Effects of transcranial magnetic stimulation in modulating cortical excitability in patients with stroke: a systematic review and meta-analysis

Background Transcranial magnetic stimulation (TMS) has attracted plenty of attention as it has been proved to be effective in facilitating motor recovery in patients with stroke. The aim of this study was to systematically review the effects of repetitive TMS (rTMS) and theta burst stimulation (TBS) protocols in modulating cortical excitability after stroke. Methods A literature search was carried out using PubMed, Medline, EMBASE, CINAHL, and PEDro, to identify studies that investigated the effects of four rTMS protocols—low and high frequency rTMS, intermittent and continuous TBS, on TMS measures of cortical excitability in stroke. A random-effects model was used for all meta-analyses. Results Sixty-one studies were included in the current review. Low frequency rTMS was effective in decreasing individuals’ resting motor threshold and increasing the motor-evoked potential of the non-stimulated M1 (affected M1), while opposite effects occurred in the stimulated M1 (unaffected M1). High frequency rTMS enhanced the cortical excitability of the affected M1 alone. Intermittent TBS also showed superior effects in rebalancing bilateral excitability through increasing and decreasing excitability within the affected and unaffected M1, respectively. Due to the limited number of studies found, the effects of continuous TBS remained inconclusive. Motor impairment was significantly correlated with various forms of TMS measures. Conclusions Except for continuous TBS, it is evident that these protocols are effective in modulating cortical excitability in stroke. Current evidence does support the effects of inhibitory stimulation in enhancing the cortical excitability of the affected M1. Supplementary Information The online version contains supplementary material available at 10.1186/s12984-022-00999-4.

disinhibition is detectable, regardless of chronicity [1], which is believed to have a facilitatory role in motor recovery [2]. For instance, the effects of rehabilitation training correlate with reduced intracortical inhibition, reflected by reduced short-interval intracortical inhibition (SICI) and long-interval intracortical inhibition [3,4]. With respect to the cortical reorganization within the unaffected M1, as suggested by neuroimaging studies, the unaffected M1 in stroke patients became overactivated during movement execution of the affected hand [5]. Along with recovery of motor functions, stroke patients tend to regain the interhemispheric balance over the bilateral sensorimotor cortices [6]. However, meta-analyses found nonsignificant differences in a series of TMS measures, such as rMT, aMT and MEP amplitudes, compared to healthy controls [1].
In view of the association between cortical excitability and motor impairment of the hemiplegic arm, noninvasive brain modulation by repetitive TMS (rTMS) has attracted plenty of attention as it is effective in facilitating motor relearning and motor recovery after stroke. In accordance with the interhemispheric imbalance model, hemiparesis is caused not only by damaged corticospinal output from the affected M1, but also by excessive transcallosal inhibition from the unaffected to the affected M1 which can be measured by interhemispheric inhibition (IHI) and the ipsilateral silent period (iSP) [7]. Therefore, two unilateral modulatory approaches have been proposed-exciting or inhibiting the affected and unaffected M1-to counterbalance bilateral cortical excitability [8]. Early studies have revealed that both a single session of low frequency rTMS (LF-rTMS) and high frequency rTMS (HF-rTMS) were effective in improving the motor performance of the hemiparetic hand when they were applied to the unaffected and affected M1 [9,10], respectively. From a neurophysiological perspective, LF-rTMS was shown to significantly reduce the MEPs of the unaffected M1 and the IHI from the unaffected to affected M1 [9,10], while HF-rTMS had a direct facilitatory effect on the affected M1 [10].
Theta burst stimulation (TBS) is a unique form of rTMS that is usually delivered at subthreshold intensities over a short conditioning period [11,12]. In healthy people, intermittent TBS (iTBS) was shown to enhance cortical excitability outlasting the stimulation period by almost 30 min, while opposite effects were shown after continuous TBS (cTBS) [13]. In stroke, iTBS to the affected M1 and cTBS to the unaffected M1 were shown to increase and decrease the MEPs, respectively [14]. Clinical studies suggested that both iTBS and cTBS were effective in modulating modulate the cortical excitability in patients with acute stroke [15]; iTBS, but not cTBS, was also able to change the cortical excitability in patients with chronic stroke [14,16,17].
Since the early introduction of TMS in the treatment of stroke [9], the clinical effects of various rTMS protocols have been well reviewed [18], and LF-rTMS to the unaffected M1 was shown to be the most effective form of treatment [19]. However, previous meta-analyses and reviews have generally focused only on clinical effects; effects from a neurophysiological perspective were less systematically and statistically reviewed [18]. Because of the inextricable relationship between cortical reorganization and motor recovery, it is necessary to consider how rTMS modulates cortical excitability, and whether the interhemispheric imbalance model is a valid hypothesis underlying the two therapeutic approaches. Therefore, the current systematic review and meta-analysis were conducted to evaluate the effects of four forms of rTMS (namely, LF-rTMS, HF-rTMS, iTBS, and cTBS) on a range of TMS measures of cortical excitability, including rMT, aMT, MEPs, SICI, intracortical facilitation (ICF), and iSP of bilateral M1s. Moreover, not only the accumulated effects of multiple sessions of rTMS, but also the effects of a single session of stimulation were evaluated. We also summarize the correlation between cortical excitability and motor improvement after multiple sessions of stimulation.

Methods
This study was reported in accordance with the preferred reporting items for systematic reviews and meta-analysis statements [20].

Search strategy
A systematic literature search was conducted by the first author using the following electronic databases: PubMed, Medline, EMBASE, CINAHL, and PEDro. The keyword combination of "((stroke) OR cerebrovascular accident) OR (hemiparesis) OR (hemiplegia)) AND ((transcranial magnetic stimulation) OR (TMS) OR (rTMS))" was used for the literature search. A manual search was also conducted, including screening the reference lists of previous systematic reviews and searching for the same keywords in Google Scholar. The last search was conducted on Jun 5, 2021.

Selection criteria and data extraction
This study concerns the effects of both a single session and multiple sessions of rTMS. Therefore, the applied inclusion criteria were: (1) for studies investigating multiple sessions of rTMS, appropriate control groups must be employed; for studies investigating a single session of rTMS, a comparison between pre-rTMS and post-rTMS measures was indispensable; (2) studies Once the study selection was completed, two authors independently extracted relevant data which were entered into two customized forms. Any discrepancies regarding data extraction were resolved through discussion.

Quality assessment
A checklist proposed for critically appraising the quality of TMS procedures was used for all the studies [21]. Moreover, the Physiotherapy Evidence Database (PEDro) rating scale was used to appraise the general methodological quality of the studies with a parallelgroup design [22].

Data analysis
With respect to those studies ineligible for metaanalysis (e.g., skewed data), their main findings were qualitatively analyzed and integrated with the results of the meta-analyses. If there were two or less studies identified for a single analysis objective, we would not perform a meta-analysis but qualitatively described the results only. Since most studies investigating a single session of rTMS normalized the post-rTMS MEP amplitude to the pre-rTMS value, expressed as a percentage and SD, our meta-analyses computed a pooled mean of the percentage and its 95% confidence intervals (CI). Otherwise, absolute change scores were used to compute Hedges' g, which corrected the possible bias of the small sample sizes. The Higgins' I 2 statistic was used to check heterogeneity across studies; an I 2 value below 50% was considered to reflect low levels of heterogeneity, while an I 2 value above 50% indicated high levels of heterogeneity. Given that the characteristics among the included studies were not exactly identical, a random-effects model was used for all the metaanalyses [23]. Publication bias was statistically examined using Egger's linear regression test and by visually inspecting the funnel plot. The level of significance was set at p < 0.05 for all statistical analyses. Comprehensive Meta-analysis 3.0 software (Englewood, NJ, USA) was used to perform all the meta-analyses in the current study.
The methodological quality of the TMS procedures is presented in Supporting Information (Additional file 1: Tables S1 and S2). Overall, the characteristics of the participants, coil type, and intensity were well reported. However, medication, CNS drugs, coil orientation, current direction, pulse shape, time interval between MEPs, and subject attention were missing in most studies. In addition, only a few studies used a navigation system to monitor the real-time position of coils in relation to the head of the participants. The PEDro score indicated that the included studies with a parallel-group design had good general methodological quality (Table 3), with a mean score of 7.2 (SD = 1.18). The funnel plots used in the publication bias examination are presented in Supporting Information (Additional file 1: Figs. S1 to S17).

Low frequency rTMS to the unaffected M1
Pomeroy et al. applied one Hz rTMS to the affected M1, rather than to the unaffected M1, and found that MEP responses of the hemiparetic arm were more frequently evoked after receiving real rTMS than sham rTMS [46]. Recently, Wang et al. found that HF-rTMS applied to the unaffected M1 yielded greater improvement on the MEP amplitude than LF-rTMS, suggesting a compensatory role of the unaffected M1 in the motor recovery of severe-impaired patients [75].

Resting/active motor threshold
Previously, significant modulatory effects on rMT induced by a single session of LF-rTMS were not found  [24,28,36]. However, our meta-analysis has shown the significant effect of multiple sessions of LF-rTMS in decreasing the rMT of the affected M1 (Hedges' g = − 0.61, 95% CI = − 1.08 to − 0.14, p = 0.011, I 2 = 66.21%; Fig. 2), and no publication bias was identified (p = 0.187). It suggested that the decreased rMT could be sustained for three months after the intervention [50,62], while no significant long-term effects were also reported [58]. Likewise, multiple sessions of LF-rTMS also decreased the aMT of the affected M1 immediately [47,71].

Intracortical inhibition/facilitation
Despite the heterogeneity of the inter-stimulus intervals and intensities, four studies consistently found no significant changes in regard to the SICI [24,26,32,36], cortical silent period [32,36], and ICF [32] of the affected M1 after a single session of LF-rTMS. Similarly, no significant changes were found in the unaffected M1 [36,55,60]. Only one study conducted by Mello et al. revealed   that the ICF of the unaffected M1 was upregulated after 10 sessions of LF-rTMS in patients with stroke at the acute/subacute stage [60].

Transcallosal inhibition
A limited number of studies have suggested that iSP recorded in the affected hand reduced immediately after a single session [9,28] and 10 sessions [50] of LF-rTMS. For the unaffected hand, no significant effects on iSP were found [28,32]. By measuring IHI, decreased inhibition from the unaffected to the affected M1 was also noted [32].

Intracortical inhibition/facilitation
It was consistently found that a single session of HF-rTMS did not change the SICI or the ICF of the affected M1 [26,29,30].

Intermittent theta burst stimulation to the affected M1
The effects of a single session of iTBS were extensively studied [14-17, 27, 37-40], and four studies investigated the effects of multiple sessions of iTBS on cortical excitability [53,54,57,64,68]. iTBS was given at subthreshold intensities, and almost all the studies delivered 600 pulses, except the study by Hsu et al. in which patients received 1200 pulses [53].
For the effects of multiple sessions of stimulation, iTBS combined with physical therapy was beneficial to the rMT and aMT of the bilateral M1s [53,54,57,64]. It is noteworthy that the post-iTBS measurements were conducted one day after the last session of the intervention in two studies [53,64], rather than immediately after the stimulation in those studies investigating the effects of a single session of iTBS.
Largely inconsistent findings were reported regarding the MEP amplitudes of the bilateral M1s after multiple sessions of iTBS [53,54,57,64]. For the long-term effects, a previous study revealed that iTBS had superior effects in sustaining the increased cortical excitability in the affected M1 for almost three months [68].

Intracortical inhibition
Preliminary evidence did not support the modulatory effects of iTBS on SICI [38] and the cortical silent period of the affected M1 in patients with stroke [40].

Continuous theta burst stimulation to the unaffected M1
A study conducted by Di Lazzaro et al. suggested that a single session of cTBS to the unaffected M1 had similar effects as iTBS to the affected M1, simultaneously increasing the rMT and aMT of the unaffected M1 and decreasing those of the affected M1 [15]. However, another study, also conducted by Di Lazzaro et al. showed that 10 sessions of cTBS did not effectively modulate cortical excitability, reflected by the nonsignificant changes in the rMT, aMT, and MEP amplitudes of the bilateral M1s [52]. These nonsignificant findings were also confirmed by Neva et al. in which neither the SICI, the ICF, nor the iSP of the bilateral M1s benefited from cTBS [72].

Correlation between the changes in TMS measures and motor improvement
A total of 12 studies performed various correlation analyses between TMS measures and motor impairment [43, 48-50, 56, 57, 59, 61-63, 70, 71, 75]. Cross-sectional analyses indicated positive correlations between the cortical excitability of the affected M1 and the severity of motor impairment of the hemiplegic arm [61,63], while that of the unaffected M1 was negatively correlated with motor impairment of the hemiplegic arm [71]. Furthermore, motor improvement was significantly correlated with a reduction in the cortical excitability of the unaffected M1 [56,57,59]. The decrease in iSP from the unaffected to the affected M1 and the increase in MEP amplitudes in the affected M1 were correlated with arm motor improvement [50,62]. A previous study has also shown the predictive value of MEP amplitudes for subsequent motor improvement [48].
Two studies conducted these analyses in patients with severe arm impairments. Positive correlation between the cortical excitability of the unaffected M1 and arm motor impairment was noted after receiving LF-rTMS [49]. The changes in MEP latency in the unaffected M1 were negatively correlated with motor improvement [75].

Discussion
The aim of the current study was to systematically review the effects of four commonly used rTMS protocols on the cortical excitability of patients with a unilateral stroke lesion. LF-rTMS to the unaffected M1 is the most extensively investigated one in the literature. Our meta-analyses indicated that it was effective in increasing the MEP amplitude and decreasing the rMT of the affected M1, with opposite effects regarding the unaffected M1. Applying HF-rTMS to the affected M1 increased cortical excitability of the stimulated M1, while no significant effects were found on that of the  [1]. Most recently, a longitudinal study revealed that the premovement IHI did not significantly differ from that of healthy controls at the acute/subacute stage of stroke, however, the premovement IHI was abnormally higher at the chronic stage [79]. Therefore, the authors argued that the excessive IHI in chronic stroke patients may be a consequence of cortical reorganization, but it was not causally associated with motor impairment [79]. To the best of our knowledge, this study suggested that motor impairment at the acute/ subacute stage was not caused by excessive premovement. However, the study by Xu et al. [79] cannot rule out the possibility that the interhemispheric imbalance developed at the chronic stage, presumably caused by the maladaptation of cortical reorganization or learned non-use, may prevent the recovery of motor function, since many studies have supported the clinical effects of LF-rTMS to the unaffected M1 in chronic stroke patients. In our meta-analyses, we found that LF-rTMS to the unaffected M1 could significantly increase the cortical excitability of the affected M1, accompanied by decreased cortical excitability of the unaffected M1. A small number of studies also documented that reduced iSP and IHI after LF-rTMS were highly correlated with motor improvement [9,28,50], which would further reinforce the interhemispheric imbalance model. In a large-scale randomized controlled trial by Harvey et al. [80], the authors failed to find superior effects of LF-rTMS applied to the unaffected M1 against sham stimulation in improving poststroke upper limb motor functions. The failure to reject the null hypothesis points out the importance of studying the characteristics of treatment responders to LF-rTMS. Carey et al. found that the responders to LF-rTMS (applied to the unaffected M1) had greater hand function and greater preservation volume of the ipsilesional posterior limb of the internal capsule at baseline, compared with the nonresponders [81]. Besides, disrupting either the unaffected M1 or the unaffected dorsal pre-motor cortex by TMS worsened the contralateral motor performance in stroke patients with severe motor impairment [82], and that HF-rTMS applied to the unaffected M1 could facilitate motor improvement in stroke patients with severe motor impairment [75]. These studies suggested that the unaffected hemisphere plays different roles in motor recovery according to the severity of motor impairment [75,81,82]. Di Pino et al. proposed a bimodal balance-recovery model to explain the differential roles of the unaffected hemisphere in poststroke motor recovery [8]. Patients with mild motor impairment can benefit from inhibitory brain stimulation applied to the unaffected hemisphere due to high level of structural preservation, whereas the unaffected hemisphere, because of substantial damage to the ipsilesional corticospinal tract, has a significant role for compensation. Recently, Lin et al. found that IHI from the unaffected to affected M1 was negatively correlated with upper limb function in less impaired patients, while positive correlation was found in more impaired patients [83]. This supports further the bimodal balancerecovery model. In our current systematic review, only one study in which HF-rTMS was applied to the unaffected M1 in patients with severe motor impairment, and this study showed that improvement in upper limb function was correlated with shortened latency of unaffected MEP [75]. Because MEPs from the affected M1 was not detectable [75], it remained unknown whether the excitability of the affected M1 was altered after HF-rTMS applied to the unaffected M1. Regarding the mechanism, recent studies are still debating the pathway of the unaffected M1 in facilitating motor recovery via callosal connections, or direct pathway from the unaffected hemisphere [84,85]. Traditional measures of intracortical and interhemispheric excitability/inhibition rely highly on electromyography outputs, which are usually not detectable in patients with severe motor impairment. Alternatively, combining TMS and electroencephalography can directly probe the cortical reactivity after TMS pulses [86,87], which might be useful in the investigation of interhemispheric communication for those patients with severe motor impairments.
As aforementioned, the neurophysiological effects of LF-rTMS to the unaffected M1 can be explained by the mechanism of rebalancing interhemispheric inhibition. It is necessary to understand whether hyperactivity or excessive inhibition flow from the unaffected M1 is a prerequisite for LF-rTMS applied to the unaffected M1 aiming to modulate the cortical excitability of the affected M1. Healthy people are assumed to have balanced interhemispheric inhibition. However, many previous studies have found that LF-rTMS and cTBS not only suppressed the MEP amplitude of the simulated M1, but also enhanced that of the non-stimulated M1 [88,89]. Increased cortical excitability within the non-stimulated M1 is likely relevant to the elevated intrinsic excitability of the excitatory interneurons responsible for glutamatergic non-NMDA receptors [90]. Given the above findings in healthy people, it is reasonable to expect the cortical excitability of the affected M1 to benefit from inhibitory stimulation to the unaffected M1, even though the unaffected M1 is not hyperactivated or does not inhibit the affected M1 excessively during resting and premovement at the acute and subacute stages after stroke [1,79]. If this is the case, the interhemispheric imbalance model may not be the unique hypothesis of inhibitory stimulation to the unaffected M1, and future studies are warranted.
A large body of evidence suggests that the cortical excitability of the affected M1 decreased after stroke, and motor improvement is associated with the increase of excitability in the affected M1 [1]. Our meta-analyses indicated that both HF-rTMS and iTBS are useful in increasing the cortical excitability of the unaffected M1. Furthermore, the effect of simultaneously suppressing the unaffected M1 was noted after iTBS, but not after HF-rTMS. For a direct comparison, a previous study has shown the superior effects of iTBS in increasing the MEPs of the stimulated M1, compared to HF-rTMS in healthy people [88], but comparable effects were also reported [91]. Animal studies suggested that different TMS protocols may have specifically different effects in modulating neurogenesis and protein expression, which may potentially account for the different effects across neurophysiological and clinical measures [92].
In addition to unilateral-hemispheric stimulation, some new protocols integrating these regular forms of rTMS have been studied in stroke. For instance, the dual-hemisphere stimulation consisting of inhibitory rTMS to the unaffected M1 and excitatory rTMS to the affected M1 was more effective in enhancing motor performance and cortical excitability [74] and reducing the SICI of the affected M1 than unilateral-hemispheric stimulation [26]. Another protocol, LF-rTMS primed with HF-rTMS, also showed encouraging effects in reducing intracortical inhibition within the affected M1 [32]. Besides, functional rTMS triggered by electromyogram may induce greater excitability changes than passive stimulation protocols [29].

Limitations
This review was not free from limitations. First, we must be cautious when interpreting the findings relevant to LF-rTMS. The high IHI elicited by paired-pulse TMS was found during the premovement period but the majority of previous TMS measures were conducted at rest. Therefore, the outcomes-paired-pulse induced IHI at rest and single-pulse induced iSP during sustained isometric contractions, in our review might not be conclusive in explaining excessive inhibition driving from the unaffected M1 [89]. Second, substantial heterogeneity and publication bias were identified in some meta-analyses, probably due to the small sample sizes in most included studies and the way in which the patient characteristics, TMS protocols, and methodologies were not identical. Third, the available studies mostly focused on the effects on TMS measures, which may limit our findings to patients with mild to moderate impairments only. This is because the MEPs of the affected M1 are usually not recordable in stroke patients with severe motor impairments.

Conclusions
The current study systematically reviewed existing research investigating the effects of four forms of rTMS in modulating the cortical excitability of bilateral M1s. LF-rTMS to the unaffected M1 is the most extensively studied protocol, while cTBS is the least studied one. Although recent studies have argued for the rationale of inhibitory stimulation applied to the unaffected M1, our analyses suggested that LF-rTMS not only suppressed the cortical excitability of the unaffected M1 but also simultaneously enhanced that of the affected M1. HF-rTMS enhanced the cortical excitability of the affected M1 only. Preliminary evidence also supported the effects of iTBS in rebalancing bilateral cortical excitability in stroke. Our findings support the bimodal balance-recovery model in patients with mild motor impairment, more studies are needed to investigate the neurophysiological effects of HF-rTMS applied to the unaffected M1 in patients with severe motor impairment.