Imaging TMS: Antidepressant mechanisms and treatment optimization

Abstract

With the antidepressant efficacy of Transcranial Magnetic Stimulation well-established by several meta-analyses, there is growing interest in its mechanism of action.

TMS has been shown to engage, and in some cases, normalize functional connectivity and neurotransmitter levels within networks dysfunctional in the depressed state.

In this review, I will suggest candidate biomarkers, based on neuroimaging, that may be predictive of response to TMS. I will then review the effects of TMS on networks and neurotransmitter systems involved in depression. Throughout, I will also discuss how our current understanding of response prediction and network engagement may be used to personalize treatment and optimize its efficacy.

Introduction

Since its first use for the treatment of major depression, transcranial magnetic stimulation (TMS) has targeted frontal lobe dysfunction (George & Wassermann, 1994). Choosing to stimulate the left dorsolateral prefrontal cortex (DLPFC) came from the theory of hypofrontality in depression, which was supported by evidence from PET-imaging (Buchsbaum et al., 1986; Martinot et al., 1990) and post-stroke depression (Starkstein & Robinson, 1989). Indeed, early case reports found that stimulating this target was effective for treating depression (George et al., 1995; Pascual-Leone, Rubio, Pallardó, & Catalá, 1996), leading to verification in open label studies and rigorous sham controlled trials (George et al., 2010; 1997; O’Reardon et al., 2007). Meta-analyses have well-established the efficacy of TMS targeting this region (Schutter, 2009; Slotema, Blom, Hoek, & Sommer, 2010).

Early studies into the antidepressant mechanisms of TMS demonstrated patterns of frontal lobe perfusion at baseline that correlated with TMS response and that TMS modulated frontal perfusion (Speer et al., 2009; 2000; Teneback et al., 1999). More recent work has focused on circuit-based mechanisms of emotional regulation, reward processing, anhedonia and psychomotor behavior in depression. Although biomarkers have been found, and to some extent replicated, the utility of these markers awaits prospective TMS trials in which depressed subjects are selected for the presence or absence of the biomarker. First, this will provide more robust evidence TMS engages a circuit known to be abnormal. Second, over time, it will help us determine if personalizing treatment by pairing individuals who have the “appropriate predictive biomarker with the appropriate target” leads to better response rates.

This review will highlight some of the advances in using neuroimaging to understand antidepressant mechanisms of TMS. This will not be a comprehensive review of those advances as excellent reviews have already been published (Noda et al., 2015; Silverstein et al., 2015). I will try to provide a critical review of studies using resting state fMRI and neurochemical imaging modalities, to identify gaps in understanding and to identify directions and hypotheses for future research. Although bipolar depression and depression associated with dual diagnosis are clinically important areas with high morbidity and in which TMS may play an important role, I will limit this review to major depressive disorder.

Baseline functional connectivity

The depressed state is characterized by elevated functional connectivity of the default mode network (DMN), a collection of midline frontal and parietal areas as well as lateral parietal structures. This network becomes active and oscillates at 0.1Hz when the subject is doing nothing and becomes inactive during periods of purposeful behavior (Fox & Raichle, 2007). The network can be assayed using functional MRI (fMRI) and the strength of active connections between areas determined. In depression, this “functional connectivity” is elevated (Greicius et al., 2007; Sheline et al., 2009; Sheline, Price, Yan, & Mintun, 2010), the degree to which has been correlated with rumination scores (Zhu et al., 2012). This “hyperconnectivity”, has been highly replicated across studies (Northoff, 2016a). Hyperconnectivity of the DMN has been shown to improve after electroconvulsive therapy (ECT) (Perrin et al., 2012) and with serotonin-norepinephrine reuptake inhibitors (van Wingen et al., 2014).

Personalized medicine aims to discover reliable measurements, which can be taken from individual patients, that will be able to predict the likelihood that a patient will respond to a variety of prospective treatments (Kapur, Phillips, & Insel, 2012). In the quest for such predictive biomarkers of TMS for depression, several investigators have evaluated differences in baseline, pre-TMS resting state fMRIs of patients who subsequently either responded or did not respond to a course of TMS. Three studies in individuals with TRD had convergent findings. Stimulation targeting the dorsomedial prefrontal cortex (DMPFC) led to better response in patients with higher baseline functional connectivity between DMPFC and sgACC and between sgACC and DLPFC and lower baseline functional connectivity of cortico-thalamic, cortico-striatal and cortico-limbic projections (Salomons et al., 2013). A second study demonstrated similar findings using a left DLPFC target (Liston et al., 2014). This study found that higher functional connectivity between the sgACC and DLPFC, DMPFC, VMPFC, mOFC and bilateral posterior parietal cortex was associated with treatment response. A third group replicated the predictive value of sgACC-DLPFC hyperconnectivity for response to 10Hz TMS over the left DLPFC (Baeken et al., 2014).

The strength of functional connectivity of the sgACC, a structure repeatedly found to be overactive and hypermetabolic in depression (Fu, Steiner, & Costafreda, 2013; Mayberg et al., 1997), to dorsomedial and dorsolateral prefrontal cortex, is a candidate predictive biomarker for response to TMS (Baeken et al., 2014; Liston et al., 2014; Salomons et al., 2013). These studies average measurements of functional connectivity across groups of patients, while being able to say less about the individual patient. However, the case toward using baseline functional connectivity of the individual patient as a predictive biomarker was strengthened by a pair of studies by Fox and colleagues. In the first, this group compared treatment efficacy of previously published studies of TMS for depression that used left DLPFC targets that were localized using different methods. Using a set of normative resting state functional connectivity data, the group found that the best antidepressant response across studies was associated with increasing functional connectivity between the target location and the left sgACC (Fox, Buckner, White, Greicius, & Pascual-Leone, 2012). These results were then reproduced in a small sample of depressed patients. In the second study, the group showed the feasibility of targeting the region within the left DLPFC with maximal functional connectivity to the sgACC by demonstrating that functional connectivity between these two regions was stable between subsequent days (Fox, Liu, & Pascual-Leone, 2013). Additionally, the optimal TMS target for depression was strongly functionally connected to the sgACC stimulation site for deep brain stimulation (Mayberg et al., 2005). This relationship between targets of noninvasive and deep brain stimulation held true for several neuropsychiatric disorders (Fox et al., 2014).

Plasticity of functional connectivity

Several studies, including those introduced in the previous section, have demonstrated that TMS normalizes cortical circuits with abnormal baseline functional connectivity in depression. Studies from our lab showed that daily 10Hz TMS over the left DLPFC for 25 days normalized connectivity between the sgACC and several nodes of the DMN and the cognitive executive network (CEN) (Liston et al., 2014). There were no connectivity changes between the left DLPFC and the CEN. Although there was no correlation demonstrated between the size of the change in connectivity and the degree of improvement of depressive symptoms, this study was not powered to test for this. A second study using the same left DLPFC stimulation site and parameters led to similar changes in functional connectivity. However, change in connectivity was specific to treatment responders (Baeken et al., 2014). Responder-specific findings in the latter study may have been attributable to the medication washout required in the Baeken study which may have made subjects more sensitive to both clinical and neurophysiological effects of TMS.

The study by Salomons et al used a different stimulation site, the DMPFC, a similar sample size of 25 patients, on stable medication and a 4 week course of 10Hz TMS. As in the previous two studies, connectivity of the sgACC was reduced after TMS. Reduction of sgACC connectivity to the mid-cingulate, caudate, and insula was correlated with treatment response. Additionally, increases in dmPFC-thalamus connectivity correlated with improvement in depression. The authors suggest that TMS may be normalizing deficits in cognitive control, although this clinical outcome was not measured explicitly (Salomons et al., 2013).

The studies using similar methods of open label TMS in depressed cohorts, provide convergent evidence that TMS treatment is associated with normalization of the DMN, a self-monitoring network involved in the negative ruminative state of depression. Two of the 3 studies also demonstrate (to different degrees) that the cognitive executive network (CEN) and the interactions of the CEN and DMN also normalize with TMS. Resting state fMRI is easy to acquire and thus may be particularly suitable as a biomarker for predicting and tracking treatment response. However, future research should also address how changes in specific cognitive domains may be mediating the antidepressant effects of TMS by employing RDoC (Insel et al., 2010) and capturing behavioral change within these domains during a course of TMS.

Although fMRI provides some mechanistic insight into the effects of TMS on neuroplasticity, the BOLD signal used in fMRI is based on blood oxygenation, while the physiologic underpinnings are at the neuronal signaling level (Logothetis, Pauls, Augath, Trinath, & Oeltermann, 2001). Cerebral blood flow change and functional connectivity have been found to correlate with power in the alpha band of the EEG and quantitative EEG methods have been developed to study connectivity (Leuchter, Cook, Hunter, Cai, & Horvath, 2012). It has been suggested that the changes in functional connectivity observed with BOLD may be generated by “entraining” cortico-cortical and cortical-subcortical networks to a frequency which helps to “reset” the oscillation frequency of the network and restore its normal function (Leuchter, Cook, Jin, & Phillips, 2013; Leuchter, Hunter, Krantz, & Cook, 2015b). Future studies of the mechanism of action should incorporate pre/post EEG to correlate BOLD findings with frequency entrainment. Attempts have been made to “personalize” the stimulus to the patient’s individual alpha frequency, with mixed results (Leuchter et al., 2015a).

Functional connectivity is, in part, determined by structural connectivity between neural regions (Anderson, Hoy, Daskalakis, & Fitzgerald, 2016; Gong & He, 2015). Abnormalities in white matter structure may disrupt signal propagation beyond the stimulation site to more distributed networks. For example, the effects that occur “at a distance” in the sgACC with DLPFC stimulation and in downstream networks depend on physical, axonal pathways. Preliminary data from my lab suggests that abnormalities in the structural connectome, as measured by graph-theoretic nodal degree and clustering coefficient predict response of depression to TMS (Deng et al., 2016).

Monoamine systems

As in the case of psychopharmacology, studies in both animals and humans implicate neurotransmitter systems in the antidepressant mechanisms of TMS. These include the monoamine neurotransmitter systems, dopamine and serotonin, as well as the amino acid neurotransmitter systems, GABA and glutamate.

Dopamine depletion is known to correlate with depression (Hamon & Blier, 2013; Savitz & Drevets, 2013) and dopamine reuptake inhibitors may produce their antidepressant effects by increasing the availability of dopamine at striatal synapses (Argyelán et al., 2005). Studies in healthy human subjects (Strafella, Paus, Barrett, & Dagher, 2001) and in subjects with depression (Pogarell et al., 2006; 2007) using radiotracers to the D2 receptor show that TMS stimulating the left DLPFC results in increased dopamine release in the striatum. This effect has also been observed in the rat dorsolateral striatum and nucleus accumbens (Kanno, Matsumoto, Togashi, Yoshioka, & Mano, 2004; Zangen & Hyodo, 2002). As dopamine signaling in the striatum is important for reward processing, which is compromised in anhedonia, and for motor control, which is impaired in psychomotor symptoms of depression (Grace, 2016; L. K. Tremblay et al., 2005), it is possible that this mechanism of dopamine enhancement may play a role in the antidepressant mechanism of TMS. Work on whether functional connectivity between the frontal stimulation site and the striatum is predictive of antidepressant response is mixed, with one study using DMPFC as the target showing decreased connectivity predictive of response (Salomons et al., 2013), while another using DLPFC as the target showing increased connectivity predictive of response (Avissar, Powell, Casey, Liston, & Dubin, 2015), suggesting “access” to the striatum may be necessary for treatment response.

Similarly, TMS over the left DLPFC at 10Hz modulates serotonin release in healthy volunteers throughout the limbic system, including in the cingulate gyrus and cuneus, parahippocampal gyrus and insula (Sibon et al., 2007). TMS over prefrontal targets in rat led to increases in serotonin release in hippocampus suggesting there is some overlap in mechanism between TMS and monoamine reuptake inhibitors (Juckel, Mendlin, & Jacobs, 1999; Levkovitz, Grisaru, & Segal, 2001). However, serotonin modulation by TMS has yet to be demonstrated for individuals with depression.

If monoamine neurotransmitters are demonstrated to play an important role in the antidepressant mechanism of TMS, this would raise several questions about overlapping mechanisms between TMS and medication treatments. For example, is treatment resistance due to different mechanisms of monoamine release? Does TMS enhance monoamine release through direct neuronal stimulation instead of reuptake inhibition or monoamine oxidase inhibition? Such PET findings would also open another avenue to target TMS by choosing a site that maximizes downstream dopamine release.

GABA and Glutamate systems

Several lines of evidence implicate GABA, the primary inhibitory neurotransmitter in the brain, in the pathophysiology of depression. MR-Spectroscopy reveals low GABA levels in the limbic system in severe depression in both adults (Croarkin, Levinson, & Daskalakis, 2011; Sanacora, Mason GF, Rothman DL, et al, 1999) and adolescents. In the sgACC, GABA levels have correlated inversely with anhedonia symptoms in adolescents (Gabbay et al., 2012). Levels of glutamate decarboxylase (GAD) enzymes, which synthesize GABA, are low in post-mortem brains of individuals who suffered depression (Hasler et al., 2007) as are GABAergic interneuron counts (Maciag et al., 2010). Single and paired pulse TMS-based assays of cortical inhibition have also been used to probe GABA_A and GABA_B receptor activity and have found that depression severity correlates with level of GABA signaling deficit (Levinson et al., 2010). GABA has also been implicated in animal models of learned hopelessness suggesting depression may result from an inhibitory/excitatory imbalance (Yizhar et al., 2011).

GABA homeostasis may be related to other biomarkers of depression. For example, GABA levels in the medial PFC have been found to correlate with reductions in functional connectivity of the default mode network (Northoff et al., 2007), raising the question that these two biomarkers of depression may be causally related (Northoff, 2016b; Northoff & Sibille, 2014). Additionally, multiple antidepressant treatments have been shown to elevate (or normalize) GABA levels, including ECT (Sanacora et al., 2003) and SSRIs (Sanacora, Mason, Rothman, & Krystal, 2002). More recent MR-spectroscopy results from ketamine studies add to the evidence that GABA modulation seems to be a final common pathway of psychopharmacological and neuromodulation treatments of depression (Milak et al., 2015). Further, GABA’s intermediate mechanism of action may be to modulate abnormal circuit connectivity in depression.

TMS targeting the left DLPFC at 10Hz has been shown to increase GABA in the MPFC in treatment-resistant depression in adults (Dubin et al., 2016). Although TMS in this study was open label, the effect was limited to treatment responders. Although baseline GABA level was not predictive of treatment response, this study was not powered to test this hypothesis. Glutamate and glutamine metabolism is also dysregulated in depression (Sanacora, Treccani, & Popoli, 2012) although this has been more difficult to study, given technical limitations of MR spectroscopy of separating the resonances of these two neurotransmitters. TMS has been found to increase glutamate concentration in the MPFC in normals (Michael et al., 2003). However, results after TMS for depression have been less conclusive, with one study showing a lack of effect on glutamate in the MPFC in depressed adults (Dubin et al., 2016) and another showing elevation in the glutamine/glutamate ratio immediately after a course of TMS and again at 6 month follow up in both DLPFC and MPFC (Croarkin et al., 2016), an effect in which the Gln/Glu ratio change was correlated with symptomatic improvement.

In summary, there is evidence that TMS over the left DLPFC modulates the GABA and glutamate systems and this modulation has been correlated with response to treatment for depression. Studies in this area have been limited by being open-label, by using the standard treatment protocol and by using MR spectroscopy with its necessary selection of specific a priori regions of interest. Future studies could benefit from addressing these shortcomings, as well as using correlation with TMS-based neurophysiological measures, in the same patients (S. Tremblay et al., 2013). Larger studies may be able to determine the predictive value of neurotransmitter levels for treatment response. Finally, maps of neurotransmitter levels (for example, GABA levels) may be coupled with functional or structural connectivity maps to determine if targeting surface-convexity regions that project to regions of GABA deficit optimizes treatment response.

Second-Generation Targeting Strategies

How can we apply our current understanding of the aspects of neural structure and function that predict outcome of TMS to optimize and personalize treatment for depression? One treatment optimization strategy is to focus on the temporal domain, including number of sessions, length and spacing of individual sessions and temporal structure of the stimulus, including frequency and inter-stimulus train intervals. Another strategy is to optimize the spatial targeting of the stimulus so as to most effectively change abnormally functioning neural circuits. One approach has been to design magnetic coils that penetrate to deeper neural structures; for example, the H-Coil, which can penetrate to 5cm and directly stimulate at that depth (Levkovitz et al., 2009). However, this coil does so at the expense of focality, which may explain why it is not more clinically effective for depression than superficially stimulating TMS (Levkovitz et al., 2015). This suggests that taking a “surgical” approach to engaging the appropriate circuits instead of a “shotgun” approach, is important for treatment response.

Refining this targeted approach will be an iterative process as more is learned about how best to access downstream circuits from the cortical surface. However, as reviewed here, enough is currently known to hypothesize and test potential targets. For example, DLPFC targets that we know are optimally connected to the sgACC could be studied in sham-controlled and head-to-head studies with currently-accepted targets. A treatment protocol would start with a resting state fMRI, which would be immediately analyzed online for connectivity to the sgACC (the striatum could be another candidate “deep target”). Resting state fMRI in this patient may reveal that sgACC-DLPFC connectivity is normal, or even lower than normal, which is a poor predictor of response according to published studies (Baeken et al., 2014; Liston et al., 2014). However, connectivity between the DMPFC and sgACC may be abnormally high in this patient, which would indicate that this patient may have a higher likelihood of response to TMS targeting the DMPFC (Salomons et al., 2013).