Neurocircuitry basis of motor cortex-related analgesia as an emerging approach for chronic pain management

Abstract

Aside from movement initiation and control, the primary motor cortex (M1) has been implicated in pain modulation mechanisms. A large body of clinical data has demonstrated that stimulation and behavioral activation of M1 result in clinically important pain relief in patients with specific chronic pain syndromes. However, despite its clinical importance, the full range of circuits for motor cortex-related analgesia (MCRA) remains an enigma. This review draws on insights from experimental and clinical data and provides an overview of the neurobiological mechanisms of MCRA, with particular emphasis on its neurocircuitry basis. Based on structural and functional connections of the M1 within the pain connectome, neural circuits for MCRA are discussed at different levels of the neuroaxis, specifically, the endogenous pain modulation system, the thalamus, the extrapyramidal system, non-noxious somatosensory systems, and cortico-limbic pain signatures. We believe that novel insights from this review will expedite our understanding of M1-induced pain modulation and offer hope for successful mechanism-based refinements of this interventional approach in chronic pain management.

Chronic pain is a common, distressing, and intractable disease affecting up to 30% of the population, generating a high patient and health system burden¹. The contemporary biopsychosocial model that describes pain as a dynamic consequence of biological, psychological, and social factors stimulates the development of interdisciplinary chronic pain management. However, the non-response rate is high with most of the available techniques². Neuromodulation offers a promising alternative for the management of refractory pain disorders. Stimulation targets identified span a wide range of structures along pain-related circuits³. Among them, the primary motor cortex (M1) emerges as one of the most appealing targets owing to the simplicity and safety of the procedure⁴.

The earliest argument favoring M1 in pain processing dates back to the 1950s when Penfield et al. first described that pain recurrence after primary somatosensory cortex (S1) removal could be diminished following subsequent M1 resection in a patient with central post-stroke pain (CPSP)⁵. Subsequently, cortical removal of post- and pre-central facial representations resulted in long-term pain relief in three patients with trigeminal neuralgia⁶. Preclinical studies were conducted in parallel with neurosurgical explorations. Pyramidal stimulation was shown to inhibit afferent transmission in the spinal dorsal horn (SDH) in cats⁷, and sensorimotor cortex stimulation exerted inhibitory influences on the excitability of primate spinothalamic neurons⁸. Early attempts to apply these findings via internal capsule stimulation in clinical settings were relatively successful⁹, but were discontinued due to surgical complications.

M1 for modern pain management drew special interest from clinicians only after Tsubokawa’s work in the early 1990s. In an effort to develop effective treatments for central pain refractory to spinal cord and deep brain stimulation, they examined the modulatory effects of potential alternative cortical targets on thalamic hyperactivity in a cat model of thalamic deafferentation. A surprising discovery was that complete and long-term inhibition of thalamic hyperactivity was induced by stimulation of M1 instead of S1^{10, 11}. Based on these findings, they treated drug-resistant thalamic pain via epidural M1 stimulation (EMCS), with satisfactory pain control in all seven cases¹². Since this pioneering work, this technique has opened up new frontiers in the surgical treatment of intractable chronic pain syndromes of varying origins. The conditions most amenable to EMCS are refractory complex regional pain syndrome (CRPS) and neuropathic pain (NP)^{3, 13}.

However, EMCS is rarely performed nowadays due to its invasiveness and high costs. Non-invasive brain stimulation (NIBS) paradigms, including repetitive transcranial magnetic stimulation (rTMS) and transcranial direct current stimulation (tDCS), were introduced to target M1 for pain management non-invasively. M1-rTMS, initially serving as a prognostic test for subsequent efficacy during an EMCS procedure, was formally introduced to mitigate CPSP by Lefaucheur et al. in 2001¹⁴. The first documented use of M1-tDCS was for managing NP secondary to spinal cord injury by Fregni et al. in 2006¹⁵. Due to the excellent safety profile, NIBS has undergone exponential growth in pain medicine. Published guidelines on M1-rTMS supported its use for NP and fibromyalgia (FM)¹⁶, similarly for recent M1-tDCS guidelines¹⁷. Likewise, behavioral activation of M1 using motor representation techniques such as motor execution, motor imagery, and motor observation, have demonstrated pain reduction in healthy subjects¹⁸ and some chronic pain syndromes¹⁹. Moreover, the combination of M1 stimulation and motor representation techniques is an optimization opportunity to enhance M1 engagement for satisfactory pain control, which has been progressively confirmed by recent trials^{20, 21}. Here we utilize the term motor cortex-related analgesia (MCRA) to reflect a general reference for these paradigms that target M1 activation to modulate pain perception.

Past decades witnessed the progression of MCRA from anecdotal reports to evidence-based data in relation to neurosurgical approaches, NIBS, and behavioral techniques. Notwithstanding positive results reported, frustrated voices concerning its efficacy never diminished owing to methodological shortcomings and limited sample sizes of some studies³. The unmet clinical need also reflects our insufficient knowledge of the neurobiological underpinnings of MCRA, which may contribute to poor patient selection and the failure of MCRA in some patients undergoing this treatment. Thus, curiosity about the neural basis underpinning MCRA motivates neuroscientists worldwide. The unraveling of MCRA mechanisms started with functional neuroimaging in humans, with a range of brain regions identified (Supplementary Table 1). Positron emission tomography further extended our knowledge regarding molecular underpinnings of MCRA in vivo with appropriate radiotracers (Supplementary Table 2). Several hypotheses (e.g., the role of thalamus, descending control system, and opioidergic circuitry) derived from these seminal studies^{12, 22, 23} were verified in experimental models of M1 stimulation. Animal models allowed for research efforts for unmasking MCRA neurocircuitry through direct examination of hypotheses drawn from current pain theories, with a kaleidoscope of neurobiological techniques including single-unit recording, small animal imaging, pharmacological interventions, and post-mortem histological assessment. Hopefully, the advent of optogenetics and virus tracing enables cell- and projection-specific visualization and manipulation of M1 neurons in mice, and neuroscientists are now directing their focus toward a state-of-the-art integration of transsynaptic neuronal tracing, projection-specific neuromodulation, and electrophysiology for a thorough delineation of behaviorally relevant circuit intricacies for MCRA²⁴.

Despite a series of excellent reviews touching on this issue^{4, 22, 25, 26}, surprisingly few summaries discuss MCRA from the viewpoint of neurocircuitry. This review ties previous and recent knowledge from patients and healthy subjects with experimental animal data on MCRA, and speculatively discusses possible neurocircuitry rationale for the management of chronic pain disorders based on contemporary pain theory (Box 1), covering the endogenous pain modulation system, the extrapyramidal system, the thalamus, non-noxious somatosensory systems, and cortico-limbic pain signatures. Finally, we conclude by pointing out M1 as the critical hub integrating endogenous pain and motor interaction. We hope this review will provide the groundwork for future research and discussion among neuroscientists and clinicians in the domain of MCRA.

BOX 1. Neuronal circuitry for pain processing.

Nociceptive inflow through the spinal dorsal horn conveyed by primary nociceptors undergoes extensive modulation by local circuits involving excitatory and inhibitory interneurons, before being relayed to projection neurons that target higher-order brain centers. The activities of primary afferents and nociresponsive neurons are also subject to supraspinal descending control with its pro- and anti-nociceptive components, including but not limited to serotonergic, noradrenergic, and dopaminergic fibers. These fibers act on both pre- and post-synaptic sites to control the gain of excitability during nociceptive transmission^{139, 142}.

Projection neurons within laminae I and III-VI constitute the major output from the dorsal horn to the brain. These neurons carry noxious information toward primary and second somatosensory cortices by way of ventral posterior lateral/medial nucleus, and toward second somatosensory cortex and posterior insula through posterior thalamic nucleus, which is thought to encode the sensory-discriminative aspect of pain. The centromedian-parafascicular nuclei receive indirect projections from the dorsal horn via the spinoreticulothalamic tract, contributing to affective-motivational qualities of pain experience¹⁴³. Apart from intralaminar thalamic nuclei, the mediodorsal thalamic nucleus, a major component of the medial thalamus, receives nociceptive inputs from the spinal dorsal horn, subnucleus reticularis dorsalis, and parabrachial nucleus, encoding pain affect via projections to the anterior cingulate cortex^{144, 145}.

Although the thalamus is the primary target of spinal dorsal horn neurons in primates, the majority of projection neurons in rodents terminate in the parabrachial nucleus and in turn get access to forebrain limbic structures, including amygdala, anterior cingulate cortex, and insula, which are relevant to emotional components of pain and pain-related autonomic reactions. Ascending information also accesses brainstem neurons to engage descending feedback systems. As the core of this system, the periaqueductal grey regulates spinal pain processing by employing inferior brainstem nuclei including the rostroventromedial medulla, locus coeruleus, and subnucleus reticularis dorsalis¹⁴².

According to current functional subdivision of the nociceptive system, the exteroceptive branch detects external threats and drives reflexive-defensive reactions to prevent or limit injury, while the interoceptive branch senses the disruption of body integrity, produces strong aversive emotional components, and drives self-caring responses to reduce suffering¹⁴⁶. Pain itself also has a cognitive component, requiring anticipation, attention, learning, recall of past experiences, and active decision making in the psychosocial context. Related neural processes involve a cerebral network through the anterior cingulate cortex, prefrontal cortex, hippocampus, and amygdala¹⁴⁷. Given the above, pain involves a distributed group of brain structures instead of a single area, with each presumably involved in distinct aspects of pain processing¹⁴⁸. It is now generally accepted that the final experience of pain is the denouement of dynamic interactions between the dorsal horn circuitry engaged to convey pain signals and the modulatory actions collected from higher brain centers whose activity can be influenced by emotion, motivation and other cognitive states that ultimately exacerbate or mitigate overall pain experience associated with specific noxious stimuli^{72, 149, 150}.

Neurocircuitry of motor cortex-related analgesia

Generally, M1 stimulation alters the electrical state of individual cortical motor neurons, affects local microcircuitry in favor of enhanced motor output, and induces synaptic plasticity changes for long-term pain control. The interrogation of the internal circuitry of M1 recruited by stimulation-triggered electrical fields has identified input elements of the pain-relieving circuitry²⁷, but this is outside the scope of the present account. Besides modifications in local microcircuitry, M1 stimulation elicits changes in distant pain structures. The superior location and long-range outputs of M1 provide the anatomical basis for its control over the SDH and central pain-processing nuclei in a top-down style or horizontally (Box 2). Possible sources of M1-related neurocircuitry for MCRA and the strength of evidence are provided in Supplementary Table 3.

BOX 2. Primary motor cortex structure and connectivity.

1. Primary motor cortex structure

M1 was thought to form a continuous somatotopic homunculus extending down the precentral gyrus. According to the homunculus, the lower-limb area dips into the longitudinal fissure and has a weak appearance, while the head or upper-extremity is coded by regions on the more accessible lateral convexity surface. In particular, the face representation is characterized by a disproportionately large area compared to the body¹⁵¹, and this explains why M1 stimulation is more effective for orofacial pain than lower-limb pain^{152, 153}. Recent imaging data showed the classic homunculus was punctuated by regions with distinct structure and function. These regions exhibited strong functional connectivity to the cingulo-opercular network, critical for action, autonomic functions, error-related activity, and pain. Thus, current opinion esteems M1 as two parallel systems interleaved: effector-specific regions for isolating fine motor control and the somato-cognitive action network for integrating body control (motor and autonomic) and action. The latter is required if an individual is to achieve its goals through movement while avoiding injury and maintaining physiological allostasis¹⁵⁴.

Implications for MCRA:

The integrate-isolate pattern provides a substrate for two types of pain-resolving efficacy of MCRA. One of these, subject to somatotopic influences, modulates the sensory aspect of pain, whereas another one, devoid of somatotopic drive, may engage higher-order mechanisms involving affective and cognitive appraisal of pain. In support of such duality, the clinical efficacy of M1 stimulation has been alternately reported to prevail on the sensory or the affective dimension of pain (see refs. in supplemental materials). Interestingly, a few studies also reported improved autonomic dysregulation related with chronic pain by M1 stimulation^{12, 97, 155}. In the field of animal studies, the overwhelming majority of existing MCRA studies evaluated pain experience by measuring reflexive-defensive behaviors, with only a few ones emphasizing behavioral anxiety and aversion along with neuropathic pain^{24, 30}. Regarding cognitive effects of M1 stimulation in pain control, anecdotal observations revealed seemingly contradictory conclusions without any mechanistic explorations^{156, 157}.

2. Primary motor cortex connectivity

Knowledge regarding the position of M1 in sensorimotor networks is a prerequisite for understanding circuitry determinants of MCRA. The most basic classification of M1 neurons distinguishes excitatory pyramidal neurons and local inhibitory interneurons. Defined by projection targets, pyramidal neurons could be further subdivided into IT neurons, PT neurons, and CT neurons. IT neurons, mainly located in L2-L5, project ipsi- or bilaterally within the cortex and striatum via the corpus callosum and external capsule. IT neurons in L2/3 only have cortico-cortical projections, whereas those in L5 have both cortico-cortical and cortico-striatal projections. PT neurons, exclusively found in L5B, project to brainstem and spinal cord via the internal capsule, cerebral peduncle, and PT. L6 CT neurons exclusively innervate the thalamus^{158, 159}. Within the cortical column of M1, neural signals tend to flow into superficial layers and then travel to deep layers. L2/3 IT neurons project to L5 IT neurons and L5B PT neurons in specific parallel pathways (interlaminar excitation), and L5 IT neurons unidirectionally engage L5B PT neurons (intralaminar excitation), placing PT neurons downstream of their IT counterparts in microcircuit connectivity¹⁵⁹.

Long-range inputs converge onto M1 through cortico-cortical, thalamocortical, and neuromodulatory projections with characteristic laminar profiles from L2 to L6, and we direct readers elsewhere for further information^{35, 160}. Here we elaborate on M1 outputs which are almost exclusively glutamatergic. L5 pyramidal neurons provide substantial subcortical structures, including the striatum, thalamus, and subthalamic nucleus, with a copy of cortical motor outflow en route to their caudal-most destinations in the brainstem and spinal cord via collateral branches³⁵. Corticospinal tract fibers emanating from M1 terminate primarily in laminae VI–IX, with minor fibers in the DCN and scarce to no fiber in laminae III–V in rodents and primates. L5 pyramidal neurons have direct projections to contralateral spinal motoneurons for fine movement control^{38, 39}. Among brainstem targets, RN and RF are the origins of descending motor control, pontine nuclei are the recipient part of the cortico-cerebello-cortical motor loop, and SC, APT, and substantia nigra also serve as key nodes in the network for goal-directed motor behaviour¹⁶⁰. Importantly, projections are also established with pain-related nuclei, and this is the case of the PAG and SRD³⁵.

M1 efferents toward the basal ganglia form three distinct pathways known as direct, indirect, and hyperdirect pathways. M1 neurons that connect with direct and indirect pathways synapse on distinct sets of striatal neurons: striatal input neurons of the direct pathway that express dopamine receptor D1 project to basal ganglia output nuclei SNr and GPi, while those that express dopamine receptor D2 within the indirect pathway first synapse at the GPe and STN en route to the SNr and GPi¹⁶⁰. As the convergence point for direct and indirect pathways, GPi neurons in turn influence M1 activity via thalamocortical projections. The direct pathway is presumed to facilitate desired movement representations, while the indirect pathway inhibits unwanted movement. The hyperdirect pathway toward the STN also enables modulation of thalamocortical activity via the GPi, allowing for rapid ‘braking’ of M1 outputs¹⁶¹.

The VA-VL complex is the most prominent thalamic recipient for L6 inputs. Other thalamic components receiving M1 projections include the Po, MD, VB, CeM-PF, CL, and CM, all of which have proven to be involved in somatosensory processing^{35, 84}. Cortico-thalamic projections are conceptualized to function as a feedback loop that gates motoric and sensory inputs by modulating response properties of thalamic relay neurons¹⁶². Notably, M1 also forms direct connections with the TRN and ZI^{35, 84}, two diencephalic structures with critical importance in controlling sensory information flow from the thalamus to the cortex⁷⁶. The axons of L2/3 pyramidal neurons extensively project to bilateral S1 in superficial layers. Other reported cortical output regions include adjacent motor areas, IC, ACC, orbital cortex, etc^{35, 84}.

Implications for MCRA:

First, somatotopically organized M1 outputs toward the spinal cord, S1, striatum, and thalamus (see refs. in supplemental materials) provide the neuroanatomical basis for the somatotopic feature of MCRA^{29, 163}. Second, M1 efferents, especially those form the whisker region of rodent M1, are characterized by bilateral projections, which is considered to support bilateral whisking behavior coordination (see refs. in supplemental materials). For MCRA, M1 stimulation applied to the side contralateral to the pain is routinely recommended in case of focal pain^{29, 163}. Intriguingly, ipsilateral analgesia is also well-documented in neuropathic pain rat models, despite shorter duration than contralateral EMCS¹⁶⁴. Thus, ipsilateral EMCS was usually adopted in basic studies to eliminate movement-related artifacts when measuring pain threshold^{33, 79}. Bilateral diffuse MCRA is also seen in non-invasive brain stimulation paradigms in the treatment of chronic widespread pain, although contralateral effects may be slightly higher^{165, 166}. Third, there exists a functional dichotomy of circuit connectivity of M1 outputs in regulating distinct components of pain, with L6 and L5-related pathways specifically modulating emotional pain and sensory hypersensitivity, respectively²⁴. These insights will help improve neurostimulation therapies for optimal pain control.

3. Graphical illustration

The figure showes layer-specific and cell type-specific local excitatory connections (top) and long-range axonal projections of each pyramidal cell-type (bottom) of M1. At bottom, nuclei related with sensory processing are indicated by red fonts, while those esteemed as motor-related nuclei indicated by black. Abbreviations: ACC, anterior cingulate cortex; APT, anterior pretectal nucleus; CeM-PF, centromedian-parafascicular thalamic complex; CL, centrolateral thalamic nucleus; CM, centromedial thalamic nucleus; CT, cortico-thalamic neurons; DCN, dorsal column nucleus; GPe, globus pallidus pars external; GPi, globus pallidus pars internal; IC, insular cortex; IT, intratelencephalic neurons; MD, mediodorsal thalamus; PAG, periaqueductal gray; Po, posterior thalamic nucleus; PT, pyramidal tract neurons; RF, reticular formation; RN, red nucleus; S1, primary somatosensory cortex; SC, superior colliculus; SNc, substantia nigra pars compacta; SNr, substantia nigra pars reticulata; SRD, subnucleus reticularis dorsalis; TRN, thalamic reticular nucleus; VA-VL, ventroanterior and ventrolateral thalamic complex; VB, ventrobasal thalamic complex; ZI, zona incerta.

Endogenous pain modulation system

Convergent evidence suggests top-down modulation of spinal pain processing in MCRA. In animal studies, morphological data have shown that MCRA decreased the expression of functional markers of activated neurons, such as FOS, within the SDH in naïve and NP rats^28–31; electrophysiological studies further documented the inhibitory effects of electrical M1 stimulation on primary afferent fibers and spinothalamic neurons in primates^{7, 8}, as well as spinal wide dynamic range (WDR) and nociceptive specific neurons in naïve and NP rats^32,33. In human studies, the paradigm of conditioned pain modulation (CPM) was introduced to decipher the mechanistic effects of MCRA at the spinal level, which was significantly enhanced by non-invasive M1 stimulation in both healthy and pain populations³⁴, suggestive of the potential role of subnucleus reticularis dorsalis (SRD) or descending pain control in MCRA (Box 3). Indeed, the existence of M1-SRD projections³⁵ provides an anatomical basis for the involvement of SRD in MCRA, and future work is required to link behavior to neurocircuitry mechanisms. Next we focus on these canonical descending control pathways (Figure 1a).

BOX 3. Effects of somatosensory modalities on pain.

1. Cross-modal pain modulation

Interactions of multiple sensory inputs are pivotal for individuals to orchestrate proper global reactions in response to complicated sensory cues from the environment. Pain is subject to cross-modal inhibition from vision, sound (see refs. in supplemental materials), and somatosensation. Here we focus on the inhibition of pain by innocuous mechanical and thermal sensations.

Spinal circuits for somatosensation:

Primary afferent neurons transmit somatosensory information from skin, muscle, and joints to the spinal cord. Among them, mechanosensory neurons account for proprioception, touch and a subset of nociception. Proprioceptive neurons that sense muscle stretch and tension mainly employ thickly myelinated Aα fibers to transmit signals to medulla through direct dorsal column pathway. Touch sensory neurons, or Aβ low-threshold mechanoreceptors, are sensitive to vibration, pressure, stretch of skin, and movement of hairs. They send a branch that ascends to the brainstem through direct dorsal column pathway. In parallel, they synapse with dorsal horn projection neurons, which send ascending axons along the dorsal column (indirect pathway). Neurons in both direct and indirect pathways synapse with dorsal column nuclei that are bound for contralateral thalamus via the medial lemniscal pathway. In addition, there is a third population of mechanosensory neurons that senses noxious mechanical insult (high-threshold mechanoreceptors). Apart from mechanosensory neurons, other neurons sense temperature, a large fraction of which is also nociceptive neurons for detecting hot or cold temperatures. These thermal and nociceptive (including thermal and mechanical pain) neurons utilize thinly-myelinated Aδ and unmyelinated C fibers and synapse with second-order dorsal horn neurons, which ascend within the anterolateral column pathway to the contralateral thalamus. Distinct thalamic neurons then relay touch, pain, thermoreception, and proprioception to the primary somatosensory cortex^{143, 167}.

Regarding primary afferents terminating areas in the spinal cord, C fibers terminate predominantly in laminae I–II and transmit noxious heat and mechanical information. Aδ fibers transmit a mixture of noxious and innocuous tactile and cold information and terminate mainly within laminae I and V. However, Aβ fibers carrying the bulk of innocuous tactile information terminate in laminae III–V. Aα fibers terminate widely throughout laminae IV–VI and the ventral horn, where they contribute to sensory-motor loops¹⁶⁸.

Pain inhibition by thermoreception:

Thermal grill illusion, in which an array of interspersed warm and cool bars causes a strong sensation of heat, provides a good illustration of cross-modal pain inhibition by thermoreception. A widely recognized explanation for this phenomenon is that activity in warm fibers interferes centrally the ability of cool fibers to suppress signals from cold-sensitive nociception¹⁶⁹. This unmasking mechanism accounts for the pathogenesis of central poststroke pain. Patients with central poststroke pain usually experience a dramatic loss of warm and cool sensation frequently accompanied with unremitting burning pain within the affected body part, mimicking that of thermal grill. These disturbances are proposed to be caused by the thermosensory loss which consequently releases integrated polymodal nociceptive activity at the thalamic level¹⁷⁰.

Pain inhibition by mechanoreception:

Pain relief by light mechanical stimulation is a common experience. This pain-tactile interaction prompted the original development of the gate control theory, which denotes that large fibers (Aβ) establish a gate by feedforward activation of inhibitory interneurons that acts to attenuate small fiber (C) pain pathways¹⁴². Correspondingly, dorsal column stimulation and peripheral stimulation were developed as a direct spin off this theory for pain control^{3, 171}. Apart from the spinal dorsal horn, insular and primary somatosensory cortices also serve as an interface enabling bottom-up modulation of pain perception by touch^{172, 173}, and a tactile-specific thalamocortical pathway from the ventroposterior medial nucleus to the S1 barrel cortex has recently been proven to mediate whisking-induced face analgesia¹⁷³. Whereas the pathophysiology of central poststroke pain highlights the involvement of spinothalamic tracts¹⁷⁰, evidence also exists that the dysfunction of the dorsal column-medial lemniscal pathway is causally associated with the development of neuropathic pain¹⁷⁴. Although thus far there is less circuitry underpinnings of proprioception-pain interaction, somatosensory thalamus stimulation has early been developed for the management of deafferentation pain based on the theory that pain is caused by lack of propriocepive stimuli reaching the thalamus¹⁷¹.

Implications for deafferentation pain:

Since nociceptive neurons in the spinothalamic system are subject to inhibitory control from the system mediating non-noxious sensory information at multiple levels of the central nervous system, one putative action mode of deafferentation pain is that the loss of sensory afferent signals leads to deafferentation hyperactivity of nociceptive neurons⁶⁷. Echoing such hypothesis, peripheral nerve stimulation, spinal cord stimulation, and sensory thalamus stimulation were intended to relieve pain in which the deafferented level was below the targeting site by compensating for the lack of normal sensory input^{67, 171}.

2. Conditioned pain modulation

Differing from the generally modest decrease in pain induced by mechanical stimuli, stronger analgesia usually arises when the additional stimulus is itself painful. Furthermore, the efficacy of the noxious “conditioning” stimulus is not contingent on its distance on the body from the test stimulus¹⁷⁵. This “pain inhibits pain” paradigm is referred to as conditioned pain modulation, which was previously described as diffuse noxious inhibitory control for assessing endogenous pain control in animals. This concept derived from early observations that electrophysiological responses of spinal wide dynamic range neurons to somatic noxious stimuli were inhibited following introduction of an extra-segmental noxious stimulus¹⁷⁶. It is driven by a pro-nociceptive reverberating spinal-medulla loop integrated at the subnucleus reticularis dorsalis¹⁷⁶, which serves as an integrative relay for descending pain facilitation from the anterior cingulate cortex, hypothalamus, rostroventromedial medulla, and locus coeruleus¹⁷⁷. In addition, descending dopaminergic and serotoninergic controls participate in the loop subserving this phenomenon¹⁷⁸.

Figure 1. Possible neurocircuitry mechanisms underlying recruitment of descending pain control system and modulation of thalamic pain processing by MCRA.

a. Activation of the PAG-RVM pathway via direct or indirect (through ZId glutamatergic neurons) pathways from M1 evokes the release of 5-HT in the SDH, which may exert presynaptic inhibition of primary afferent fibers via 5-HT_1AR. The descending dopaminergic system could also be engaged by MCRA via indirect pathways. Dopaminergic fibers originating from A11 may inhibit the activity of primary afferents and PNs via pre- and post-synaptic D₂Rs, respectively. b. MCRA may inhibit thalamocortical pathways through TRN GABAergic neurons. Direct M1 inputs may also suppress the activity of pain-related thalamic projection neurons via local GABAergic neurons in higher-order animals. In addition, M1 indirectly activates the NAcC via the MD to specifically negatively modulate emotional valence associated with chronic pain. Abbreviations: 5-HT_1AR, serotonin 1A receptor; A11, A11 hypothalamic nucleus; DA, dopamine; D₂R, dopamine 2 receptor; DRG, dorsal ganglion; Glu, glutamate; M1, primary motor cortex; MD, mediodorsal thalamus; IN, interneuron; NAcC, nucleus accumbens core; NAcSh, nucleus accumbens shell; PAG, periaqueductal gray; pIC, posterior insular cortex; PN, projection neuron; Po, posterior thalamic nucleus; RVM, rostral ventromedial medulla; S1, primary somatosensory cortex; S2, secondary somatosensory cortex; SDH, spinal dorsal horn; TRN, thalamic reticular nucleus; VB, ventrobasal thalamic complex; CeM-PF, centromedian-parafascicular thalamic complex; ZId, dorsal zona incerta; ZIv, ventral zona incerta.

Spinal nociception control by M1 stimulation was once believed to be driven by direct M1-spinal cord projections³², which was seemingly supported by the positive correlation of the integrity of corticospinal tract (CST) and the absence of severe paresis with the likelihood of positive outcomes following MCRA in CPSP control^36,37. However, unlike its sensory counterpart, M1 provides scarce projections to the SDH and medulla dorsal horn across species^38–40, excluding the possible involvement of direct M1-spinal cord pathway in MCRA. Instead, the CST may mediate motor system repair through axon sprouting in CPSP patients⁴¹. It is broadly established that the CST originating from M1 sends collaterals to a host of brainstem nuclei associated with descending pain control^{42, 43}. Conceivably, indirect M1-SDH projections may play a leading role in spinal nociception control.

Periaqueductal grey (PAG) integrates inputs from multiple cortical regions and mediates cortical modulation of nociception⁴⁴. Neuronal subpopulations of the PAG have been identified as segregated drivers of pain modulation, with glutamatergic neurons producing analgesia in opposition to GABAergic inhibitory interneurons (INs). GABAergic INs exert tonic inhibition over glutamatergic neurons, which are output neurons that project to the RVM for descending pain inhibition^{45, 46}. There is a wealth of imaging literature underscoring functional activation of PAG in patients with NP during EMCS^{22, 23, 47}, which was corroborated by preclinical data showing enhanced neuronal firing rates and FOS expression in ipsilateral PAG induced by EMCS^{29, 48, 49}. Recently, Gan et al. observed that optogenetically activating the M1-PAG pathway robustly attenuated mechanical allodynia in NP mice, with increased FOS expression noted in both GABAergic and non-GABAergic postsynaptic neurons. Transsynaptic labeling revealed that more than 80% of PAG neurons receiving direct synaptic inputs from M1 are non-GABAergic, suggesting that M1 activation shifts the neuronal excitation-inhibition balance in the PAG toward increased activation. Further tracing data showed PAG neurons receiving M1 inputs projected to the RVM and LC²⁴, highlighting their possible involvement in descending modulation of spinal nociception by MCRA.

RVM serves as a gateway for descending pain control from PAG. Serotonin (5-HT) released from the terminals of descending RVM projections acts on various spinal 5-HT receptors, the effect of which varies from pro- to anti-nociception depending on receptor subtypes. 5-HT_1AR is considered to be among those subtypes that predominantly promote anti-nociception⁴². Morphological studies indicated that EMCS induced the activation of RVM neurons³¹, and increased 5-HT-expressing neurons in the RVM^{28, 31} as well as spinal 5-HTergic terminals⁵⁰ in healthy and NP rats. Functional involvements were further documented by pharmacological block of EMCS analgesia with an intra-RVM injection of GABA_AR agonist or spinal injection of 5-HT_1AR antagonist⁵¹. Considering these, serotoninergic projections from the RVM acting on spinal 5-HT_1AR mediate descending anti-nociceptive effects of MCRA.

As a noradrenergic component of supraspinal anti-nociception, LC receives direct inputs from PAG for descending pain modulation. Noradrenaline released from descending LC-spinal pathways attenuates pain mainly by inhibitory actions on α₂-adrenoceptors on central terminals of primary afferents and spinal pain-relaying neurons⁴². Although EMCS increased the firing rates of LC neurons, pharmacological block of descending noradrenergic influence failed to attenuate EMCS analgesia in sham and NP rats⁵², and LC-projecting neurons in the PAG constituted less than 10% of neurons that were transsynaptically labeled from incoming M1-PAG projections²⁴. Considering these, coeruleospinal pathways may not be critical for MCRA.

As a major source of descending dopaminergic projections, the hypothalamic A11 nucleus receives innervation from brainstem (e.g., the PAG and parabrachial nucleus) and cortical (e.g., ACC) areas. Descending dopaminergic pain inhibition is mediated through spinal dopamine receptor D2 (D₂R) activation⁴². The involvement of this pathway in MCRA was substantiated by pharmacological data showing that blocking A11 or spinal D₂Rs reverted EMCS-induced inhibition of spinal nociceptive processing in NP rats³³. Given the lack of direct M1-A11 projections, indirect pathways may contribute to A11 activation.

Taken together, the neurocircuitry substrates underlying MCRA mimic that of CPM. As is known, the imbalance between descending facilitatory and inhibitory pathways, with higher activity in the former, contributes to the development of chronic pain⁵³, and MCRA may restore this imbalance by reinstating endogenous control of pain. In clinical settings, the evaluation of pain modulation capability serves as a step forward in individualizing pain medicine⁵⁴. It is therefore of interest for future studies to validate the CPM biomarker as a prognostic factor of MCRA.

Extrapyramidal system

Aside from motor control, the basal ganglia is integral to somatosensory processing⁵⁵. The neurocircuitry basis of extrapyramidal pain modulation is briefly discussed in Box 4. Several lines of evidence converge to support the role of striatum in MCRA. First, in line with imaging data indicating significant striatal cerebral blood flow (rCBF) increments by EMCS in NP patients⁴⁷, decreased glucose metabolism in the striatum was reversed by EMCS in NP rats⁵⁰. Second, electrical stimulation of M1 induced striatal dopamine release in cats⁵⁶, and this finding also holds for M1-rTMS in healthy individuals^{57, 58}. Finally, homozygous D₂R T/T genotype rendered participants more likely to experience CPSP reduction with M1-rTMS compared with other genotypes⁵⁹, and the involvement of striatal D₂R in MCRA was further substantiated by striatal administration of D₂R antagonists in NP rats³³. Given the above, MCRA may employ descending pain control through dopaminergic inhibition of the indirect pathway (Figure 2). The modulation of neuronal activity by M1 stimulation within the cortico-basal ganglia-thalamo-cortical motor loop also mediates motor symptom improvements in patients with Parkinson’s disease⁶⁰.

BOX 4. Extrapyramidal modulation of pain.

Early evidence:

As the principal input nucleus of basal ganglia, the striatum receives dopaminergic projections from SNc and somatotopically-organized glutamatergic inputs from frontal motor areas⁵⁵. In contrast to cortico-striatal projections which were found to modulate pain until recently (see refs. in supplemental materials), there has been growing appreciation for the role of dopaminergic innervation in striatal pain processing. Earlier observations showed lesion of nigrostriatal neurons led to neuropathic pain related with Parkinson’s disease¹⁷⁹. The importance of striatal D₂Rs was further revealed by imaging data showing that individuals with few available D₂Rs, possibly reflecting high dopaminergic transmission, likely had a high tonic level of pain suppression¹⁸⁰. Accordingly, striatal administration of D₂R (not D₁R) agonist for inhibiting the indirect pathway attenuated neuropathic hyperalgesia, which could be reversed by lesions of ipsilateral globus pallidus, SNr, and STN, implicating a pain-facilitatory role of the indirect pathway¹⁸¹. Further research indicated that the anti-hypersensitive effect of striatal D₂Rs in neuropathic pain involved suppression of RVM pro-nociceptive neurons and subsequent descending influence acting on spinal 5-HT receptors¹⁸².

Recent neurocircuitry findings:

Neurocircuitry mechanisms underlying the engagement of descending 5-HTergic projections by indirect pathway of basal ganglia is still a subject for speculation, but recent insights into STN in pain modulation provide a clue. The activity of STN neurons was enhanced in various persistent pain states. In the Parkinson’s disease model, STN overactivity was causally related to central pain processing and behavioral hypersensitivity, and STN projections to the SNr, GPi, and VP differentially facilitate mechanical and thermal pain^183–185. Transsynaptic tracing showed that these basal ganglia output nuclei innervated the thalamus (the mediodorsal thalamus and ventral posterior lateral nucleus), brainstem (the PAG, RVM, and PBN), and forebrain regions (the prefrontal cortex, anterior cingulate cortex, and amygdala), laying anatomical basis for the modulation of central pain processing and recruitment of descending control system by the basal ganglia¹⁸³. Apart from engaging descending 5-HT control, STN overactivity positively regulated chronic pain via glutamatergic STN-PBN projections¹⁸⁶. Given the above, the indirect pathway may be an effective access point for modulation of neural systems subserving chronic pain processing.

Figure 2. Scheme illustrating proposed extrapyramidal circuitry mechanisms in MCRA.

M1 stimulation inhibits the activity of indirect pathway of basal ganglia through striatal D₂Rs, thus leading to the disinhibition of descending PAG-RVM pathway as well as inhibition of pain transmission in the PBN. Direct glutamate-dopamine axon-axonal interplay in the striatum through cortico-striatal projections may mediate the recruitment of striatal DA signaling by M1 stimulation. Abbreviations: AMPAR, α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor; DA, dopamine; D₂R, dopamine 2 receptor; GABA, gamma-aminobutyric acid; Glu, glutamate; GPe, globus pallidus pars external; GPi, globus pallidus pars internal; M1, primary motor cortex; PAG, periaqueductal gray; PBN, parabrachial nucleus; RVM, rostral ventromedial medulla; SNc, substantia nigra pars compacta; SNr, substantia nigra pars reticulata; STN, subthalamic nucleus; VP, ventral pallidus.

M1 stimulation may induce striatal dopamine release through modulation of dopamine neuron firing or a direct effect of cortico-striatal projections on striatal dopaminergic terminals. But the first scenario is unlikely owing to the scarce M1-substantia nigra pars compacta projections³⁵. Cortico-striatal fibers synapse on the dendritic spines of medium spiny neurons, and these neurons are also in close proximity to dopaminergic terminals located on the same spines. This configuration allows for the modulation of striatal dopamine release by glutamate through direct axonal interrelationship or volume transmission⁶¹. Under the condition of M1-rTMS, the striatal area with significant dopamine release corresponds to the projection zone of cortico-striatal efferents from the stimulated M1. This spatially-restricted dopamine release implicates direct cortico-striatal influence on striatal dopamine terminals⁵⁸ (Figure 2).

Thalamus

The thalamus integrates nociceptive inputs conveyed from the SDH and then transmits them toward the cerebral cortex via thalamocortical (TC) tracts. However, cortico-thalamic projections should not be neglected, though relatively little is known about their contribution to pain⁶². Neuroimaging findings have pointed to a direct implication of TC tracts in MCRA. Alterations of functional connectivity between the ventral posterior lateral nucleus (VPL) and its cortical targets correlated with reductions in clinical pain with M1-tDCS in healthy volunteers and patients with FM^{63, 64}, and the integrity of TC tracts was proven to be required for anti-nociceptive effects of M1-rTMS in patients with central pain^{36, 65}. Collectively, the thalamus serves as the principal hub by which M1 regulates a distributed pain-related cortical network via TC circuits.

Decrease of thalamic blood flow contralateral to the painful side is associated with NP pathophysiology, and restoration of thalamic hypometabolism represents a promising condition to gain favorable relief by analgesic therapies including MCRA⁶⁶. Tsubokawa et al. first described drastic rCBF increments in the ipsilateral thalamus by EMCS in patients with CPSP¹². This observation was soon reproduced by independent authors in patients^{22, 47} and rats⁵⁰ with NP. The study series reported by Garcia-Larrea et al. showed that EMCS yielded the most significant rCBF increase in the ventroanterior and ventrolateral thalamus, with sub-significant changes in the medial thalamus and other cortical correlates in NP cases. However, the activation of the ventroanterior and ventrolateral thalamus was irrelevant to pain amelioration^{22, 23}. Thus, it is tempting to suggest that EMCS recruits sensory and motoric thalamus through parallel cortico-thalamic pathways, with the former serving as the node for modulation of cortical and subcortical pain-processing nuclei and the latter for alleviation of stroke-related motor disorders^{23, 67, 68}.

As referred before, the original rationale for the clinical use of M1 stimulation arose from the inhibitory effects on pathological thalamic bursting upon electrical M1 stimulation in NP cats¹². Thalamic neuronal hyperactivity has been recorded in the VPL, posterior thalamic nucleus (Po), as well as medial and intralaminar thalamic nuclei in the setting of NP^69–71. Such aberrant overactivity develops concomitantly to thalamic hypometabolism, and is also consistently restored by analgesic procedures including M1 stimulation⁷². Immunohistochemical studies have shown that EMCS decreased FOS expression in the VPL, ventral posterior medial nucleus (VPM), and centromedian-parafascicular thalamic complex in naïve or NP rats^{48, 49}. Electrophysiological data further demonstrated that EMCS decreased single-unit activities of the VPL^{50, 73}, Po⁷⁴, and centromedian-parafascicular thalamic complex⁴⁹ in naïve or NP rats. Notably, the inhibition of abnormal VPM firing was recently recapitulated by optogenetic M1 activation in rats with orofacial NP⁷⁵. It bears mentioning that perfusion increments in the thalamus are not contradictory with suppressed neuronal discharge upon M1 stimulation. As demonstrated in functional neuroimaging, hemodynamic responses could subserve excitatory and inhibitory synaptic activities. As such, M1 stimulation may predominantly activate preserved functional thalamic or extra-thalamic GABAergic processes, thus normalizing the hyperactive deafferented thalamus.

Thalamic relay neurons receive inhibitory afferents mainly from the thalamic reticular nucleus (TRN) and extra-thalamic sources including the zone incerta (ZI), anterior pretectal nucleus, substantia nigra pars reticulate (SNr), and pontine reticular formation^{76, 77}. As an integrative node for global behavior modulation⁷⁸, ZI is the best-studied source of thalamic inhibition in the domain of MCRA. The dorsal ZI predominantly harbors excitatory neurons by virtue of projections toward brainstem nuclei. By contrast, the ventral ZI (ZIv) is overwhelmingly GABAergic in nature and exerts feedforward inhibition preferentially in higher-order thalamic nuclei (e.g., the Po)^{24, 78}. Early studies based on non-specific region and cell-type modulation seemed to allow the assumption that peripheral pain signals to the Po were gated at the ZI via GABAergic projections⁶⁹, and M1 stimulation limited the priming of TC loops for nociception by activating the pain-suppressing incerto-thalamic pathway in a rodent model of central pain^{74, 79}. However, a recent study with the aid of cell-type manipulations confirmed the role of ZIv parvalbumin neurons in promoting nocifensive behaviors via inhibitory projections to the Po⁸⁰. Gan et al., using optogenetics, further demonstrated that activating the M1-ZI pathway suppressed neuropathic allodynia in mice, and M1 activation preferentially enhanced incertal glutamatergic outputs but deactivated parvalbumin neurons. Transsynaptic anterograde tracing showed that incertal neurons that received direct M1 inputs are almost non-GABAergic and projected to the RVM. Thus, M1 activation may simultaneously recruit descending pain inhibition system indirectly via ZI (Figure 1a) and curb the pain-facilitating ZIv-Po pathway for pain relief²⁴. An intra-incertal GABAergic circuit⁸¹ may underlie the inhibition of parvalbumin-expressing neurons in the context of MCRA.

As a thin sheet of GABAergic neurons covering the lateral aspect of the thalamus, TRN integrates inputs from cortico-thalamic neurons and thalamic relay neurons, and in turn provides inhibitory outputs to all individual thalamic relay nuclei for curbing thalamic information flow⁸². This inhibition enables thalamocortical operations to flexibly accommodate to diverse behavioral needs⁷⁶. TRN controls pain processing by virtue of its inhibitory projections toward the ventrobasal thalamus in rodents⁸³. Considering that M1 sends direct projections toward TRN^{35, 84}, Kobaïter-Maarrawi et al. postulated that TRN served as the key intermediate structure whereby M1 yields inhibitory effects on VPL hyperactivity during chronic pain (Figure 1b)⁸⁵.

Thalamic INs, or thalamic GABAergic neurons residing outside of TRN, represent a third form of thalamic inhibition, but their specific function remains largely unexplored⁷⁶. Since somatosensory thalamic nuclei in rodents are nearly devoid of thalamic INs in contrast to those of cats and primates⁷⁷, the role of this population in MCRA seldom received attention in early preclinical studies until Kobaïter-Maarrawi et al. discussed this issue. They observed that EMCS induced depression of the firing rate of VPL WDR neurons, with concomitant activity enhancement of non-nociceptive (NN) cells in naive cats⁸⁵. These changes were inconsistent with the suppressed activities of both VPL WDR and NN neurons induced by EMCS in rats⁴⁹. The authors attributed this discrepancy to species differences in the presence of thalamic INs. In rats, the depressive activity of these neurons after EMCS is consistent with GABAergic control exerted by extrinsic inhibitory inputs. Whereas in higher animals, unlike thalamic WDR neurons receiving spinothalamic tract (STT) afferents, NN cells in contact with medial lemniscus terminals are under the control of thalamic INs which themselves are subject to extrinsic inhibitory inputs⁸⁶. Enhancing extrinsic inhibitory control via EMCS may bias this complex system toward increased NN cell activities through a disynaptic disinhibitory pathway⁸⁵.

Given neuroanatomical variations across species, caution is necessary when extrapolating data from rodents to other species. Extrinsic GABAergic sources occupy a dominant position for thalamic modulation in rodents, while the interplay between extrinsic inhibition and local circuit INs may add complexities to the modulation of thalamic relay neurons in higher animals. Importantly, the potential role of direct cortico-thalamic projections, with thalamic INs involved, in MCRA-related thalamic suppression should not be neglected in higher animals (Figure 1b). Further thalamic circuitry dissection will be a challenge for the next generation of MCRA studies.

Primary somatosensory cortex

Both S1 and second somatosensory cortex are key components of cortical pain processing⁸⁷. However, the latter has been rarely reported in MCRA studies (Supplemental Table 1), and thus is not discussed here. S1 receives both innocuous and noxious information through TC afferents. Under the condition of chronic pain, local inhibitory interneuron networks shift their activity in favor of pyramidal neuron hyperactivity^{88, 89}. Emerging data revealed a layer-specific and bidirectional role for S1 in modulating subjective sensory experiences. Unlike layer 5 output neurons, those in layer 6 exert pronociceptive and aversive effects, challenging the stereotypes that S1 generates strictly sensory aspects of pain and cortical pyramidal neurons exhibit no functional heterogeneity in pain processing⁹⁰. Further projection-specific dissections suggest that S1 projection neurons contribute to both sensory and pain-related affective behaviors via projections to the striatum, thalamus, and ACC^{90, 91} and facilitate only behavioral hypersensitivity by projections to the SDH³⁸.

Deafferentation pain refers to a subtype of neuropathic pain caused by complete or partial interruption of afferent impulses in the somatosensory transmission pathway⁹². The notion that S1 is involved in MCRA stems from early observations that deafferentation pain control provided by neuromodulation varies with stimulation targets. For instance, thalamic pain is refractory to peripheral nerve stimulation, spinal cord stimulation, and sensory thalamus stimulation instead of EMCS. There is documentation that somatosensory neurons below the level of deafferentation could not exert normal inhibitory control over deafferented nociceptive neurons, while better pain relief would be provided by stimulation at a level more rostral to the deafferented nociceptive neurons⁶⁷. As per the ‘‘surrounding inhibition’’ concept, the activation of S1 non-nociceptive neurons by innocuous afferents leads to the inhibition of a large surrounding region of nociceptive neurons via intrinsic connections⁹³. Based on this hypothesis, Tsubokawa et al. proposed that the rostral level superiority enables M1 stimulation to exert inhibitory effects on cortical deafferented nociceptive neurons through activation of hypothetical fourth-order sensory neurons in S1⁶⁷ (Figure 3). Tantalizing support for this notion comes from studies demonstrating that M1-S1 projections contribute to sensorimotor integration in a somatotopic manner through broad employment of excitatory and inhibitory cell populations in S1^{94, 95}. Hence, it would be intriguing to ascertain if the cortical circuit from M1 plays a role in sensory gating within S1 during MCRA.

Figure 3. A proposed model for control of nociceptive systems by non-nociceptive somatosensory systems during MCRA.

M1 stimulation may facilitate non-nociceptive sensory transmission directly at cortical, thalamic and medullary levels via cortico-cortical or corticofugal projections, and indirectly at the level of SDH where second-order neurons of the STT and the indirect PSDC pathway reside via descending modulation systems. In turn, non-nociceptive sensory systems may compete with the transmission of nociceptive sensation along the route toward the cortex. Abbreviations: DCN, dorsal column nuclei; DRG, dorsal root ganglion; M1, primary motor cortex; PSDC, postsynaptic dorsal column neurons; S1, primary somatosensory cortex; SDH, spinal dorsal horn; STT, spinothalamic tract; Tha, thalamus.

To date, inconsistencies remain in clinical observations regarding the role of S1 in MCRA. Garcia-Larrea et al. failed to observe any hemodynamic responses or electrophysiological changes in S1 after EMCS in NP patients^{22, 23}. In another study performed in healthy adults, they noticed an expansion of cortical hand representation in S1 concomitant to increased pain threshold in the contralateral hand after M1-rTMS, but no significant correlation could be drawn. Therefore, they speculated that S1 might only serve as a “marker” of the ability of M1 to exert influences over cortical networks via cortico-thalamo-cortical loops⁹⁶. However, a recent study advocating for the involvement of S1 in MCRA demonstrated a correlation between CRPS pain ratings following M1-rTMS and functional changes in S1⁹⁷. Correspondingly, EMCS has been shown to suppress metabolic signals in S1 in NP rats⁹⁸ and ipsilateral somatosensory evoked potentials in naïve rats⁹⁹. It should be noted that the S1 activity changes described above were observed with the STT or thalamus spared. In this case, the deactivation of S1 may be just an epiphenomenon of inhibitory control of M1 stimulation over pain processing along the STT. Supporting this notion, decreased VPL-S1 functional connectivity was found to be linked with M1-tDCS analgesia in FM patients⁶³.

Non-nociceptive sensory systems

The relationship between the antalgic efficacy of M1 stimulation in NP and sensory discrimination changes in the painful zone emanated from the work of Lefaucheur’s group. In 2002, they noted that good responders to EMCS could be identified by the absence of alteration of thermal sensation or abnormal thermal thresholds that could be improved by EMCS within the painful area¹⁰⁰. Later, people further observed that pain amelioration by M1-rTMS improved thermal discrimination within the painful zone in patients with NP, with the improvement of warm threshold found to correlate with the magnitude of reported analgesia^101,102. Taken together, MCRA may control pain transmission by reinforcing non-noxious thermal pathways, at least when these afferents are partially preserved¹⁰⁰.

The involvement of non-noxious mechanical sensations in MCRA remains a matter of debate. While several studies indicated no significant improvement in tactile, pressure, and vibratory discrimination along with MCRA in patients with NP^100–102, anecdotal evidence supported the restoration of tactile sensation in a case of post-traumatic facial NP¹⁰³. Interestingly, Reyns et al. linked the restoration of defective post-movement beta synchronization, an indicator of sensorimotor deactivation by cortical processing of movement-related proprioceptive and cutaneous inputs, with EMCS-related NP relief¹⁰⁴. In this context, it is worth revisiting another hypothesis suggesting central pain arises due to a mismatch of normal interaction between motor intention and sensory feedback¹⁰⁵. Thus, EMCS may relieve pain by counterbalancing sensory feedback deficiency at the cortical level¹⁰⁴. Concurring with Kobaïter-Maarrawi’s report⁸⁵, early animal studies have shown that electrically stimulating M1 selectively amplified the signal/ratio of exteroceptive messages from the moving body segments through surrounding inhibition at bulbar and thalamic lemniscal relays in cats, which was considered to improve the discriminative somatosensory aspect during movement for sensory-motor integration¹⁰⁶. In this scenario, such cortical reinforcement of movement-related sensory signals might underlie MCRA.

Overall, the evidence in support of the role of non-nociceptive sensory systems in MCRA is based on clinical observations instead of mechanistic studies. Therefore, the enthusiasm for this hypothesis may be tempered. Nevertheless, here we would like to propose a circuitry model linking this system to MCRA in Figure 3. As we know, neurocircuitry underpinnings of brain modulation of innocuous mechanical and thermal sensations remain largely a virgin land in neuroscience, mainly owing to the lack of validated behavioral paradigms in rodents. Although it is easy to speculate that M1 may facilitate sensory transmission at the level of dorsal column nuclei, thalamus, and S1 via direct excitatory projections, disynaptic or multisynaptic routes with GABAergic disinhibitory mechanisms involved as illustrated at the level of thalamus⁸⁵ should not be ignored. In addition, considering supraspinal descending modulation of pain⁴², it is tempting to speculate that touch and thermoreception are also subject to descending facilitation that could be recruited by MCRA. Then, the compensation of non-nociceptive somatosensory transmission may compete with nociceptive signaling at multiple levels of the neuroaxis (Box 3). This scheme is streamlined for the sake of simplicity, and we are expected to see additional layers of complexity in the crosstalk between these sensory modalities.

Cortico-limbic emotional systems

Accumulating imaging data have shown that M1 stimulation induced hemodynamic responses in other cortico-limbic pain signatures, with the most consistent effects occurring in the ACC, IC, and prefrontal cortex (Supplementary Table 1). Accordingly, there is a growing stream of morphological^{24, 30, 48, 107} and functional^{73, 108} evidence in support of their involvement in MCRA in animal models of NP. However, still is less clear the mechanism of action by which M1 stimulation influences these regions. Apart from the cortico-thalamo-cortical loop, imaging data implied that M1 might affect cortical areas via intracortical connections¹⁰⁹, despite the scarce anatomic connectivity^{35, 84}. Instead, opioidergic mechanisms may better explain the effects of MCRA on distant structures. The involvement of opioidergic signaling in MCRA was first reported by Maarrawi et al. who observed that clinical impacts of EMCS in NP patients positively correlated to preoperative opioid-receptor availability in the PAG, thalamus, IC, ACC, and orbitofrontal cortex¹¹⁰, as well as changes in opioid receptor availability in the ACC and PAG during therapy¹¹¹. These discoveries have parallels in NIBS paradigms corroborating the importance of endogenous opioid signaling in a wide brain network for MCRA^{57, 112}. Our recent study further documented the association of mu-opioid receptor (MOR) gene polymorphisms with outcome improvements in patients with knee osteoarthritis after M1-tDCS¹¹³. Later, the elevation of serum β-EP levels after motor cortex stimulation was seen in patients and animal models with various chronic pain subtypes, although debate existed regarding its correlation with pain and mood improvement^114–118. In preclinical studies, the causal relationship has been substantiated by pretreatment of opioid antagonists systematically^{99, 119, 120} or within the PAG¹²¹. Regarding opioidergic actions in pain-related regions (Box 5), we believe it is an integral part of the mechanism whereby MCRA alleviates emotional disorders of chronic pain and enables direct (actions on PAG) or indirect (via TC and corticofugal pathways) activation of descending pain control.

BOX 5. Opioid modulation of pain circuitry.

Cellular mechanisms:

Here we address opioid analgesia at thalamic, cortical, and midbrain levels, which are beneficial toward our understanding of neurocircuitry underpinnings of MCRA. Opioids blunt both sensory and emotional aspects of pain by modulating neuronal activity in the central nervous system^{125, 187}. Mu-opioid receptors (MORs) responsible for a major proportion of opioid analgesia are widely distributed in thalamic areas and their cortical targeting areas, as well as the descending pain control pathway¹⁸⁸. At the cellular level, opioids binding to MORs presynaptically inhibit voltage-gated Ca²⁺ channel opening and postsynaptically activate G-protein-coupled inwardly rectifying K⁺ channels, resulting in reduced neurotransmitter release and membrane hyperpolarization, respectively¹⁸⁹. Endomorphin and β-endorphin are canonical, endogenous MOR-preferring ligands^{189, 190}. Endomorphin-containing cell bodies primarily reside in the hypothalamus and nucleus tractus solitarii, with their terminals distributed throughout the central nervous system¹⁹⁰. β-endorphin-expressing neurons in the arcuate nucleus release β-endorphin mainly into the PAG through direct projections and the blood through the pituitary gland upon activation¹⁹¹.

Opioid actions in the thalamus:

Behavioral analgesia related with thalamic MORs has been well-documented in both lateral and medial/midline thalamic nuclei. Emerging evidence have revealed thalamic MOR-mediated downregulation of excitatory activity in midline/medial thalamic pathways to the amygdala and ACC, which may account for opioid-induced amelioration in emotional states associated with chronic pain (see refs. in supplemental materials). However, opioids act via presynaptic MORs to inhibit the tonic inhibitory action of GABAergic inputs on submedius thalamic nucleus neurons that project to the OFC¹⁹². Hence, opioidergic modulation of thalamic neurons may be complicated by their projections toward cortico-limbic areas, with facilitation effects on thalamic projections to pain-inhibiting areas but opposing effects on those to pain-facilitation areas.

Opioid actions in the ACC:

As a cortical hub for affective pain processing, ACC integrates pain signals from the mediodorsal thalamic nucleus, amygdala and primary somatosensory cortex, and regulates emotional qualities of pain via distinct pathways differentially from behavioral hypersensitivity, with projections to mesolimbic dopaminergic regions for the generation of aversiveness and to the spinal dorsal horn, striatum, and PAG for behavioral hyperalgesia facilitation (see refs. in supplemental materials). Converging imaging evidence suggest that ACC-PAG connections also mediate placebo analgesia, attentional analgesia and MCRA by engaging descending pain control^{47, 123, 193}. ACC glutamatergic neurons are effectors of nociceptive responses, which are negatively modulated by local GABAergic INs^{91, 144}. MOR signaling in the ACC is associated with affective pain relief¹⁹⁴, and mediates reduction of ongoing pain by placebos and non-opioidergic analgesics^{91, 125}. Mechanistic studies in rodents have shown that MOR activation inhibits excitatory glutamatergic transmission in the ACC through presynaptic mechanisms. Also, cingulate opioid signaling induces downstream activation of dopaminergic neurotransmission in the nucleus accumbens, thus mediating the resolution of pain aversiveness by non-opioid analgesics¹⁹⁴.

Opioid actions in the IC:

As another cortical site for pain facilitation, insular subregions are proposed to be involved in isolated pain networks, the posterior sensory circuit and the anterior emotional network, based on the connectivity pattern¹⁷². The activation of IC pyramidal neurons is necessary for behavioral hypersensitivity, while GABAergic interneurons promote antinociception^{172, 195}. So far, insular circuit mediating nociception is largely unclear, with only an afferent midcingulate-posterior insula pathway identified that maintains nociceptive hypersensitivity through recruitment of descending 5-HTergic facilitation⁹¹. Regarding opioid analgesia, insular MOR activation specifically suppresses inhibitory synaptic transmission toward local INs through presynaptic mechanisms¹⁹⁶, thus leading to the depression of excitatory transmission and subsequent recruitment of descending inhibitory systems¹⁹⁷.

Opioid actions in the OFC:

Unlike IC and ACC, OFC negatively controls affective pain in which the activation of glutamatergic neurons ameliorates both sensory and anxiodepressive consequences of neuropathic pain^{198, 199}. Activation of excitatory inputs from the ventromedial thalamus and submedius nucleus into the OFC mediates analgesia, as does boosting the output from the OFC to the PAG^{198, 199}. For opioid analgesia, opioids act via presynaptic MORs to inhibit the tonic inhibitory action of GABAergic inputs on OFC neurons projecting to the PAG¹⁹².

Opioid actions in the PAG:

Another well-described action mode of central opioid analgesia is attributed to actions along descending brainstem-spinal dorsal horn pathways, among which the PAG serves as a key generator of central opioidergic pain-suppressing system^{42, 189, 200}. GABAergic neurons within the PAG act as a critical site of action by opioids. The combined action of MORs at pre- or post-synaptic sites by opioids is a decrease in GABAergic neuronal activity, and therefore the disinhibition of excitatory outputs from the PAG to the RVM for descending pain control¹⁸⁹. The relationship between MCRA and opioid signaling in the PAG was initially observed in neuroimaging studies^{110, 111}, and further substantiated by local pharmacological interventions in rats¹²¹. Interestingly, the activation of PAG neurons under the condition of EMCS was accompanied with decreased expression of GABA within the same area⁴⁹. The inhibition of GABAergic INs might be owing to the recruitment of local opioidergic signaling. Collectively, these data lend support to the notion that M1 stimulation may enhance the efferent activity of PAG through opioid-mediated “GABAergic disinhibition”.

Graphical illustration:

The figure showcases brain mechanisms of opioids analgesia under the condition of MCRA. a. Placebo effects recruit endogenous brain opioidergic signaling in the clinical setting of M1 stimulation. The treatment context that are perceived and interpreted by the patient’s brain triggers placebo effects, which is characterized by the activation of brain opioidergic circuitry. The released endogenous opioids then terminate pain signaling via presynaptic and postsynaptic mechanisms. b. Schematic showing suggested brain circuitry underlying opioid-induced analgesia by MCRA. Green circles represent sites where opioidergic activity bears a relation with MCRA. Thalamic MOR activation inhibits thalamic inputs to the pyramidal neurons in pain-facilitating cortical areas including the ACC and IC, but does the opposite in the pain-inhibiting area OFC. These cortical activity changes in turn engage the descending PAG-RVM pathway for pain control. c. Current knowledge of opioidergic modulation of brain pain circuits. Locally released opioids bind to presynaptic MORs and inhibit the release of glutamate from thalamic excitatory inputs, thus inhibiting the activity of ACC principal neurons. The activation of insular MORs specifically alleviates inhibitory inputs toward local GABAergic neurons which exert tonic inhibition on pyramidal neurons, thus indirectly inhibiting insular output activity. Opioids act via presynaptic MORs to inhibit neurotransmitter release from GABAergic terminals in the OFC and PAG, thus indirectly exciting descending output neurons. In addition, opioids in the PAG via postsynaptic MORs also directly inhibit local GABAergic neurons. Abbreviations: ACC, anterior cingulate cortex; GABA, gamma-aminobutyric acid; Glu, glutamate; IC, insular cortex; MOR, mu-opioid receptor; Tha, thalamus; OFC, orbitofrontal cortex; PAG, periaqueductal gray; RVM, rostral ventromedial medulla.

The lack of direct projections from M1 to opioidergic neuron-harboring brain territories (Box 5) excludes potential neurocircuitry mechanisms whereby MCRA employs endogenous opioid systems. In this scenario, the introduction of the concept of placebo analgesia remains the most impressive way to comprehend this issue since it engages similar brain correlates to MCRA. First, placebo analgesia involves the consistent increase in fMRI activation of higher-order cognitive networks (e.g., the ACC and prefrontal cortex) and descending pain control systems (e.g., the PAG) and the strengthening of functional connectivity between the ACC and PAG, all of which are associated with individual variation in placebo analgesia^{122, 123}. Second, placebo analgesia engages endogenous opioidergic signaling, the strength of which is associated with enhanced MOR activities in the ACC, IC, and PAG^122–124. Given the functional converging shared by MCRA and placebo, an emerging consensus is that MCRA induces analgesia partially via placebo effects.

Placebo analgesia refers to beneficial effects attributable to brain-mind responses to the context surrounding treatment rather than to the biological properties of the treatment itself. It is mainly shaped by cognitive expectation of the efficacy of the “novel” treatment, and by unconscious associative learning whereby prior positive experience leads to positive effects of the present treatment^{122, 123, 125}. Although clinical data have ascertained the superior analgesic effects of real vs sham stimulation, intrinsic placebo action does play a role in the benefits of M1 stimulation^{15, 126}. For refractory pain patients desperate to obtain pain relief, MCRA, as the last resort, generates the expectancy of favorable outcomes in the clinical psychosocial context. Provider-patient interactions, positive emotions, and treatment effects (e.g., paraesthesia) also contribute to placebo responses¹²⁷. In addition, the importance of placebo timing in placebo effects driven by M1-rTMS has been evaluated in chronic pain treatment. Sham rTMS induced significant analgesia when preceded by a successful active rTMS, whereas tended to worse pain when following an unsuccessful rTMS, suggesting the possible role of conditioned learning in placebo modulation¹²⁷.

With placebo effects clearly present in M1 stimulation, the specific action of the mode of real stimulation remains a significant scientific issue. In a study series, DaSliva and colleagues tended to discriminate regions and pathways that primarily drove the effects of real and sham M1 stimulation with tDCS^{63, 128, 129}. Regarding MOR activation, they noted that initial placebo and subsequent real M1-tDCS induced similar and positively correlated MOR activations in PAG and precuneus, as well as dissimilar activations in the thalamus and prefrontal cortex respectively in a cohort of healthy subjects. Although opioid release occurred with both stimulations, significant analgesic effects in thermal pain were only observed after real tDCS. These discoveries support the view that MCRA can be partially attributed to the recruitment of similar endogenous opioidergic mechanisms induced by placebo, which can be purposely fortified at both molecular and clinical levels with real tDCS subsequently added to the placebo experience¹²⁸. Thus, the value of heightening patient expectations could be considered if tDCS alone produces limited clinical benefits, since exogenous priming might facilitate real tDCS to ignite endogenous pain inhibitory system¹³⁰.

Other limbic structures worth mentioning include the nucleus accumbens, hippocampus, and amygdala (AMY), covering cognitive and affective dimensions of chronic pain¹³¹. Recently, a novel connection from M1 to the nucleus accumbens reward circuitry through a M1 layer 6-mediodorsal thalamic (MD) nucleus pathway was unraveled (Figure 1b). Unlike M1 layer 5 connections with the ZI and PAG exclusively suppressing behavioral sensitization, this pathway specifically suppressed negative affect associated with NP²⁴. The increased medial temporal metabolism was presented in patients with FM after M1-rTMS¹³² and patients with CRPS and NP after EMCS^{23, 133}. In particular, hippocampal activation was associated with improvements in emotional pain¹³². FOS immunostaining also substantiated the activation of AMY by EMCS in experimental NP models^{48, 107}. Recent imaging evidence further showed that M1-rTMS normalized altered functional connectivity between ipsilesional MD and AMY in CPSP monkeys¹³⁴. Anatomically, MD and AMY have reciprocal connections, presumed to convey the emotional significance of events¹³⁵. One major limitation in current MCRA imaging studies is the ambiguity in signaling hierarchy and network directionality, rendering it challenging to extrapolate findings from human imaging to rodent neurocircuitry studies. Hopefully, technological advances in imaging analysis may bring exciting possibilities for tackling these pitfalls¹³⁶.

Conclusions

M1 serves as a gateway for chronic pain control, spanning various loci along the neuroaxis from the SDH to the cortex. At the level of neurotransmitters, MCRA acts not just via glutamatergic and GABAergic signaling, but through a myriad of dopaminergic, serotoninergic, and opioidergic systems. At the network level, MCRA inhibits spinal pain processing through direct or indirect (via the basal ganglia, ZI, and brain opioidergic circuitry) activation of descending control systems, and modulates broad cortical/subcortical pain-related nuclei via the thalamus. At the system level, MCRA may reinforce non-noxious sensory transmission for cross-modal pain modulation at multiple levels of the somatosensory pathway, and neuropsychological processes underpinning placebo effects yield further benefits. Behaviorally, MCRA improves chronic pain across sensory and affective dimensions, enhancing patients’ overall well-being.

There are several limitations to this summary. First, the theoretical framework of motoric pain modulation provided herein is roughly based on M1 stimulation since behavioral activation procedures are nearly devoid of mechanistic understandings yet. Although electrical stimulation modulates brain function bidirectionally – both orthodromically and antidromically¹³⁷ – we have not delved into the potential role of M1 afferents. Second, considerable differences may exist in MCRA mechanisms between experimental pain and pathological pain owing to the effect of disease-related brain plasticity processes on the recruited circuits¹³⁸. Different pathophysiology of chronic pain subtypes further adds an extra layer of complexity to MCRA. This is especially exemplified in the case of NP, with the integrity of the nervous system compromised on different levels to varying degrees, in which MCRA may opt for differing circuits to control pain. Third, neuro-glial interactions along peripheral and central pain pathways fulfill key roles in pathological pain processing via neuroinflammatory signaling¹³⁹. While our understanding of MCRA here has primarily revolved around neuronal mechanisms, it is encouraging to see the growing recognition of glia’s role in MCRA mechanisms^{118, 140}. Fourth, there exists huge heterogeneity in prior mechanistic studies which could be explained by clinical heterogeneity across cohorts, the diversity of stimulation modality, methodological variability, and discrepancy in behavioral readout. Now people know far less about MCRA neurocircuitry associated with emotional improvement than that associated with sensory improvement. The actual landscape of MRCA-related neurocircuitry might be more intricate than that we have depicted here.

Current perspectives concerning the interplay between pain and motor systems have illustrated how pain interferes with motor functions via actions on multiple levels of the motor pathway. Although the participation of M1 in the development of chronic pain has not been fully clarified, different changes in M1 excitability in acute and chronic pain are believed to carry different behavioral values^{25, 141}. This review interprets the flip side of this interrelationship – how M1 activation alters pain perception and associated emotional disorders. To date, the motor system has not been considered part of the neural correlates of pain processing, despite frequent activation occurring in M1, striatum, and cerebellum reported during noxious stimuli⁸⁷. In light of the growing understanding of MCRA, we propose that as the pivotal hub integrating endogenous pain and motor interaction, M1 mediates the locomotion dimension of pain and serves as a promising target for rehabilitative strategies for chronic pain and locomotor comorbidities.

The harnessing of MCRA represents a paradigmatic shift in chronic pain management. Breakthroughs in our understanding of neural correlates of MCRA will endow us with a deeper apprehension of cortical pain processing and inevitably spur the optimization of this procedure and personalized interventions. The neurobiological basis underlying the analgesic effects reported with M1 stimulation remains to be illuminated, making this an exciting time to engage in this field.

Abbreviations:

5-HT

serotonin

ACC

anterior cingulate cortex

AMY

amygdala

CPSP

central post-stroke pain

CPM

conditioned pain modulation

CRPS

complex regional pain syndrome

CST

corticospinal tract

D₂R

dopamine receptor D2

EMCS

epidural M1 stimulation

IN

inhibitory interneurons

LC

locus coeruleus

M1

primary motor cortex

MCRA

motor cortex-related analgesia

MD

mediodorsal thalamic nucleus

MOR

Mu-opioid receptor

NIBS

non-invasive brain stimulation

NP

neuropathic pain

PAG

periaqueductal grey

Po

posterior thalamic nucleus

rCBF

cerebral blood flow

rTMS

repetitive transcranial magnetic stimulation

RVM

rostroventromedial medulla

S1

primary somatosensory cortex

SDH

spinal dorsal horn

STT

spinothalamic tract

SRD

subnucleus reticularis dorsalis

tDCS

transcranial direct current stimulation

TC

thalamocortical

TRN

thalamic reticular nucleus

VPL

ventral posterior lateral nucleus

VPM

ventral posterior medial nucleus

WDR

wide dynamic range

ZI

zona incerta

ZIv

ventral ZI