Positron emission tomography imaging relies on detecting and localization of a radionuclide used to label a molecule (e.g., receptor/transporter agonists or antagonists, enzymatic precursors) to target a specific neurotransmitter system. Various radiotracers can be used to monitor specific proteins and neurochemicals processes in the living brain non-invasively. PET-rTMS studies have been heavily focused on evaluating stimulation effects on endogenous dopamine. The majority of such studies use [¹¹C]raclopride, a D_2/3 receptor antagonist (Laruelle, 2000), as a surrogate marker for dopamine level quantification in the striatum. Evaluation of extrastriatal dopamine release can also be assessed using higher-affinity receptor radioligands like [¹⁸F]fallypride (Hernaus and Mehta, 2016) and [¹¹C]PHNO (Wilson et al., 2005).

Positron emission tomography imaging has also been used to evaluate rTMS-induced perturbations in other neurotransmitters such as serotonin and endogenous opioids. Serotonin imaging is enabled by tracer such as [¹¹C]αMtrp synthetic analog of the essential amino acid precursor to 5HT, L-tryptophan (Trp), which provides a proxy measure for serotonin synthesis capacity in vivo. Imaging of the endogenous opioid system can be monitored using [¹¹C]carfentanil as a high-affinity μ opioid receptor antagonist.

While understanding rTMS-driven neurotransmitters release is an essential tool to understand the short- and long-term mechanisms of action of the non-invasive stimulation therapies, other tracers also permit measuring and quantifying other processes, such as inflammation or synaptic density.

Magnetic resonance spectroscopy allows to investigate concentrations of metabolites in vivo non-invasively and can thus can be used as a piece of surrogate information to infer neurotransmitter and neurochemical changes. Usually, MRS is acquired from a volume of interest containing several voxels at a time (1 cm³ is often considered adequate size). Its acquired spectral components from N-acetyl aspartate (NAA), total NAA (tNAA), a combination of NAA and NAA glutamate (NAAG), creatine (Cr), phosphocreatine (PCr), choline, myoinositol, and the combined signal of glutamate and glutamine (Glx).

Under specific acquisition and analysis conditions, it is also possible to record spectral components of glutamate and glutamine separately and gamma-aminobutyric acid (GABA). MRS measurements of GABA and glutamate concentration can be used to monitor inhibitory and excitatory neurotransmission, while the other metabolites present in the MRS spectra have been used as markers of other neurochemical processes.

Quantification of GABA is technically challenging due to its low concentration (Govindaraju et al., 2000) and spectral overlap with other metabolites that are more abundant (Puts and Edden, 2012). One possible method to separate the GABA spectra is to use scanners with higher field strength, such as 7Tesla (Gröhn et al., 2019). Alternatively, spectral editing and two-dimensional MRS methods can be used when higher field scanners are unavailable. Spectral editing methods such as MEGA-PRESS are used to reduce the information in the spectrum (Buonocore and Maddock, 2015), while two-dimensional MRS methods can spread the overlapping signals (Ke et al., 2000).

Glutamate (Glu) spectra overlap with glutamine (Gln) spectra, making it simpler to report information from both spectra together as Glx. Modifying echo times to shorter ones can improve the detection of Glu (Wilson et al., 2019), as done with sequences such as SPECIAL and sLASER (Öz et al., 2021).

Positron emission tomography evaluation of neural activity is enabled by measuring alterations in [¹⁸F]FDG, a glucose analog, and flow and oxygen metabolism using ¹⁵OH₂O and ¹⁵O radiotracers. As neural activity changes, changes in glucose consumption and metabolism are reflected, leading to an increase or decrease in the uptake and trapping of [¹⁸F]FDG. PET- ¹⁵OH₂O imaging is used to measure regional cerebral blood flow (rCBF), coupled with neural activity, unless there are severe pathological conditions where uncoupling may occur (Raichle and Mintun, 2006).

Another helpful imaging tool to evaluate neural activity is fMRI. Similar to PET- O¹⁵H₂O, this technique can monitor changes in blood flow through the measurement of the BOLD signal and change in oxygenation as an indirect measure of neural activity. When acquired at rest, fMRI can identify the correlation of spontaneous fluctuations in brain activity among spatially distributed brain regions; this analysis is defined as resting-state functional connectivity (rsFC) (Biswal et al., 2010; Smitha et al., 2017). In rTMS studies, rsFC evaluation is useful as it allows for the assessment of rTMS effects without accounting for connectivity modulation concomitant to a task or state.

Regions that show statistical dependence in BOLD signal fluctuation are known as resting-state networks (Smith et al., 2009; Doucet et al., 2011; Hamada et al., 2013; Smitha et al., 2017). These networks can be identified using several analysis techniques, seed-based analysis being the first method adopted in early work by Biswal et al. (1995). Such analysis finds the linear correlation of a pre-selected region (a seed region) with all the other voxels in the brain. While this is advantageous due to its simplicity and interpretability, its results are heavily influenced by the seed region selected.

Later techniques have allowed hypothesis-free analysis of rs-fMRI data; one of such techniques is independent component analysis (ICA) (Beckmann et al., 2005; Damoiseaux et al., 2006; Bi et al., 2018) which allows for the distinction of networks through decomposition of the signal from all brain voxels to temporal and spatial independent components. In addition, techniques such as k-means clustering (Thomas Yeo et al., 2011; Zhang and Li, 2012; Drysdale et al., 2017), graph theory (Lord et al., 2012; Khazaee et al., 2015), and others (for review see Khosla et al. (2019)), have expanded the available toolbox of hypothesis-free rs-fMRI analysis methods.

Diffusion tensor imaging can evaluate structural integrity and structural changes in white matter. It allows the acquisition of diffusion water maps in the brain and provides quantitative information through diffusion coefficients (Alexander et al., 2007; Westlye et al., 2010; Hawco et al., 2018). Diffusion maps provide microstructural information that allows the detection of changes in white matter and can also be used to evaluate alterations in neuroanatomy due to aging (Poudel et al., 2015; Lebel and Deoni, 2018), disease (Melzer et al., 2013; Poudel et al., 2015; Visser et al., 2019) or interventions (Coenen et al., 2014; Pi et al., 2019).

Fractional anisotropy (FA) measurements described the directionality of random water diffusion (Pierpaoli and Basser, 1996). FA values range from 0 (isotropic) to 1 (anisotropic), where values closer to 0 represent water molecules that diffuse freely in any direction or are equally restricted in all directions (e.g., cerebral fluid) and values closer to 1 represent water diffusion occurring in a single direction (e.g., when contained in fiber bundles). Regarding white matter FA estimates, higher values are observed in heavily myelinated tracks, given that the myelin restricts the diffusion of water molecules to the direction along the fiber tracks (Beaulieu, 2002). Another commonly used DTI measurement is mean diffusivity (MD) representing the average mobility of water molecules where higher values of MD indicate a more isotropic movement.

Diffusion tensor imaging can provide information on the local white matter bundle orientation, which can be used to reconstruct white matter connections (Lin et al., 2011) and structural connectivity (Chiang and Haneef, 2014). DTI tractography analyzed using graph theory methods can evaluate fiber count as a measurement connectivity strength (Caeyenberghs et al., 2012). Specifically, metrics to evaluate global or local connectivity such as characteristic path length, global efficiency, strength, and modularity can be obtained (Kocevar et al., 2016; Yu et al., 2018). Path length is a measurement of the information integration processing, an increase in this measurement represents a bigger number of steps for brain regions to communicate with each other. Global efficiency is a proportional inverse of path length, and it provides information about the network’s parallel information processing efficiency. Finally, modularity and strength are measurements that evaluate the network segregation and the weight of the connectivity across the regions of a network (Girvan and Newman, 2002; Baum et al., 2017).

Positron emission tomography studies of rTMS effects have primarily focused on evaluating dopamine modulation, partly because of the relevance of the dopaminergic system and the availability of tracers for dopaminergic evaluation. However, a few PET studies have evaluated the effect of rTMS on serotonin and endogenous opioid release.

An early pilot study used [¹¹C]-Alpha-methyl-tryptophan ([¹¹C]-αMtrp), a synthetic analog of the serotonin precursor, and measured “trapping” as a proxy measurement for or serotonin synthesis capacity in vivo. Their results suggested that HF-rTMS delivered over the left DLPFC could modulate serotonin metabolism and trapping in parts of the limbic cortex of healthy subjects right after a single stimulation session (Sibon et al., 2007). These results are in line with reported results from SPECT imaging, where binding of [¹²³I]-5-I-R91150, a 5-HT_2A receptor antagonist decreased following 10 sessions of rTMS delivery (Baeken et al., 2011). These changes in serotonin release correlated with symptomatic improvement, suggesting that HF-rTMS therapeutic effects might be associated with modulation of serotonin release and metabolism.

A study by Lamusuo et al. (2017) evaluated the effects of HF-rTMS in both the dopaminergic and opioid systems. While they found no dopamine release changes hours after the end of HF-rTMS, the evaluation using an agonist tracer of the μ opiate receptor ([¹¹C]carfentanil) continued to show significant changes more than 5 h after the end of the stimulations.

Although there is limited literature on the effects of rTMS on neurotransmitters other than dopamine, these PET and SPECT studies highlight the complex multi-neurotransmitter effect of rTMS. The effects of rTMS on dopamine release seem to be immediate and transient, given the lack of effects after several hours (Kuroda et al., 2010; Lamusuo et al., 2017). In contrast, studies of the serotonergic system reported immediate [right after stimulation (Sibon et al., 2007)] and longer-lasting effects [a week after 10 daily sessions (Baeken et al., 2011)]. Finally, there also seems to be an effect of rTMS in the opioid system, which lasts longer than the effect in the dopaminergic system (Lamusuo et al., 2017). These studies confirm that different neurotransmission systems have a unique response to rTMS, with modulation lasting from less than a few hours to several days.

Neuropharmacological imaging is limited to a few neurotransmitters and changes in the two transmitters that may present the most interest based on their ubiquitous distribution, GABA and glutamate, are not easily measurable by PET or SPECT. Combination with other modalities, as described below, may, however, permit one to have a glimpse into their role.

Excitatory rTMS modulates neurotransmitter levels in healthy humans. Iwabuchi et al. (2017) showed a decrease in GABA/Glx ratio in the targeted region (left DLPFC) and the anterior cingulate cortex after iTBS delivery. An early HF-rTMS recorded a decrease in Glx in the stimulated left DLPFC and an increase in the contralateral region (Michael et al., 2003). Interestingly, Glx changes in the left DLPFC negatively correlated with baseline levels, suggesting that baseline Glx might predict rTMS induced Glx changes.

Studies in healthy subjects have reported a relation between brain connectivity and changes in metabolite levels following excitatory rTMS. Iwabuchi et al. (2017) showed that the GABA/Glx ratio decreased in the right DLPFC and was associated with decreased connectivity between the DLPFC and right anterior insula in response to iTBS over the left DLPFC. Another study reported a relation between an increase in GABA/tCr in the posterior cingulate and its connectivity and the inferior parietal lobe (targeted area) after iTBS delivery (Vidal-Piñeiro et al., 2015). These results illustrate the propagation of rTMS effects from the proximal, stimulated area to distal anatomically and functionally connected regions and that GABA and Glx measurements have a significant relationship with the large-scale network changes resulting from rTMS delivery.

Stagg et al. (2009) reported an increase in GABA/NAA ratio in the stimulated motor cortex following a single session of cTBS but could not detect any changes in Glx, possibly due to limited measurement sensitivity. A later cTBS study also reported an increase in GABA concentration in the primary visual cortex, accompanied by a decreased cortical excitability (Allen et al., 2014). Similarly, LF-rTMS also led to increased GABA in the stimulated left motor cortex, accompanied by a GABA decrease in the contralateral M1 (Gröhn et al., 2019), observable after 20 min post-stimulation. These results suggest that LF-rTMS and cTBS stimulation might activate GABAergic interneurons, which in turn modulate excitability in the cortex. Table 2 summarizes MRS findings from healthy participants.

Early PET studies in healthy subjects showed that both HF- (Paus et al., 1997) and LF-rTMS (Fox et al., 1997) increased rCBF. Similar results were replicated in later studies (Paus et al., 2000; Siebner et al., 2000, 2001b; Speer et al., 2003b; Takano et al., 2004; Knoch et al., 2006).

Early work also suggested the rTMS effect over rCBF to be dose, frequency, and intensity-dependent. More specifically, it was shown that a more significant number of pulses (Paus et al., 1997, 1998), a higher rTMS frequency (Siebner et al., 2001b; Knoch et al., 2006), and a more vigorous stimulation intensity (Speer et al., 2003a,b) induced greater rCBF change. In addition, work from Knoch et al. (2006) suggested rCBF response to rTMS might also vary depending on the stimulated hemisphere. They showed that rTMS over the prefrontal cortex yields activity modulation in different distal regions based on the hemisphere stimulated.

Low-frequency-rTMS modulation of metabolism in the targeted area has been reported to be heterogeneous, with studies reporting both increase (Siebner et al., 1998, 1999, 2001b; Cho et al., 2012; Lee et al., 2013) and decrease (Cho et al., 2012; Lee et al., 2013) after LF-rTMS. HF-rTMS in healthy participants has been only shown to increase local glucose metabolism (Siebner et al., 2000).

Changes in rCBF and glucose metabolism have been mainly localized within the stimulated networks (Paus et al., 1998; Siebner et al., 2001a; Speer et al., 2003b; Takano et al., 2004). Delivery over the motor cortex both increases and decreases activity in the contralateral motor cortex, premotor areas, ACC, cerebellum, and striatum (Paus et al., 1998; Siebner et al., 1998, 2001a; Speer et al., 2003b). Stimulation in the prefrontal cortex modulated rCBF in cognitive regions (Paus et al., 2000; Kimbrell et al., 2002; Speer et al., 2003a; Knoch et al., 2006). Glucose metabolism studies also showed that LF-rTMS delivery decreases activity in the cerebellum and induces a widespread activity modulation in the prefrontal, parietal, and temporal cortex (Cho et al., 2012).

Evaluation of rsFC has been used to reveal the neural mechanisms underlying rTMS effects. In healthy controls, studies have evaluated the effects of HF, LF, and inhibitory and excitatory TBS and QPS delivered mostly over the prefrontal and motor cortices.

The effects of rTMS in rsFC are dependent on the number of stimuli delivered. Nettekoven et al. (2014) found iTBS delivery over the left motor cortex increased connectivity between the stimulated area and the ipsilateral dorsolateral premotor cortex. The increase in connectivity was higher after 1800 pulses compared to changes after 600 and 1200 pulses. An increase in connectivity was also observed between the motor cortex and bilateral areas of the somatosensory and superior parietal cortex, but changes were not dose-dependent, suggesting that the dose-dependent effect might be exclusive to some areas within the stimulated network.

Stimulation of the DLPFC induces a consistent modulation of rsFC across subjects in the anterior cingulate. In a large study carried on 60 healthy participants, Tik et al. (2017) recorded an increase in rsFC in the anterior cingulate cortex after HF-rTMS. These results were supported by a study from a different group, where similar results were observed (Xue et al., 2017).

Bilateral modulation of rsFC has also been reported after QPS delivery to the motor cortex (Watanabe et al., 2014). Excitatory QPS (Interstimulus intervals of 5 ms) decreased rsFC between the stimulated left motor cortex and the contralateral motor cortex, while inhibitory QPS (interstimulus intervals of 50 ms) increased it, illustrating the bidirectional effects of rTMS in rsFC.

Other studies have also reported differences in rsFC response to inhibitory and excitatory rTMS over the motor cortex. Eldaief et al. (2011) found that HF rTMS decreased functional connectivity within nodes of the default mode network when delivered over the posterior inferior parietal lobule, while LF increased connectivity between the targeted area and regions outside the network. A different study reported similar effects after delivery of HF- and LF-rTMS over the medial frontopolar cortex, where authors reported an increase (after HF-rTMS) and decreased (after LF-rTMS) in connectivity between the target area and the amygdala (Riedel et al., 2019). Adding to these findings, Zhang et al. (2020) reported changes in connectivity observed after HF to be more widespread, while LF-rTMS modulated connectivity mostly found in regions within the stimulated hemisphere. This suggests that HF and LF rTMS not only can induce different directional effects but also distinct topological changes.

Similarly, modulation of the motor network with LF rTMS and cTBS, 2 inhibitory forms of TMS stimulation differing by their frequency (1 Hz vs. 50 Hz), has been shown to induce distinct directional effects. Stimulation of the supplementary motor area using LFr-rTMS increased rsFC within the motor network, while cTBS decreased it. Nevertheless, these two stimulation paradigms had similar topographical and temporal effects, with effects being observed after 6 to 10 min post-stimulation (Ji et al., 2017). Similarly, cTBS delivery over the left somatosensory cortex decreased connectivity within the motor network (Mastropasqua et al., 2014; Valchev et al., 2015). Contrary to cTBS studies, HF studies have failed to induce changes in rsFC after stimulation of the supplementary motor area (Hendrikse et al., 2020).

Finally, a cTBS study by Mancini et al. (2017) suggested that recorded changes in functional connectivity might depend on the time they are evaluated. They measured rsFC at 5 and 15 min post stimulation and reported a decreased functional connectivity of the temporal lobe at the 5 min and a change in precuneus connectivity at 15 min.

Diffusion tensor imaging evaluations have reported FA and MD values modulation. In a study by Kozel et al. (2011), authors found 4–6 weeks of 10 Hz stimulation over the left prefrontal cortex increased FA prefrontal white matter. Similarly, white matter FA in the right superior frontal gyri has been reported to significantly increase after HF-rTMS delivery over the left DLPFC (Tateishi et al., 2019).

Similar to results from functional connectivity studies, a DTI study showed pre-rTMS structural changes in white matter tracks related to therapeutic improvement. A study by Fu et al. (2021) showed baseline FA in fiber tracts from the DLPFC to the insula correlated with symptom improvement in mood disorders patients. The authors found the same correlation when comparing pre-rTMS functional connectivity of the DLPFC and the insula and therapeutic effects. These results illustrate the close relation between fMRI and DTI derived measurements, and more importantly, the similarity between functional and structural connectivity changes induced by rTMS delivery. In a different study, baseline DTI evaluation found reduced white matter FA in the left middle frontal gyrus compared to healthy controls, suggesting abnormalities of white matter in this region might be part of mood disorders pathology (Peng et al., 2012). Authors also reported that 4 weeks of HF-rTMS over the left DLPFC increased FA and induced therapeutic effects; these changes in FA were correlated with therapeutic improvement. Taken together, these results illustrate rTMS capacity to modulate structural changes in regions related to the mood disorders’ pathophysiology and suggest such modulation might facilitate rTMS therapeutic effects.

High-frequency-rTMS has been shown to enhance FA values as soon as 4 days after stroke onset. In a recent clinical trial, the treatment group (those with subcortical stroke) received 10 days of 5 Hz rTMS over the ipsilesional motor cortex. Following the 10 days of rTMS, the treatment group showed increased FA in the contralesional corticospinal tract, the pontine crossing tract, the middle cerebellar peduncle, the contralesional superior cerebellar peduncle, the contralesional medial lemniscus, and the ipsilesional inferior cerebellar peduncle compared to the sham group (Li et al., 2018b). Measures of MD in the superior longitudinal fasciculus and the uncinate fasciculus can be enhanced following a combination of acupuncture therapy and LF-rTMS over the contralesional motor cortex (Zhao et al., 2018). Another study showed that HF- rTMS (10 Hz) to ipsilesional motor cortex enhanced FA values of motor-related gray matter, which correlated with decreased motor impairment. These increases in FA in motor-related white matter were greater than those in a conventional physical therapy group that did not receive HF-rTMS (Guo et al., 2016). Changes in FA values along with symptom improvement suggest rTMS induced recovery of the motor-related system and white matter connections to a greater extent than usual therapy alone.

Yamada et al. (2018) showed that 15 daily LF-rTMS and occupational therapy improved motor function and observed an increase of generalized FA in the white matter of lesioned motor cortex. Similarly, Yang et al. (2015) reported changes in FA following chronic rTMS delivery. Specifically, the authors compared the therapeutic and structural effects of different rTMS frequencies (1 Hz, 10 Hz, and cTBS) in stroke patients with unilateral spatial effect: subjects who received cTBS delivery showed the greatest therapeutic improvement which was accompanied by an increase in FA and MD in the white matter of frontal fasciculus and the capsula interna and externa. Therapeutic effects were also observed after 1 and 10 Hz stimulation, but no changes in FA and MD were reported.

In individuals with aphasia after stroke, five sessions of iTBS applied to the affected hemisphere on separate days showed increases in FA values in several regions of the brain previously shown to have increased fMRI activation during language activation after iTBS (Allendorfer et al., 2012). Specifically, FA was enhanced in inferior and superior frontal gyri, anterior corpus callosum, the right midbrain, and bilaterally near temporal, parietal and posterior cingulate regions in those with aphasia due to stroke.

Diffusion tensor imaging analyzed using graph theory analysis showed that 12 sessions of LF-rTMS over the motor cortex combined with occupational therapy increased path length in the stimulated network in stroke patients (Ueda et al., 2019). These findings suggest rTMS delivery led to more multifaceted neural circuits and, as mentioned by the authors, these new connections might be a compensatory mechanism to cover for the impaired motor paths due to stroke. Graph-theoretical analysis was also used to evaluate DTI measurements before and after rTMS delivery over the left DLPFC in children with cerebral palsy. Zhang et al. (2021) reported an increase in the clustering coefficient of the frontal gyrus and the pallidum, and a decrease of the path length of the right putamen and thalamus, suggesting rTMS stimulation might have improved the efficacy of information exchange through the striatal thalamic projections. These results indicate rTMS delivery modulation of structural connectivity within the corticostriatal pathway and, along with results reported by Ueda et al. (2019), illustrate the complexity of the rTMS effect in brain structure, integrity, and connectivity.

Even though DTI studies of rTMS effects are limited, they show the ability of rTMS to induce changes in white matter microstructure. Several studies have reported modulation of FA values, changes that are thought to represent microstructural changes of white matter tracks (Li et al., 2018b). Through graph theory analysis, studies were also able to measure changes in structural networks and found LF-rTMS might lead the generation of more complex neural circuits, which might ultimately indicate an increase in microstructural connections across the evaluated network. The studies summarized here illustrate the value of DTI-rTMS studies and the ability of rTMS delivery to induce microstructural changes in brain connectivity.