Located mainly in L2/3, L5a, and L6a, the IT neurons constitute approximately 80% of the PN population (Bakken et al., 2021; Zhang et al., 2021). In the human motor cortex, the relative proportion of IT neurons is significantly larger than that in mice and marmosets (Bakken et al., 2021), suggesting the essential contribution of IT neurons in the control of human movement. IT neurons can be divided according to the projection area into corticofugal projection neurons (projecting to subcortical structures in both hemispheres), associative projection neurons (projecting to other cortical areas in the ipsilateral hemisphere), and commissural projection neurons (projecting to homologous cortical area in the contralateral hemisphere) (Fame et al., 2011; Greig et al., 2013; Sohur et al., 2014). The L2/3 cortical IT neurons in the motor cortex are mostly commissural and associative projection neurons, which interconnect by synapses, receive input from other cortical areas, and project to the bilateral cortex and striatum or target the L5 extratelencephalic projecting (ET) and IT neurons as a pivot in the sensorimotor integration (Mao et al., 2011; Hooks et al., 2013; Lin et al., 2018). IT neurons in L5a and L6a are mostly corticofugal projecting, with different developmental and functional properties from L2/3 IT neurons (Greig et al., 2013). In the human temporal cortex, the L5 IT neurons display tuftless dendritic morphology and nearby projection property (Kalmbach et al., 2021), whereas the projection of L6 IT neurons in the human motor cortex remains largely unknown. Together, L2–6 IT neurons send their projections within the telencephalon, targeting the bilateral neocortex, striatum, and subcortical structures and, in particular, the contralateral neocortex (Brown and Hestrin, 2009). L2/3 commissural projecting IT neurons of the motor cortex project through the corpus callosum and other white matter (WM) commissures (e.g., the anterior commissure), forming axodendritic synapses at all cortical layers (yet dominating in L2/3) of the contralateral hemisphere (Lin et al., 2018; Tasic et al., 2018).

Extratelencephalic projecting neurons, also known as the pyramidal tract neurons (PTN) in the M1, concentrate in L5b and contribute to the CST. In primates, the quantity of ET neurons is approximately eightfold lower than that of L2/3 IT neurons. With a more diffuse projection than the IT neurons, ET neurons target not only the telencephalon but also the ipsilateral subthalamic nucleus, brainstem, and spinal cord (Kita and Kita, 2012; Callaway et al., 2021). Importantly, ET neurons contain no interhemispheric projection (projection to the contralateral telencephalon), as proven by the retrograde tracing study by Reiner et al. (2003). However, in the rodent motor cortex, ET neurons received interhemispheric monosynaptic excitatory input from contralateral motor-related areas (with weak input from somatosensory areas), as reported in a retrograde labeling study (Mao et al., 2011; Hooks et al., 2013). While the input to L5 ET neurons in the human motor cortex is not clearly addressed, it is shown that ET neurons can be excited by receiving integrated signal from multiple IT neurons in both hemispheres, although the signal propagation is irreversible due to the nature of synapse connection. That is, whereas IT neurons can excite ET neurons, ET neurons cannot excite IT neurons in reverse. As a subset of L5 ET neurons, the anatomically and physiologically specialized Betz cells were also found in the premotor cortex (PM) and supplementary motor area (SMA) of the human brain with similar transcriptomic definition (Rivara et al., 2003; Bakken et al., 2021; Takata et al., 2021), consistent with the reported origin from the premotor cortex and the supplementary motor area of the CST (Rouiller et al., 1996; Seo and Jang, 2013). Collectively, L5 ET neurons in the motor cortex receive multiple inputs from both hemispheres and have a wide-range projection to cortical and subcortical areas. Additionally, L5 ET neurons in the motor-related areas (M1, PM, and SMA) share cellular similarity and may have close internal interactions so as to function as the cortical motor network.

Corticothalamic neurons project to the thalamus exclusively from L5a/L6a of the neocortex while connecting with IT neurons in the nearby lamina (i.e., L5a and L6a IT neurons) (Behrens et al., 2003; Kim et al., 2014; Yamawaki and Shepherd, 2015; Callaway et al., 2021). In the multiple circuit-analysis research by Yamawaki and Shepherd (2015), it was observed that the CT–IT connection presented only in L6, but not in L2/3. Although CT may interact with PTN through disynaptic circuitry involving an inhibitory IN, the direct connection of CT-PTN is almost absent, despite the vicinity of L6a, L5a, and L5b. However, in the mouse barrel cortex, CT neurons are reported to receive strong and focused innervation from L4 PN (Tanaka et al., 2011; Qi and Feldmeyer, 2016); however, whether this innervation exists in the motor cortex remains unknown. Moreover, despite the thalamus acting as a pivot of both the ascending and descending pathways and processing both afferent and efferent information (McFarland and Haber, 2002; Behrens et al., 2003), thalamocortical projections have almost no interaction with the projections of CT neurons at the ventrolateral nucleus (Yamawaki and Shepherd, 2015), consistent with the structural model of the cortico-thalamo-cortical loop (for a recent review, see Shepherd and Yamawaki, 2021). In terms of interhemispheric connectivity, while Molinari et al. (1985) reported bilateral components of the CT fibers from the motor cortex in rats and cats, a recent cell census proposed by the Brain Initiative Cell Census Network (BICCN) suggested that CT projections contain no interhemispheric components (Callaway et al., 2021). For interaction with INs, L6 CT neurons innervate local parvalbumin-expressing (PV⁺) and somatostatin-expressing (Sst⁺) INs, resulting in general inhibition in superficial layers (Baker et al., 2018). With a lower amount than IT neurons, the CT neurons with distinct neural properties are considered the third major subclass of neocortical principal neurons after IT and ET.

The L5/6 NP neurons and L6b neurons are mostly defined by transcriptomic signatures (Tasic et al., 2018; Langseth et al., 2021); their physiological properties are largely unknown. The L5/6 NP neurons, populating about 4% of the entire neocortical PNs, have no long-distance projections like the IT, ET, and CT neurons (Tasic et al., 2018; Bakken et al., 2021). Moreover, the L6b neurons (comprising 5% of total PN population) in the mouse visual cortex have been observed to send projections to the thalamus or anterior cingulate cortex (Tasic et al., 2018). However, there is no evidence reporting the physiology and function of the two subsets of principal neurons in the human neocortex owing to the difficulty in assessing activities in deep cortical layers in vivo. A viable investigation method analyzing L6b and NP neuron activity is needed to unveil their connectivity and physiological roles in the neocortical neural circuits.

The fast-spiking PV⁺ IN is the most common and well-studied neocortical inhibitory IN, distributed in L2–L6 of the neocortex (Bakken et al., 2021). PV⁺ INs stem from MGE, and can be morphologically divided into two subtypes: the basket cell (BC, greater quantity) and the chandelier cell (ChC, smaller quantity). Both BCs and ChCs have fast-spiking electrophysiological properties, yet the firing pattern (latency, frequency, etc.) differed between the two subtypes (Povysheva et al., 2013). For innervation target on the PN, it was reported that BCs targeted the cell soma and proximal dendrites, whereas ChCs innervated the axon initial segment (AIS) of the PN (Szentagothai and Arbib, 1974; Inan and Anderson, 2014).

Contrary to the common inhibitory effects elicited by BC, the postsynaptic effects elicited by ChCs are more intriguing. A series of rodent and human studies reported that ChCs received input from the PN of the ipsilateral and contralateral hemispheres (Lu et al., 2017). Woodruff et al. (2011) performed a patch-clamp study and reported that L2/3 ChCs were activated by the electrical stimulation of L1 (where most of the ChC dendrites reside), with different timing-dependent feedforward effects of both inhibition (ChC activation 5 ms prior to PN activation) and facilitation (ChC activation 15–30 ms prior to PN activation). Consistent with this study (Woodruff et al., 2011), ChCs were reported to excite postsynaptic PNs by depolarizing the AIS under certain postsynaptic membrane potential states (Szabadics et al., 2006; Woodruff et al., 2009, 2011), despite its classification as “GABAergic inhibitory IN.” In the developed neocortex, the existence of axo-axonic ChC provides an important strategy to optimize PN output via GABA_A-receptor-induced excitation, as a specific GABA_A receptor (GABA_AR-α2) is selectively enriched in the AIS of neocortical PNs (Gulledge and Stuart, 2003; Howard et al., 2005; Szabadics et al., 2006; Gao and Heldt, 2016).

Nevertheless, it should be noted that neocortical BCs stem from two different origins of CGE and MGE (Marin-Padilla, 1972; Butt et al., 2005; Gelman et al., 2011), with those from CGE expressing cholecystokinin (CCK), a neuropeptide of serotonin 3A Receptor (5HT_3AR) family. PV⁺ BCs exceed CCK⁺ BCs in both quantity and size (Wang et al., 2002), and BCs are further classified as nest basket cells (mainly PV⁺), large basket cells (mainly PV⁺), and small basket cells (mainly CCK⁺) (for reviews, see Druga, 2009; Armstrong and Soltesz, 2012). Further description of CCK⁺ basket cells can be found in Section “Serotonin 3A Receptor-Expressing (5HT_3AR⁺) Inhibitory Interneuron.”

The second major GABAergic IN subclass (∼30% of the entire IN population) in the human motor cortex resides in L2–L6, expressing the neuropeptide somatostatin (Kowall and Beal, 1988; Wonders and Anderson, 2006; Bakken et al., 2021). Sst⁺ INs stem from MGE and predominantly consist of non-fast-spiking Martinotti cells and a minor proportion of non-Martinotti cells, with the latter demonstrating diverse morphological properties such as bitufted cell soma and long-range axonal projections (Rudy et al., 2011; Jiang et al., 2015; Tremblay et al., 2016; Urban-Ciecko and Barth, 2016).

The majority of Martinotti cells concentrates in L5 and L6, receiving excitatory synaptic input from L5 PN axon collaterals and forming synapses with other PN dendrites, and forming disynaptic inhibitory circuits between neighboring L5 PNs and translaminar recurrent inhibitory circuits targeting L2/3 IT neurons (Wang et al., 2004; Silberberg and Markram, 2007; Kätzel et al., 2011; Zolnik et al., 2020). In frequency-dependent disynaptic inhibition, it has been reported that fast (immediate) disynaptic inhibition was mediated by fast-spiking PV⁺ basket cells, whereas non-fast-spiking Martinotti cells mediated delayed disynaptic inhibition between L5 ET neurons (Silberberg and Markram, 2007; Berger et al., 2009; Obermayer et al., 2018).

In contrast to Sst⁺ Martinotti cells targeting L1, the quasi-fast spiking (fast-spiking pattern intermitted by random silences) Sst⁺ non-Martinotti cells targeting L4 were discovered over a century later than Martinotti cells (Ma et al., 2006; Xu et al., 2013). Of note, although the human M1 has been canonically characterized by the lack of the specific Layer 4, the existence of L4 neuron phenotypes in M1 (as in the middle temporal gyrus) was proven by the recent transcriptomic profiling study by BICCN (Bakken et al., 2021), aligning with the histological evidence from animals and humans (Cajal, 1899; Yamawaki et al., 2014; Barbas and García-Cabezas, 2015). Morphologically, non-Martinotti cells have ramified axons inside L4, and their dendritic projections do not extend to L1 (Muñoz et al., 2017; Scala et al., 2019). Thus, it is inferable that this type of non-Martinotti cells may coexist with L4-phenotype neurons scattered in L2/3 and L5 of the human M1. For synaptic connectivity, Sst⁺ non-Martinotti cells in L4 of mouse somatosensory cortex made more connections with PV⁺ INs than with PNs and this connection caused disinhibition of PV⁺ INs, controlling the processing of afferent input from the thalamus (Xu et al., 2013; Nigro et al., 2018). However, the specific role of non-Martinotti cells in the human motor cortex is not clear, as the presence of L4 in the motor cortex has not yet been firmly defined.

Apart from PV⁺ and Sst⁺ INs, almost all of the remaining 30% of cortical GABAergic INs express 5HT_3AR, reside in the supragranular layers (L1–L3, especially abundant in L1) and specifically target PV⁺ and Sst⁺ INs in the neocortex (Chameau and van Hooft, 2006; Lee et al., 2010; Rudy et al., 2011; Bakken et al., 2021). 5HT_3AR⁺ INs originate from CGE, and correspondingly, CGE-derived neurons specifically express 5HT_3AR as the molecular marker (Wester and McBain, 2014; Wester et al., 2019). Similar to the heterogeneity of PV⁺ and Sst⁺ INs from MGE, CGE-derived INs also have highly diverse morphology and electrophysiological profiles (Lee et al., 2010; Miyoshi et al., 2010; Vucurovic et al., 2010). For example, in a genetic fate mapping study of CGE-derived INs by Miyoshi et al. (2010), nine types of morphologically and physiologically different INs were identified in the mice somatosensory cortex. In the same study (Miyoshi et al., 2010), it was also reported that around 75% of CGE-derived INs expressed reelin and vasoactive intestinal peptides (including calretinin and CCK, under the category of 5HT_3AR), and had bipolar, single/double-bouquet, or multipolar morphology. The reelin-expressing INs were found to be late-spiking neurogliaform cells and INs with single bouquet or multipolar morphology, whereas INs expressing calretinin and CCK were burst-spiking double-bouquet or bipolar INs. While the double-bouquet cells mainly target fast-spiking PV⁺ BCs and ChCs (Hioki et al., 2018), single bouquet cells were reported to form synapses that selectively inhibited the activity of L2/3 INs, thus disinhibiting L5 ET neurons (Jiang et al., 2013). In particular, the multipolar CCK-expressing INs were classified as small basket cells, for its morphological and target resemblance with PV⁺ BCs (Armstrong and Soltesz, 2012; Goff and Goldberg, 2019). However, CCK⁺ BCs significantly differed from PV⁺ BCs by their high response sensitivity to acetylcholine, serotonin, or cannabinoids (Földy et al., 2007; Lee et al., 2010). Furthermore, CCK⁺ BCs had a non-fast-spiking profile with asynchronous GABA release onto the postsynaptic neurons, whereas PV⁺ BCs are fast-spiking cells with high GABA release synchronicity (Daw et al., 2009; Bartos and Elgueta, 2012).

M1 paired-pulse TMS paradigms have been widely used to investigate cortical facilitation and inhibition. Although studies using multipulse paradigms have also been proposed, the basic strategy of multipulse protocols can be regarded as an integration of two or more single- and paired-pulse paradigms. Supported by pharmacological evidence [for reviews, see Ziemann et al. (2015), Darmani and Ziemann (2019)], it has been widely accepted that different TMS paradigms preferentially activate different excitatory/inhibitory circuits (Figure 3). As cortical inhibition has been extensively investigated and the results are highly systematic, in the present review, we summarize the patterns and interaction of cortical inhibition in Figure 3, and we specifically focus the discussion on cortical facilitation phenomena, which, at present, remain poorly understood.

As presented in Figure 3, cortical inhibition has been extensively investigated and evidenced to have relationship with cortical GABA activity, which can alternatively be assessed non-invasively using magnetic resonance spectroscopy (MRS). Specifically, the prior study that connected TMS and MRS measurements have revealed that SICI at 1-ms ISI was negatively correlated with GABA level in the sensorimotor cortex (SMC GABA level) while SICI at 2.5-ms ISI did not show specific correlation (Stagg et al., 2011). Accordingly, the authors speculated that the 1-ms SICI and SMC GABA level correlation may reflect extra-synaptic GABA tone, which is different from those involved in 2.5-ms SICI. A subsequent study that investigated the comprehensive relationship between TMS-measured SICI, LICI, and CSP supported the previous results, showing no relationship between 3-ms SICI, 100-ms LICI, and SMC GABA level (Tremblay et al., 2012). Tremblay et al. also reported a moderate positive correlation between CSP duration and SMC glutamate + glutamine level, as well as a positive correlation of SMC GABA and glutamate levels. Based on the prior pharmaceutical evidence that the 2–4-ms SICI may reflect low-threshold GABA_A receptor activity (Ilic et al., 2002), it can therefore be inferred that the mechanisms of 1-ms SICI and 2–4-ms SICI may also come from different ways that GABA acts upon the neurons (i.e., extra-synaptic tone or GABA receptor activity), even though the targeting neurons may be the same. However, contrasting results emerged from recent studies investigating the effect of aging. Hermans et al. (2018) reported reduction of 1- and 3-ms SICI in the group of older adults with no alteration of SMC GABA level compared with younger adults, along with no correlation observed between SICI and SMC GABA level in both groups. Additionally, another study reported reduced SMC GABA level in older adults compared with younger population, along with the task-related SICI (3-ms ISI) reduction of the dominant hemisphere. Similarly, no correlation was found between SICI and MRS outcomes (Cuypers et al., 2020). Regarding the disconnection between MRS-measured SMC GABA level and 2–4-ms SICI outcomes, two reasons can be considered. Firstly, as the MRS measures the general GABA level of the sensorimotor cortex due to the minimum of a 3 × 3 × 3-cm voxel size to acquire reliable outcome (Mullins et al., 2014), possibility exists that the GABA level from the somatosensory cortex (included in MRS outcomes) may affect the MRS-TMS correlation while the paired-pulse TMS reflects mainly the GABA activity in the M1. Secondly, since TMS is an external magnetic stimulation measured through the CST output (MEP, SCEP, etc., reflecting merely the CST-related pathway GABA activity) and MRS outcomes reflect a general GABA level within the SMC, the correlation can be blurred by the possibility that MRS outcomes may come from circuits and neural populations that are not involved in TMS.

Apart from the inhibitory neural circuits, excitatory circuits give rise to paired-pulse cortical facilitation phenomena, which can be subdivided regarding the paired-pulse interstimulus interval (ISI) into (1) short interval intracortical facilitation (SICF), (2) long interval intracortical facilitation (LICF), and (3) interhemispheric facilitation (IHF) (Figure 3). For SICF, it is reported that the ISI with peak MEP facilitation was consistent with the timings of the I-wave; thus, SICF has been used widely as a non-invasive method to evaluate I-wave recruitment, acting as a substitute for invasive SCEP and SMU recordings (Tokimura et al., 1996; Ziemann et al., 1998; Hannah and Rothwell, 2017; Van den Bos et al., 2018; Ziemann, 2020). However, the controversial results from the SICF recruitment curve (the curve plot of SICF-MEP amplitude in different ISIs) in the AP and PA directions shed brought to light some uncertainty in the correspondence between I-waves and the SICF curve. Specifically, the SICF curve assessed by AP-TMS showed enhanced I1-wave-latency facilitation, whereas PA-TMS yielded comparable facilitation for all I-wave latencies (Delvendahl et al., 2014), contradictory to the SCEP evidence showing preferential late I-wave recruitment by AP-TMS. Meanwhile, a recent study reported the facilitation of SICF by the presence of short interval afferent inhibition (SAI) (Saravanamuttu et al., 2021), which also contradicted the PSTH evidence showing inhibition of late I-waves (especially late I-waves in PA-TMS) by SAI (Ni et al., 2011a). Consequently, it may be problematic to analogize SICF to I-waves, because the causal link between SICF and I-wave recruitment is merely the timing consistency, which is inadequate for equalizing the two physiological processes. For LICF, the origin and underlying circuitry remains unknown, despite its early discovery in 1993 (Kujirai et al., 1993) and broad application in basic and clinical TMS studies. One of the complex conundrums regarding LICF is that almost all paradigm-interaction studies found no interaction of LICF with other TMS paradigms (Chen, 2004; Lee et al., 2007; Reis et al., 2008; Udupa et al., 2010; Ni et al., 2011c), not even corticospinal descending volleys in SCEP recordings (Ni et al., 2011b). However, an early study (Nakamura et al., 1997) reported late I-wave facilitation in the LICF paradigm, and another two (Kobayashi et al., 2003; Ni et al., 2007) reported LICF facilitation by long interval intracortical inhibition (LICI) and long interval afferent inhibition (LAI). As the cortical origin of LICF has long been known to be different from those generating the high-frequency I-waves (reflected in the SCEP) (Kujirai et al., 1993; Nakamura et al., 1997; Di Lazzaro et al., 2006), it is therefore inferable that LICF was facilitated by LICI mechanisms (upon which LICI and LAI act), possibly due to selective disinhibition of the silenced PNs by LICI. For IHF, little evidence has been provided since the report of M1-IHF (hand motor area) existence by paired-pulse TMS at an ISI of 4–5 ms (Ferbert et al., 1992; Ugawa et al., 1993; Hanajima et al., 2001). Although Sommer et al. (2012) observed inconsistent M1-IHF at 2 ms ISI paired-pulse TMS that was diminished by sodium channel blocker carbamazepine, the 2 ms ISI brought uncertainty about the observed IHF as it was significantly lower than common M1–M1 interhemispheric transfer time by TMS (8.8–12.2 ms) (Cracco et al., 1989). As IHF is thought to be masked by massive cortical inhibitory mechanisms (especially the powerful IHI), the investigation into IHF appears unviable due to its inconsistency and vulnerability to IHI in paired-pulse TMS paradigms. However, it should be noted that IHF is a phenomenon actually present in cortical activities, as the excitatory attribute of callosal fibers has long been proved (Kawaguchi, 1992; Conti and Manzoni, 1994). That is, excitatory callosal axons project to local inhibitory INs, generating IHI, which, as an alternative can be interpreted as a structure with “superficial IHI above deeper callosal excitation.” Accordingly, if the “superficial” inhibitory effects could be reduced or eliminated, IHF presence may therefore be expected and observed. This insight has gradually gathered interest, and researches attempting to measure online and offline IHF emerged in recent years. In a recent study by Belyk et al. (2021) using paired-pulse IHF with 10 and 50 ms ISI, the authors reported paradoxical facilitation presenting with increased conditioning stimulus (CS) intensity; therefore, it was suggested that when CS is sufficiently intensive, it may overwhelm the contralateral local inhibitory circuits and eventually demonstrate IHF in the common IHI paradigm. Together, we suggest that the previous consideration of “difficult to investigate” for IHF has now reached the prime stage for investigation, to determine the expected clinical application perspectives of IHF in stroke, brain injury, neurodegenerative diseases, and compensation of impaired brain function.

In humans, the cortical motor network consists of M1, dorsal and ventral PM (PMd and PMv), SMA, cingulate motor area, somatosensory cortex (S1), and the inferior parietal lobule (IPL) (Volz et al., 2015; Ruddy et al., 2017). Unlike the M1 paired-pulse paradigm selectivity, dual-site TMS targeting M1 and non-M1 motor areas yielded highly stimulus intensity- and ISI-dependent results. For the PM-M1 interaction, the first report by Civardi et al. (2001) using dual-site TMS and PSTH demonstrated that the motor response of M1-test stimulus (TS) was inhibited by a preceding PMd-CS (90% AMT) at 6 ms ISI, and that when the CS intensity was increased to 110–120% AMT, the effect became facilitatory. Subsequently, MEP of M1-TS was also reported to be facilitated by a delayed ipsilateral suprathreshold PMd-CS at 1.2 and 4.4 ms ISI in humans, yet the facilitation at 1.2 ms ISI was abolished during muscle activation (note that the CS intensity, ISI timing, and discrimination in muscle activation resembled that of M1-SICF) (Ortu et al., 2008; Groppa et al., 2012a,b). For interhemispheric interaction, Baumer et al. (2006) reported IHF effects by conditioning the contralateral PMd with a subthreshold pulse of 60/80% AMT at 6/8 ms ISI, respectively. Similar to the ipsilateral PMd-M1 interaction, the IHF effect also turned into IHI when the CS intensity was set to 90–110% RMT and applied prior to the TS by an 8–10 ms ISI (Mochizuki et al., 2004). Based on this basal intracortical facilitatory and interhemispheric inhibitory effect of suprathreshold PMd-TMS, a further triple-pulse TMS study by Koch et al. (2007) revealed that conditioning the left PMd with paired-pulse TMS also altered the outcome of this basal facilitation and inhibition that PMd exerted on ipsilateral and contralateral M1. Specifically, the pairing of an 80–100% (but not 70%) RMT PMd-CS1 and a 110% PMd-CS2 by a 5 ms ISI canceled the 6 ms PMd-ipsilateral M1 facilitation (i.e., the M1-TS was applied 6 ms after CS2), as well as the 8 ms PMd-contralateral M1 inhibition at both 1 ms and 5 ms CS1-CS2 ISI. Contrary to the basal intracortical facilitatory and interhemispheric inhibitory effects that PMd exerts on M1, PMv tends to elicit more intracortical inhibitory and interhemispheric facilitatory effects on M1 than PMd, as proven by intracortical microstimulation (ICMS) study (Tokuno and Nambu, 2000; Côté et al., 2017), which suggests a substantial functional difference between PMv and PMd. Correspondingly, as reported in rest-state dual-site TMS studies, conditioning ipsilateral PMv 8 ms (80% RMT) or contralateral PMv 40 ms (90 and 110% RMT) before M1-TS led to MEP inhibition (Buch et al., 2011; Côté et al., 2017; Fiori et al., 2017). For the SMA-M1 interaction, the glutamatergic connectivity between M1 and SMA has long been proven in primate and human studies (Tokuno and Nambu, 2000). In humans, this excitatory connectivity typically underpins bimanual coordination, action preparation, and motor imagery (Al-Wasity et al., 2021; Neige et al., 2021). Using dual-site TMS, the SMA-M1 facilitation was confirmed that SMA-CS (140% AMT) applied 6/7/8 ms prior to M1-TMS induced stable rest MEP facilitation at all ISIs in younger adults but not in older adults (Rurak et al., 2021a). However, another study also adopted 140% AMT SMA-CS and a subsequent M1-TS by 6/8 ms ISI, but the MEP facilitation was observed only in younger adults with 6 ms ISI; older adults showed no facilitation at any ISIs (Green et al., 2018). Overall, the experimental evidence for motor network dual-site TMS suggests a close circuit wiring inside the motor network, although the results depend highly on the dual-site TMS paradigm even when assessed in the rest state. As the task-related motor network connectivity and interaction may be much more complicated due to the huge diversity of task types, we do not perform further review of the task-dependent motor network connectivity. Nevertheless, possible reasons for this varying evidence can be summarized as (1) TMS coil size limitation and (2) results contamination by direct CST projections from the non-M1 motor areas. In (1), except for two PMd-M1 dual-site TMS studies (Groppa et al., 2012a,b) in which a customized high-focal TMS coil was used and one S1-M1 dual-site TMS study (Brown et al., 2019) that used figure-of-eight coils with an inner diameter of 25 mm, all other aforementioned dual-site TMS studies targeting M1/PM/SMA adopted a figure-of-eight coil with ≥50 mm loop diameter to perform the non-M1 TMS. Therefore, it is technically difficult to simultaneously aim M1 and ipsilateral nearby areas with two 50/70 mm coils, because the PM-M1 and SMA-M1 distances are both less than 30 mm, and even less than 20 mm in the case of ipsilateral M1-S1 (Mochizuki et al., 2004; Holmes and Tamè, 2018; Brown et al., 2019; Rurak et al., 2021b). For reason (2), while the fibers with largest diameter and fastest conducting velocity in CST come from M1, it was also reported that the CST contained direct projections from SMA, PMd, parietal cortex (mainly terminated in the brain stem), S1, and PMv, presented in descending order of the axon conduction velocity (Firmin et al., 2014; Innocenti et al., 2019). Accordingly, we may expect the change in M1-MEP to be a result of CST output change from other cortical areas instead of M1 excitability. As the axon conducting velocities from different cortical areas vary from 6.01 to 15.49 m/s in the monkey CST (Innocenti et al., 2019) and this variability may result in the iteration of corticospinal descending volleys, future studies are expected to examine the descending volley modulation under motor network dual-site TMS protocols.