFC was first localized using a positron emission tomography scan for each individual and then targeted using high-frequency rTMS. However, neither this study () nor other similar studies (,) were able to demonstrate superiority of this approach relative to conventional targeting approaches. Possible reasons for the failure of this approach include poor spatial resolution of positron emission tomography imaging (), unknown reproducibility of the targets, or the fact that individual foci of prefrontal hypometabolism are not the best antidepressant target. Interestingly, the hypometabolic maximum was located in the right DLPFC in 33% () to 73% () of individuals, which challenges the traditional valence hypotheses of mood lateralization in depression, in line with other recent studies (,,, , , ).

Finally, at least 1 trial of TMS for depression has targeted individualized sites of functional MRI (fMRI) activation. This trial used an n-back working memory task to identify individualized TMS targets for each patient, but failed to show a difference in antidepressant response or imaging biomarkers relative to sham ().

Localization of psychiatric symptoms, including depression, has gradually shifted from a focus on individual brain regions, such as the DLPFC, to a focus on distributed brain networks (,,,, , ). For example, lesion locations associated with depression are not solely located in the left DLPFC, but instead map to a distributed brain network that is centered on the left DLPFC (). Similarly, recent network models of depression include a variety of cortical and subcortical brain regions (). In parallel, it has become clear that the effects of rTMS are not restricted to the stimulated region, but propagate to affect the network of regions connected to the stimulation site (, , ,,, , , ,,).

One valuable neuroimaging technique for visualizing brain networks is resting-state functional connectivity (FC) MRI (,). Brain regions are interconnected to form intrinsic networks, characterized by shared temporal fluctuations in spontaneous brain activity. These distributed networks can be delineated, even in the absence of external stimuli, and are therefore termed resting-state networks. One advantage of recording these networks at rest is that there is a reduced need for patient compliance and avoidance of confounds related to task performance or instructions (). Resting-state networks may also exhibit higher reproducibility compared with conventional task-based imaging (). The FC within and between these networks is often altered in psychiatric and neurological conditions (,,), while partial normalization may occur following successful treatment (,), including brain stimulation (, , ,). It is now clear that different regions across the spatial extent of the DLPFC map on to different distributed brain networks (), potentially explaining a portion of the response variability associated with conventional TMS targeting techniques (Figure S1).

The shift in focus from brain regions to brain networks has motivated various new approaches to improve TMS for depression. These include the use of TMS to modify connectivity, with the goal of correcting network abnormalities in depression (, , ,,,). Another is the use of brain connectivity to predict which patients will respond to TMS treatment (,, , ). Finally, brain connectivity has been used to identify rTMS targets for depression based on the connectivity of the stimulated target to other brain regions (, , ,,).

An example of the latter approach relates to the observation that antidepressant outcomes were better when stimulation was serendipitously delivered at sites of the DLPFC that displayed stronger negative (anticorrelated) FC with the subgenual cingulate cortex (SGC) (Figure 3) (). The SGC is a region positioned at the anterior-inferior end of the cingulum bundle with extensive connections across prefrontal and limbic structures that have been implicated in depression (). It is associated with abnormal emotional regulation and processing and has been linked to depression and clinical response across diverse antidepressant treatment modalities (, , ).

The association between DLPFC-SGC FC at the stimulation site and treatment response has additionally been replicated across 3 geographically distinct clinical cohorts, with findings robust across different populations, methodologies, scanners, stimulators, and DLPFC targeting approaches (5.5-cm, cognitive activation, Beam F3) (,). More specifically, a 60% to 70% reduction in depressive symptoms occurred when individuals were stimulated near the DLPFC site of maximal FC anticorrelation with SGC, while those stimulated farther away showed no response or worsening of depressive symptoms [c.f. Figure 1C in ()]. The association between SGC FC and treatment response is also unique to individuals receiving active and not sham stimulation ().

Based on these associations, it has been suggested that the DLPFC site most anticorrelated with the SGC could represent an optimal TMS target for depression (, , ). The DLPFC coordinates of maximal SGC-FC anticorrelation are at Montreal Neurological Institute coordinates x = −42, y = 44, z = 30 (, , ). In line with this reasoning, a recent large randomized trial of TMS used neuronavigation to target this coordinate with the goal of optimizing antidepressant response (). However, the goal of this trial was to compare two different forms of active rTMS, not to validate this target. Whether neuronavigated TMS to this target results in stronger or more consistent antidepressant response compared with conventional scalp-based targeting remains to be tested in a dedicated clinical trial. It also remains unclear whether connectivity to the SGC is the only or the most important connection for defining an optimal TMS target. Connectivity of the TMS target to other brain regions implicated in depression may also be important (,,).

The above TMS target based on SGC connectivity may not be optimal for all patients owing to individual differences in brain connectivity. The above target was based on group connectivity, averaged across 1000 healthy individuals. This averaging allows for robust maps that help counteract the low signal-to-noise ratio of the SGC region (); however, it ignores potentially important individual differences in connectivity.

Many have suggested that rTMS might be improved by personalizing the stimulation site based on individual differences in connectivity (,,,,,,, , , , ). Recent work demonstrates that SGC FC shows considerable interindividual variation across the spatial extent of the DLPFC (,,). Indeed, prefrontal regions show some of the highest levels of interindividual variation in terms of cytoarchitecture, structural morphology, neural function, and connectivity (,, , ,, , ). However, these individual differences are likely to prove useful as TMS targets only if one can overcome signal-to-noise limitations of single-subject fMRI data ().

The past decade has seen substantial advances in fMRI acquisition, preprocessing, and noise-reduction strategies () as well as in our capacity to model the relationship between stimulation site, network engagement, and treatment response (described further in the Supplement). With sufficient data, clear individual differences in functional network architecture are evident (Figure S1) (,,, , ); however, several of these studies used many hours of MRI scanning, which is not practical for clinical patients. Further, limbic areas such as the SGC are particularly prone to signal-to-noise problems, limiting our ability to identify robust TMS targets (). In fact, recent work indicated that single-subject TMS targets derived from individual connectivity to a small region of interest in the SGC are not reproducible enough to provide an advantage over group-based connectivity (,). This signal-to-noise hurdle may be overcome through creative strategies to obtain reproducible individualized targets. In one such strategy, connectivity to a network of limbic regions, rather than just the subgenual alone, resulted in robust individualized targets (), findings that have since been replicated and extended (,).

A number of small clinical trials have begun using SGC-FC TMS targets derived from individualized connectivity (Table 1). Some of these trials have reported very high response and remission rates (,,,), but whether personalization exceeds the efficacy of conventional targeting methods remains to be established in a dedicated clinical trial. It is important to note that these studies have been performed in small cohorts and typically without a comparison target or sham group. It is also possible that the placebo effect of rTMS is greater when additional technologies such as MRI scanning and neuronavigation are used. Future research with larger trials and comparator groups is warranted. These trials would further benefit from validation of the accuracy and reproducibility of the individualized targeting method [e.g., as per ()].

The TMS target based on SGC connectivity may not be the best target for all depression symptoms (Figure 4A). Emerging evidence suggests that separate symptom clusters might respond to stimulation of different brain circuits (,,,). Anterolateral DLPFC sites with anticorrelated FC to the SGC may be more effective in moderating dysphoric symptoms such as sadness, decreased interest, and suicidality, whereas posterior DLPFC and medial prefrontal targets appear more effective for ameliorating anxiosomatic symptoms such as insomnia, decreased libido, and irritability (Figure 4B) (). Different depression symptoms may therefore benefit from different TMS targets.

Thus far, our review has focused on high-frequency rTMS to the left DLPFC; however, other TMS sites and parameters have been selected for depression. The most common alternative is low-frequency TMS to the right DLPFC, which has produced comparable clinical outcomes to high-frequency left DLPFC TMS in randomized trials (,, , ). Traditionally, high- and low-frequency rTMS paradigms have been delivered with the aim of increasing or decreasing cortical activity and potentially normalizing metabolic abnormalities in depression. However, whether 1-Hz rTMS reduces cortical activity (when measured in terms of regional cerebral blood flow, etc.) remains uncertain (,, , , , , ) and may depend on stimulation intensity, flipping from inhibitory to excitatory at higher intensities (,, , , ) [for a discussion of the physiological mechanisms mediating this phenomenon see ()]. A similar phenomenon has been described with continuous theta burst stimulation, which elicits excitatory, rather than inhibitory, effects at higher stimulus intensities (), as typically employed clinically (). Consistent with this observation, studies comparing the effects of low- and high-frequency rTMS to the left DLPFC have often reported similar antidepressant efficacy (, , , , , ), with high-frequency stimulation generally delivering more robust effects (,). In sum, the assumption that rTMS normalizes an imbalance in cortical activity between the left and right DLPFC may be too simplistic. Connectivity-based studies of TMS sites in the right DLPFC, similar to those reported here for the left DLPFC, may shed light on whether the same connections mediate rTMS response in the two hemispheres. It is worth noting that high- and low-frequency rTMS do not implicitly evoke opposite changes in FC; indeed, the effects on connectivity explicitly depend on stimulation site and frequency and are difficult to anticipate prior to experimentation (, , , , ,,).