Electroacupuncture activates inhibitory neural circuits in the somatosensory cortex to relieve neuropathic pain.

Electroacupuncture (EA) has been accepted to effectively relieve neuropathic pain. Current knowledge of its neural modulation mainly covers the spinal cord and subcortical nuclei, with little evidence from the cortical regions. Using in vivo two-photon imaging in mice with chronic constriction injury, we found that EA treatment systemically modulated the Ca2+ activity of neural circuits in the primary somatosensory cortex, including the suppression of excitatory pyramidal neurons, potentiation of GABAergic somatostatin-positive interneurons, and suppression of vasoactive intestinal peptide-positive interneurons. Furthermore, EA-mediated alleviation of pain hypersensitivity and cortical modulation were dependent on the activation of endocannabinoid receptor 1. These findings collectively reveal a cortical circuit involved in relieving mechanical or thermal hypersensitivity under neuropathic pain and identify one molecular pathway directing analgesic effects of EA.

Introduction

Electroacupuncture (EA) effectively relieves various types of pain disorders as suggested by large amounts of clinical evidence (Ulett et al., 1998; Hinman et al., 2014; MacPherson et al., 2017; Scarborough and Smith, 2018; Hershman et al., 2018; Witt et al., 2005). To provide mechanistic explanations for the analgesic effects of acupuncture, classical views argue for the involvement of endorphins (Clement-Jones et al., 1980; Ho and Wen, 1989), and recent studies suggest the possible involvement of central opioids, monoamines, and neuropeptides (Zhang et al., 2014; Kim et al., 2009; Han, 2003). In particular, endocannabinoid receptor 1 (CB1R) pathway was recently found to be involved in central analgesic effects of EA (Yuan et al., 2018; Shou et al., 2013). All these neuromodulators may affect neural plasticity of the sensory pathway in the spinal cord (Xing et al., 2007; Koo et al., 2008) and brain (Zhao, 2008; Zhang et al., 2012) to reduce pain hypersensitivity. With regard to specific brain regions, EA-induced analgesia can be contributed by different subcortical nuclei, such as periaqueductal gray (PAG) (Guo et al., 2004), preoptic area (Ma et al., 2011), and habenular nucleus (Lee and Beitz, 1993). However, the somatosensory cortex has drawn little attention until a recent human brain imaging study suggested the rewiring of primary somatosensory cortex (S1) after continuous EA intervention (Maeda et al., 2017). Therefore, EA's analgesic functions may also involve the modulation of cortical circuits.

Pyramidal neurons (PNs) in S1 receive and send projections from and to various cortical and sub-cortical nuclei (Aronoff et al., 2010), making S1 a higher modulatory center for spatial and graded pain sensations (Bushnell et al., 1999). Recent findings further suggest the participation of S1-originated corticospinal tract (CST) in innervating dorsal horns of spinal cord to mediate pain sensitivity (Liu et al., 2018), illuminating a top-down modulatory pathway. Under neuropathic pain, both humans (Alshelh et al., 2016) and animal models (Liu et al., 2018; Cichon et al., 2017) have displayed the hyperactivity of S1 neurons, suggesting that the modulation of S1 activity might alleviate pain disorders.

Within S1, PNs are innervated by GABAergic interneurons (INs), forming a local circuit to code sensory inputs (Harris and Mrsic-Flogel, 2013; Huang and Paul, 2019). Two-photon in vivo imaging approach has illustrated the hyper-activation of layer 5 pyramidal neurons (L5PNs) and inhibition of INs of S1 in mice with neuropathic pain (Cichon et al., 2017). More importantly, the activation of somatostatin-positive INs (SST-INs) effectively reduced allodynia and hyperalgesia (Cichon et al., 2017). Such information provides a panoramic view of cortical neural circuits for pain sensation, and since the activation of cortical GABAergic transmission is sufficient to alleviate chronic pain behaviors (Eto et al., 2012), we thus propose that EA might modulate the excitatory-inhibitory (E/I) balance of S1 local circuits to alleviate neuropathic pain.

To test this hypothesis, we performed two-photon imaging of neuronal Ca2+ activities on both PNs and INs of S1, in a mouse neuropathic pain model receiving persistent EA treatment. In addition to reducing mechanical or thermal hypersensitivity in pain model animals, EA effectively suppressed the hyperactivity of L5PNs, and potentiated cortical GABAergic networks. Furthermore, we found that the CB1R activation was necessary for EA-mediated analgesic effects. These results collectively reported a cortical mechanism of EA in alleviating neuropathic pain.

Results

EA relieves mechanical and thermal hypersensitivity under neuropathic pain

To substantiate analgesic effects of EA on patients with neuropathic pain, we adopted a mouse chronic constriction injury (CCI) model. One week after CCI surgery, mice were subjected to 7-day EA treatment (one session per day). During each EA treatment session, two acupoints, GB30 (Huantiao) and GB34 (Yanglingquan), were inserted with needles connecting to a pulsed current (amplitude: 1 mA; duration: 0.1 ms; frequency: 2 Hz; total length: 30 min; Figures 1A and 1B; see Transparent methods for more details). We found that either CCI surgery or EA treatment did not significantly change the animals' general locomotor activity or anxiety levels (Figure S1). When examining the mechanical pain threshold, CCI mice rapidly and persistently presented hypersensitivity under von Frey or brush stimuli test. In contrast, mice with EA treatment showed a marked elevation of mechanical pain thresholds (Figures 1C and 1D). The sham EA treatment (by inserting two needles into the adjacent regions of acupoints without applying the current) did not result in significant improvements of mechanical hypersensitivity (Figures 1C and 1D), suggesting specificity of EA-induced pain-relieving effects.

Figure 1

Electroacupuncture effectively relieves mechanical and thermal hypersensitivity

(A) Experimental outlines for (C–G). Male C57BL/6 mice received chronic constriction injury (CCI) surgery on D0 and received daily electroacupuncture (EA) for 7 days starting from D7. Both mechanical (von Frey and brush) and thermo-related (acetone and heat-cold plate) pain behaviors were continuously monitored for 14 days.

(B) A representative picture for EA treatment on mice. Two acupoints, GB30 (Huantiao) and GB34 (Yanglingquan), were adopted using standard acupuncture needles connected with electrical currents (1 mA, 2 Hz).

(C) Mechanical threshold of hindlimb retraction reflex using von Frey apparatus from D1 to D14 (two-way ANOVA, group f actor: F(3, 32) = 125.7, p < 0.001; time factor, F(11, 352) = 61.72, p < 0.001; ∗∗∗p < 0.001 using Bonferroni post-hoc comparison; ns, no significant difference; n = 7, 8, 12, and 9 animals for sham CCI, CCI, CCI + EA, and CCI + sham EA group, respectively).

(D) Total scores for the brush-induced reflex, with higher score indicating stronger response (two-way ANOVA, group factor: F(3, 32) = 162.4, p < 0.001; time factor, F(11, 352) = 23.60, p < 0.001; ∗∗∗p < 0.001 using Bonferroni post-hoc comparison; ns, no significant difference; n = 7, 8, 12, and 9 animals for sham CCI, CCI, CCI + EA, and CCI + sham EA group, respectively).

(E) Scores for acetone-induced reflex on D14, with higher score indicating stronger response (1-way ANOVA: F(3, 31) = 62.91, p < 0.001; ∗∗∗p < 0.001 between sham CCI and CCI group; ∗∗p = 0.0024 between CCI and CCI + EA group; ns, no significant difference between CCI and CCI + sham EA group; Tukey's post-hoc comparison was used; n = 8, 8, 10, and 9 animals for sham CCI, CCI, CCI + EA, and CCI + sham EA group, respectively).

(F) Lowest temperature (heat threshold) of limb retraction on the heat plate (two-way ANOVA, group factor: F(3, 32) = 41.26, p < 0.001; time factor, F(6, 192) = 31.53, p < 0.001; ∗∗∗p < 0.001 using Bonferroni post-hoc comparison; ns, no significant difference; n = 7, 8, 12, and 9 animals for sham CCI, CCI, CCI + EA, and CCI + sham EA group, respectively).

(G) Threshold of mice on the cold plate test (two-way ANOVA, group factor: F(3, 32) = 118.0, p < 0.001; time factor, F(6, 192) = 62.25, p < 0.001; ∗∗∗p < 0.001 using Bonferroni post-hoc comparison; ns, no significant difference; n = 7, 8, 12, and 9 animals for sham CCI, CCI, CCI + EA, and CCI + sham EA group, respectively).

(H) Both heat and cold thresholds of mice on D14 were plotted to give the temperature sensation range (=heat threshold – cold threshold; one-way ANOVA: F(3, 32) = 73.40, p < 0.001; ∗∗∗p < 0.001 using Tukey post-hoc comparison; ns, no significant difference; n = 7, 8, 12, and 9 animals for sham CCI, CCI, CCI + EA, and CCI + sham EA group, respectively). Data are presented as mean ± SEM.

Given the existence of thermal hypersensitivity with neuropathic pain (Jensen and Finnerup, 2014), we investigated if EA also relieves such thermal stimuli-associated pain behaviors. Using the acetone dropping test, we found that EA, but not sham treatment, reduced the excessive occurrence of hindlimb retraction reflex in CCI mice (Figure 1E). Moreover, during a conventional heat/cold plate test, CCI model mice exhibited hypersensitivity by presenting lower thresholds upon heat stimuli, and higher thresholds when facing cold challenges, and EA treatment restored the normal cold-heat sensory range (Figures 1F–1H). These data collectively suggest that EA relieves both mechanical and heat/cold hypersensitivity under neuropathic pain.

EA treatment reduces pyramidal neuron hyperactivity in S1

To investigate the changes of neural circuits underlying the effects of EA in relieving pain hypersensitivity, we studied the activity of S1, which is potentiated across different neuropathic pain models (Cichon et al., 2017; Liu et al., 2018). Examining the expression of immediate early genes (IEGs), c-Fos, we found a significant increase in the number of activated neurons in S1 at 14 days after CCI surgery (D14), whereas EA treatment prevented such neuronal activation by CCI, especially at layer 5 (Figures S2A–S2C and Table S1). Double immunostaining further showed that a large number of excitatory neurons (Neurogranin+) in S1 were activated by CCI treatment and were inhibited after EA intervention (Figure S2D and Table S1). Therefore, EA potentially relieves the hyperactivity of excitatory neurons in S1 in neuropathic pain.

To provide in vivo evidence for the neuronal activity of S1, we employed two-photon Ca2+ imaging on L5PNs at the end of 7-day EA treatment (Figure 2A). By transfecting the fluorescent Ca2+ indictor GCaMP6s into L5PNs using an adeno-associated virus vector (Figure 2B), we recorded Ca2+ activities of pyramidal neuron somas in awake, head-fixed animals. From the continuous recording of 65∼80 live neurons per group (from 4∼5 animals per group), CCI animals had significantly higher Ca2+ activities compared with sham control group, whereas EA treatment remarkably alleviated Ca2+ hyperactivation of L5PNs (Figure 2C). Further data analysis showed that both total integrated Ca2+ activities (Figures 2D and 2E) and peak values of single Ca2+ spike (Figures 2H and 2I) were significantly elevated in CCI group, and were suppressed after EA but not sham EA treatment. On the other hand, the frequency of Ca2+ spikes (defined as the total spike number per minute) was not significantly changed across groups (Figures 2F and 2G).

Figure 2

EA prevented hyperactivity of L5PNs in vivo

(A) Schematic diagrams for experiments in (C to I). Male C57BL/6 mice received transfection of AAV-Camkii-GCaMP6s into the S1 hindlimb (S1HL) region on one side, and were prepared for CCI surgery on the contralateral side on D7. EA treatment was performed from D14, and two-photon imaging was performed at 24 h after the conclusion of EA (D22).

(B) Transfection sites showing GCaMP6s expression mainly in layer 5. Scale bar, 500 μm.

(C) Time-elapse heat maps showing Ca2+ activities of L5PNs during a 150-s recording window. All data have been normalized to resting levels (ΔF/F0). A total of 70 representative neurons were plotted in each group.

(D and E) Total integrated Ca2+ activities of L5PNs. (D) Bar chart showing averaged total integrated Ca2+ activities (sham CCI: 175.8 ± 11.13, CCI: 369.9 ± 21.45, CCI + EA: 193.3 ± 10.17, CCI + Sham EA: 388.0 ± 23.00; Kruskal-Wallis test statistic = 115.6, p < 0.001; ∗∗∗p < 0.001 using Dunn's post-hoc comparison; ns, no significant difference; n = 79, 67, 82, and 74 neurons, N = 4, 5, 5, and 4 animals for sham CCI, CCI, CCI + EA, and CCI + sham EA group, respectively). (E) Histogram showing frequency distribution of total integrated Ca2+ activities.

(F and G), Total Ca2+ spike number (per min) of L5PNs. (F) Averaged Ca2+ spike frequency (sham CCI: 1.88 ± 0.11, CCI: 2.06 ± 0.13, CCI + EA: 1.63 ± 0.07, CCI + sham EA: 1.56 ± 0.09; ns, no significant difference from Kruskal-Wallis test; n = 62, 66, 56, and 74 neurons, N = 4, 5, 5, and 4 animals for sham CCI, CCI, CCI + EA, and CCI + sham EA group, respectively). (G) Histogram for the frequency distribution of Ca2+ spike frequency.

(H and I) Peak values of Ca2+ activities in L5PNs.

(H) Averaged peak values of Ca2+ activities (sham CCI: 144.4 ± 15.30, CCI: 288.7 ± 17.69, CCI + EA: 104.2 ± 6.92, CCI + sham EA: 410.6 ± 25.62; Kruskal-Wallis test statistic = 141.8, p < 0.001; ∗∗∗p < 0.001 using Dunn's post-hoc comparison; n = 68, 73, 61, and 80 neurons, N = 4, 5, 5, and 4 animals for sham CCI, CCI, CCI + EA, and CCI + sham EA group, respectively).

(I) Frequency distribution of peak values of Ca2+ activities. Data are presented as mean ± SEM.

L5PNs vertically extend their apical dendrites to form highly branched networks in the cortical superficial layer, where large amounts of dendritic spines reside to receive presynaptic inputs from both cortical and sub-cortical nuclei (Manns et al., 2004). We next examined the Ca2+ activity of dendritic spines in apical tuft (Figure 3A). Similar to the findings of L5PNs, CCI treatment potentiated Ca2+ spikes in apical spines, whose activities were significantly suppressed by EA intervention (Figure 3B). Further analysis showed that total integrated Ca2+ values (Figures 3C and 3D), Ca2+ spike frequency (Figures 3E and 3F), and peak values (Figures 3G and 3H) were all elevated in CCI model, and EA intervention mainly suppressed the frequency but not peak values, suggesting possible presynaptic inhibition. Combining in vivo recording data from both soma and dendritic spines of L5PNs, EA treatment relieved the hyperactivity of Ca2+ spikes induced by CCI. As L5PNs are the major inter- and sub-cortical projecting cells in S1 (Manns et al., 2004), EA-induced modulation on L5PNs may further affect the activity of other brain regions in the pain pathway.

Figure 3

EA intervention modulates the activity of L5PN apical spines

(A) Schematic diagrams showing experimental timelines in (B–H), similar with those in Figure 2A.

(B) Top, representative apical dendrites (DS) and spines (S1 to S5) under two-photon imaging. Scale bar, 2 μm. Bottom, recording traces of Ca2+ activities after normalization (ΔF/F0). Pink shades indicate high levels of synchronization among spines on the same dendritic branch.

(C and D) Total integrated Ca2+ activities of apical spines. (C) Bar chart showing averaged total integrated Ca2+ activities (sham CCI: 118.9 ± 3.91, CCI: 250.3 ± 28.09, CCI + EA: 175.8 ± 4.13, CCI + sham EA: 202.1 ± 6.04; Kruskal-Wallis test statistic = 140.5, p < 0.001; ∗∗∗p < 0.001 using Dunn's post-hoc comparison; ns, no significant difference; n = 89, 163, 167, and 151 spines, N = 4, 5, 6, and 4 animals for sham CCI, CCI, CCI + EA, and CCI + sham EA group, respectively). (D) The frequency distribution of total integrated Ca2+ activities.

(E and F) Spike numbers per minute of Ca2+ transients. (E) Averaged Ca2+ spike frequency (sham CCI: 1.38 ± 0.07, CCI: 2.71 ± 0.12, CCI + EA: 1.59 ± 0.05, CCI + sham EA: 3.13 ± 0.16; Kruskal-Wallis test statistic = 131.0, p < 0.001; ∗∗∗p < 0.001 using Dunn's post-hoc comparison; ns, no significant difference; n = 82, 161, 158, and 116 spines, N = 4, 5, 6, and 4 animals for sham CCI, CCI, CCI + EA, and CCI + sham EA group, respectively). (F) The frequency distribution of Ca2+ spike frequency.

(G and H) Peak values of Ca2+ spikes. (G) Averaged Ca2+ peak values (sham CCI: 216.1 ± 15.01, CCI: 336.4 ± 14.52, CCI + EA: 312.1 ± 14.52 CCI + sham EA: 263.9 ± 11.19; Kruskal-Wallis test statistic = 46.66, p < 0.001; ∗∗∗p < 0.001 using Dunn's post-hoc comparison; ns, no significant difference; n = 100, 201, 192, and 138 spines, N = 4, 5, 6, and 4 animals for sham CCI, CCI, CCI + EA, and CCI + sham EA group, respectively). (H) Histogram showing the frequency distribution of Ca2+ peak values. Data are presented as mean ± SEM.

In addition to L5PNs, we also recorded the activity of PN in layer 2/3 (L2/3 PNs) of S1. Both peak values and frequencies of Ca2+ spikes were increased by CCI treatment and were suppressed after EA intervention (Figure S3). In summary, EA treatment relieves the hyperactivity of PNs in S1 of mice with neuropathic pain.

EA normalized cortical inhibitory network

Cortical PNs are under a tight control of various subtypes of INs. As we found significantly elevated c-Fos activity of S1 GABAergic neurons in EA-treated mice (Figure S4 and Table S1), it should be valuable to monitor in vivo activities of INs. As SST-INs intensively innerve L5PN apical spines (Urban-Ciecko and Barth, 2016), whose Ca2+ activities were suppressed after EA (Figure 3), we first examined the activity of SST-INs in S1 after transfecting Cre-dependent GCaMP6s into those cells (Figures 4A and 4B). The soma of SST-INs in CCI group presented lower total Ca2+ activities, which were enhanced by EA treatment (Figure 4C). Specifically, such modulation on total integrated Ca2+ activities (Figures 4D and 4E) were mainly attributed to the EA-induced re-elevation of peak values (Figures 4H and 4I) but not the frequency of Ca2+ spikes (Figures 4F and 4G). These results suggested that CCI leads to the inhibition of SST-INs soma, and EA treatment potentiates GABAergic transmissions.

Figure 4

EA treatment potentiates somatostatin interneurons

(A) Experimental outlines in (C–I). Somatostatin (SST)-Cre mice received unilateral injection of AAV-dio-GCaMP6s virus into S1, followed by CCI surgery on D7 (contralateral side) and EA intervention. The activity of SST-interneurons (SST-INs) was recorded at the end of EA treatment (D22).

(B) Transfection sites of GCaMP6s. Scale bar, 100 μm.

(C) Heat maps showing normalized Ca2+ activities (ΔF/F0) of SST-INs soma during the 150-s recording period. A total of 45 representative cells were plotted for sham CCI and CCI group, and 30 representative cells were displayed for CCI + EA group.

(D and E) Total integrated Ca2+ activities in SST-INs soma. (D) Averaged total integrated Ca2+ activities (sham CCI: 254.3 ± 22.02, CCI: 170.0 ± 16.32, CCI + EA: 327.6 ± 38.03; Kruskal-Wallis test statistic = 19.09, p < 0.001; ∗∗p = 0.0031, ∗∗∗p < 0.001 using Dunn's post-hoc comparison; n = 63, 59, and 45 neurons, N = 4, 5, and 5 animals for sham CCI, CCI, and CCI + EA group, respectively). (E) Histogram for the frequency distribution of total integrated Ca2+ activities.

(F and G) Ca2+ spike frequency (spike number per min) of SST-INs soma. (F) Averaged Ca2+ spike frequency (sham CCI: 3.19 ± 0.35, CCI: 3.28 ± 0.26, CCI + EA: 3.75 ± 0.30; ns, no significant difference from Kruskal-Wallis test; n = 90, 84, and 70 neurons, N = 4, 5, and 5 animals for sham CCI, CCI, and CCI + EA group, respectively). (G) Histogram showing the frequency distribution of Ca2+ spike frequency.

(H and I) Peak values of Ca2+ activities in SST-INs soma. (H) Averaged peak values of Ca2+ activities (sham CCI: 242.6 ± 34.38, CCI: 146.3 ± 21.88, CCI + EA: 273.5 ± 35.60; Kruskal-Wallis test statistic = 16.01, p < 0.001; ∗∗p = 0.0079, ∗∗∗p < 0.001 using Dunn's post-hoc comparison; n = 81, 75, and 47 neurons, N = 4, 5, and 5 animals for sham CCI, CCI, and CCI + EA group, respectively). (I) Frequency distribution of peak values of Ca2+ activities. Data are presented as mean ± SEM.

We further tracked the activity of SST-INs axons, which form direct synaptic connections with apical spines of L5PNs (Urban-Ciecko and Barth, 2016). Similar phenotypes were identified by two-photon imaging, as total integrated Ca2+ activities and peak values, but not Ca2+ spike frequency, were remarkably inhibited in CCI group and were restored by EA treatment (Figure S5). The potentiated axonal Ca2+ activities indicate higher probability of GABA release at SST-PN synapse. These data agreed with previous reports (Cichon et al., 2017) by supporting a disinhibition of L5PNs under CCI, and further suggested that EA exerts its analgesic effects possibly by activating SST-INs to suppress PNs.

Within the local cortical circuit, vasoactive intestinal peptide-positive interneurons (VIP-INs) innervate SST-INs, adding one extra layer of activity regulation (Karnani et al., 2016). We next investigated if EA intervention also affected VIP-INs activities (Figures 5A and 5B). We found that VIP-INs displayed a similar phenotype as those in L5PNs: CCI remarkably potentiated their somatic Ca2+ activities, whereas EA treatment inhibited those neurons (Figure 5C). Total integrated Ca2+ activities (Figures 5D and 5E), Ca2+ spike frequencies (Figures 5F and 5G), and peak values (Figures 5H and 5I) were all increased in CCI model group and were depressed by EA intervention. We further recorded the activity of VIP-INs axons, which form synaptic connection with SST-INs (Karnani et al., 2016). Results showed similar patterns, in which the Ca2+ activity of VIP-IN axons was increased after CCI and was suppressed following EA intervention (Figures S6A–S6C). Such modulatory effects were mainly attributed to the changes of Ca2+ spike frequency but not peak values within the axonal region (Figures S6D–S6G), indicating potentiated GABA release at VIP-SST synapse. Taken together, both somatic and axonal data from cortical INs suggest a dual-layered modulation on cortical GABAergic activities, as EA treatment potentially suppresses VIP-INs, leading to the activation of SST-INs and the suppression of PNs as the consequence.

Figure 5

EA alleviates hyperactivation of vasoactive intestinal peptide-interneurons

(A) A brief experimental outline in C–I Vasoactive intestinal peptide (VIP)-Cre mice were unilaterally transfected with AAV-dio-GCaMP6s virus into S1, treated by CCI surgery on D7, and received EA intervention as previous experiments. The activity of VIP-interneurons (VIP-INs) was recorded at the end of EA treatment (D22).

(B) Transfection sites of GCaMP6s within S1. Scale bar, 100 μm.

(C) Heat maps for time series recording of normalized Ca2+ activities (in ΔF/F0) of VIP-INs soma. A total of 60–80 representative cells were plotted from each group and a 150-s recording period was adopted.

(D and E) Total integrated Ca2+ activities of VIP-INs soma. (D) A bar chart showing total integrated Ca2+ activities (sham CCI: 396.9 ± 23.22, CCI: 678.3 ± 44.26, CCI + EA: 377.8 ± 23.08; Kruskal-Wallis test statistic = 34.88, p < 0.001; ∗∗∗p < 0.001 using Dunn's post-hoc comparison; n = 87, 69, and 72 neurons, N = 4, 5, and 4 animals for sham CCI, CCI and CCI + EA group, respectively). (E) Histogram showing the frequency distribution of total integrated Ca2+.

(F and G) Ca2+ spike frequency (calculated as the spike number per min) of VIP-INs soma. (F) Averaged Ca2+ spike frequency (sham CCI: 2.15 ± 0.08, CCI: 3.56 ± 0.22, CCI + EA: 1.71 ± 0.13; Kruskal-Wallis test statistic = 56.89, p < 0.001; ∗∗∗p < 0.001 using Dunn's post-hoc comparison; n = 79, 52, and 64 neurons, N = 4, 5, and 4 animals for sham CCI, CCI, and CCI + EA group, respectively). (G) The frequency distribution of Ca2+ spike frequency.

(H and I) Peak values of Ca2+ activities of VIP-INs soma. (H) Averaged peak values of Ca2+ activities (sham CCI: 292.8 ± 19.26, CCI: 495.8 ± 42.00, CCI + EA: 279.2 ± 26.09; Kruskal-Wallis test statistic = 23.56, p < 0.001; ∗∗∗p < 0.001 using Dunn's post-hoc comparison; n = 93, 60, and 80 neurons, N = 4, 5, and 4 animals for sham CCI, CCI, and CCI + EA group, respectively). (I) Histogram showing the frequency distribution of peak values of Ca2+ spikes. Data are presented as mean ± SEM.

CB1R pathway underlies EA's modulation on cortical circuits and analgesia

After characterizing the cortical circuitry modulation by EA, we next explored the molecular mechanism. Using RNA sequencing to screen genes with differential expressions in S1 under both CCI and EA treatment, we adopted Kyoto Encyclopedia of Genes and Genomes database to analyze the functional clusters that were most significantly affected by EA. Results showed that the retrograde endocannabinoid signaling pathway was remarkably changed (Figure 6A). As the upstream regulator of this pathway, CB1R has been implicated in pain processing across cortical regions (Woodhams et al., 2017), and recent data also support the role of CB1R in midbrain nuclei for reversing pain hypersensitivity (Yuan et al., 2018). We thus investigated if CB1R plays a role in EA-mediated cortical circuitry modulation and relief of mechanical/thermal hypersensitivity induced by neuropathic pain. The immunostaining first revealed the expression of CB1R on somas and neurites of VIP-INs (Figure 6B). The protein quantification further showed that CCI treatment remarkably decreased CB1R levels in S1, whereas EA intervention elevated its expression level (Figures 6C and 6D). Those results converged to support a CB1R-induced neuromodulation on Ca2+ activities of VIP-INs and their downstream neurons.

Figure 6

Endocannabinoid receptor 1 pathway underlies pain relief effects of EA

(A) Clustered enrichment of functional pathways based on RNA sequencing and Kyoto Encyclopedia of Genes and Genomes (KEGG) database.

(B) Immunostaining images showing the distribution of cannabinoid receptor 1 (CB1R) on VIP-INs soma (upper panels) and puncta (lower panels) in S1. Scale bar, 100 μm in upper panels; 50 μm in lower panels.

(C) Representative gel images from western blotting using CB1R-specific antibody and total protein extracts from S1.

(D) Relative protein expression levels of CB1R. One-way ANOVA: F(2, 10) = 5.781, p = 0.0215; ∗p = 0.0421 between sham CCI and CCI group; ∗p = 0.0249 between CCI and CCI + EA group using Tukey post-hoc test; n = 4, 5, and 4 animals from sham CCI, CCI, and CCI + EA group, respectively.

(E) Schematic diagram of CB1R blockade assay for (F–I). C57BL/6 mice received CCI surgery, and EA intervention was applied in conjunction with intraperitoneal (i.p.) infusion of AM251 for 7 days.

(F) Mechanical thresholds during 14-day CCI surgery and intervention (two-way ANOVA, group factor: F(2, 240) = 94.92, p < 0.001; ∗∗∗p < 0.001 using Bonferroni post-hoc comparison; ns, no significant difference; n = 7, 10, and 6 animals for sham CCI, CCI + EA, and CCI + EA + AM251 group, respectively).

(G) Lowest temperature threshold on the heat plate (two-way ANOVA, group factor: F(2, 114) = 5.688, p = 0.0116; ∗p = 0.0215 between CCI and CCI + EA group; ∗p = 0.0261 between CCI + EA and CCI + EA + AM251 group using Bonferroni post-hoc comparison; n = 7, 7, and 8 animals for sham CCI, CCI + EA, and CCI + EA + AM251 group, respectively).

(H) Highest temperature threshold on the cold plate (two-way ANOVA, group factor: F(2, 114) = 41.95, p < 0.001; ∗∗∗p < 0.001 using Bonferroni post-hoc comparison; n = 7, 7, and 8 animals for sham CCI, CCI + EA, and CCI + EA + AM251 group, respectively).

(I) Heat and cold threshold for each mouse in a single plotting (1-way ANOVA: F(2, 19) = 160.4, p < 0.001; ∗∗∗p < 0.001 using Tukey post-hoc comparison; n = 7, 7, and 8 animals for sham CCI, CCI + EA, and CCI + EA + AM251 group, respectively). Data are presented as mean ± SEM.

As functional evidence, we further used one potent CB1R agonist, WIN55, to mimic the effect of EA in reducing mechanical hypersensitivity. In CCI model mice, a single intraperitoneal (i.p.) injection of WIN55 rapidly elevated the mechanical pain threshold within 20 min (Figures S7A and S7B). Furthermore, two-photon imaging found that WIN55 treatment suppressed both the peak value and the total integrated intensity of Ca2+ spikes (Figures S7C–S7I). Such phenotypes largely recapitulated the modulatory effect of EA on L5PNs activity (Figure 2), indicating the potential role of CB1R activation underlying EA's functions in reliving neuropathic pain.

To demonstrate the necessity of CB1R activation in the pain relief function of EA, we daily infused the CB1R antagonist, AM251, during 7-day persistent EA in CCI mice (Figure 6E). After systemic drug administration, EA + AM251-treated mice gradually displayed mechanical allodynia (Figure 6F), in association with both heat and cold hyperalgesia (Figures 6G–6I). These results demonstrated the abolishment of EA's analgesic functions after CB1R blockade, indicating the necessary role of CB1R. As neural circuitry evidence, in vivo Ca2+ imaging was performed on L5PNs after 7-day treatment of AM251 plus EA treatment (Figure 7A). Data analysis showed that whereas EA treatment remarkably decreased Ca2+ transients in CCI mice, AM251 administration re-elevated L5PNs activities (Figures 7B–7D). The treatment of AM251 remarkably increased Ca2+ peak values (Figures 7G and 7H) but not the frequency (Figures 7E and 7F), similar to those in CCI model mice (Figure 2). Last, we monitored if Ca2+ activities of SST-INs were modulated by AM251. A single i.p. injection of AM251 is sufficient to suppress the activity of SST-INs soma within 20 min on CCI mice receiving 7-day EA treatment (Figure S8), reversing EA-induced potentiation of Ca2+ peak values. Those pharmaceutical assays demonstrate that the activation of CB1R pathway is indispensable for EA-induced relief of cortical hyperactivity, and for its inhibition of mechanical/thermal hypersensitivity. In summary, our current study establishes a cortical mechanism in which EA treatment activates CB1R to suppress VIP-INs, leading to the disinhibition of downstream SST-INs and further inhibition of PNs, contributing to the relief of neuropathic pain.

Figure 7

CB1R pathway is necessary for EA-repressed cortical hyperactivity

(A) Schematic diagram showing experimental timelines for (B to H). C57BL/6 mice received unilateral transfection of AAV-Camkii-GCaMP6s, and were treated with CCI surgery, plus EA intervention and 7-day i.p. infusion of AM251. On D22, two-photon imaging was performed on L5PNs in S1.

(B) Heat maps showing normalized Ca2+ activities of L5PNs. A total of 70 cells were plotted from each group with a recording duration of 150 s.

(C and D) Total integrated Ca2+ activities of L5PNs from CCI, CCI + EA, and CCI + EA + AM251 groups. (C) A bar chart showing averaged total integrated Ca2+ activities (CCI: 388.6 ± 15.46, CCI + EA: 263.7 ± 14.26, CCI + EA + AM251: 349.0 ± 19.85; Kruskal-Wallis test statistic = 28.25, p < 0.001; ∗∗∗p < 0.001 between CCI and CCI + EA group; ∗∗p = 0.0056 between CCI + EA and CCI + EA + AM251 group using Dunn's post-hoc comparison; n = 89, 72, and 79 cells, N = 4, 5, and 4 animals for sham CCI, CCI + EA, and CCI + EA + AM251 group, respectively). (D) A histogram showing the frequency distribution of total integrated Ca2+ activities.

(E and F) Spike numbers per minute of Ca2+ transients on L5PNs. (E) A bar chart showing averaged Ca2+ spike frequency (CCI: 1.37 ± 0.06, CCI + EA: 1.27 ± 0.07, CCI + EA + AM251: 0.96 ± 0.06; Kruskal-Wallis test statistic = 19.84, p < 0.001; ∗∗p = 0.0063 between CCI + EA and CCI + EA + AM251 group; ns, no significant difference between CCI and CCI + EA group using Dunn's post-hoc comparison; n = 83, 64, and 70 cells, N = 4, 5, and 4 animals for sham CCI, CCI + EA, and CCI + EA + AM251 group, respectively). (F) The frequency distribution of Ca2+ spike frequency.

(G and H) Peak values of Ca2+ spikes. (G) Averaged Ca2+ peak values (CCI: 420.0 ± 17.81, CCI + EA: 267.1 ± 12.35, CCI + EA + AM251: 390.7 ± 20.76; Kruskal-Wallis test statistic = 33.82, p < 0.001; ∗∗∗p < 0.001 using Dunn's post-hoc comparison; n = 109, 80, and 92 cells, N = 4, 5, and 4 animals for sham CCI, CCI + EA, and CCI + EA + AM251 group, respectively). (H) Histogram showing the frequency distribution of Ca2+ peak values. Data are presented as mean ± SEM.

Discussion

To our knowledge, this study provides the first piece of in vivo evidence showing how the cortical circuit is modulated by repeated EA treatment to relieve neuropathic pain. Specifically, the activation of CB1R by EA leads to presynaptic inhibition at GABAergic VIP axonal terminals, resulting in the disinhibition of SST-INs and the suppression of L5PNs, and eventually reducing mechanical or thermal hypersensitivity. These results agree with previous findings reporting the dysregulation of cortical GABAergic transmission under neuropathic pain (Cichon et al., 2017) and identify CB1R as the mediator in restoring cortical E/I balance to ameliorate neuropathic pain after EA intervention.

The activity of L5PNs is closely correlated with perception threshold, and the manipulation of L5PNs can shift the sensory sensitivity (Takahashi et al., 2016). It is thus proposed that EA might modulate activities of L5PNs to improve pain hypersensitivity under neuropathic pain. Our results supported this model by showing the depression of in vivo Ca2+ activities of L5PNs after EA treatment, in association with the relief of pain behaviors. In particular, both apical spines and somas of L5PNs presented suppression of total Ca2+ activities, and apical spines had decreased frequency of Ca2+ spikes in EA treatment group (Figures 2 and 3), suggesting decreased pre-synaptic inputs under EA intervention. We thus examined the activity of inhibitory neural network because recent studies have reported the essential role of SST-INs in maintaining synaptic plasticity and neural functions (Chen et al., 2019; Adler et al., 2019). Here we found disinhibition of PNs as the result of suppressed inputs from SST-INs in CCI model as previously described (Cichon et al., 2017). More importantly, we identified the potentiation of SST-INs by EA treatment, which remarkably elevated the peak value of Ca2+ spikes in axonal regions (Figure S5), suggesting a higher probability of presynaptic GABAergic transmission for suppressing PN activities. Further analysis showed that VIP-INs are suppressed after EA treatment, providing a disinhibitory input for the projection from SST-IN to PN, and resulting in lower PN activities. These data collectively support the important role of E/I balance for maintaining cortical sensory function.

There are few studies regarding the molecular mechanism governing the cortical hyperactivity in neuropathic pain. The CB1R pathway has been widely accepted in pain processing (Woodhams et al., 2017). Across different brain regions including the spinal cord (Yang et al., 2016), midbrain nuclei (Walker et al., 1999), and prefrontal cortex (Huang et al., 2019), the activation of CB1R exerted analgesic effects. Our data suggest that CB1R activation is necessary for EA's analgesic effects. Although the pharmaceutical manipulation of CB1R cannot exclude its effect in midbrain regions as previously reported (Yuan et al., 2018; Shou et al., 2013; Chen et al., 2018), the Ca2+ recording data of cortical INs and PNs suggested that cortical CB1R might play a role in EA-induced analgesia.

One unanswered question is how EA treatment at two peripheral acupoints leads to CB1R potentiation. Two possibilities may exist: (1) EA facilitates the endocannabinoids release, leading to the up-regulation of CB1R, and (2) other peripheral or central factors are released upon EA intervention, indirectly resulting in CB1R activation. For the former model, previous study has found the increased release of anandamide in PAG after local electrical stimulation to relieve hyperalgesia (Walker et al., 1999). It is thus possible that sensory stimulation by EA results in cortical release of endocannabinoids. To support the latter model, one study reported an orexin-dependent release of 2-arachidonoylglycerol (2-AG) signaling to activate CB1R in the ventrolateral PAG after EA stimulation on the median nerve (Chen et al., 2018). In short, a cross talk may exist between different neuromodulators to affect synaptic plasticity across the whole pain regulatory network after EA stimulation.

Another major challenge for revealing the neural mechanism of EA exists in the identification of its primary functioning sites or brain regions. Different models have been raised involving the local acupoints, spinal cord, midbrain nuclei, and cortex. Previous studies suggested that the local activation of adenosine receptor mediates anti-nociceptive effects of acupuncture (Goldman et al., 2010). In the dorsal horn neurons of the spinal cord, the evoked response was reduced by EA treatment (Kim et al., 2011). Within the descending pathway, different midbrain nuclei such as nucleus raphe magnus, PAG, habenular nucleus, and arcuate nucleus have been implicated in EA-related analgesia (Zhao, 2008; Lee and Beitz, 1993; Chen et al., 2014). Across the neocortex, prefrontal region and anterior cingulate cortex have been correlated with EA (Ong et al., 2019; Chen et al., 2014). The current study dissected the cortical inhibitory network regulating PNs' output under EA intervention, to further emphasize the critical role of sensory cortex in EA's analgesia. As the higher sensory processing center, S1 neurons project to various cortical and subcortical nuclei. Recently, a CST pathway originating from S1 has been demonstrated to directly affect the plasticity of spinal dorsal horn INs, thus affecting the pain modulation (Liu et al., 2018). It is thus likely that the suppression of L5PNs by EA treatment modulates such CST-mediated descending pathway, affecting the pain gating in the spinal cord.

In sum, we have established a cortical neural circuit that helps to explain the effect of EA in reducing mechanical or thermal hypersensitivity under neuropathic pain. Such modulatory processes are dependent on CB1R activation, which leads to potentiated inhibitory inputs onto PNs of S1. Our data add further knowledge to the growing body of nociceptive circuits and provide more evidence for adopting acupuncture in clinical management of neuropathic pain.

Limitations of the study

The current study only described the modulation of local neural circuitry in the somatosensory cortex under EA intervention, leaving the changes in subcortical network largely unclear. Moreover, whether CB1R mediates the pain-relieving effect via other brain regions or peripheral nerves is inconclusive based on current data.

Resource availability

Lead contact

Further information and requests for resources and reagents in this study can be fulfilled by contacting the Lead Contact, Li Zhang (zhangli@jnu.edu.cn).

Materials availability

This study did not generate new unique reagents.

Data and code availability

RNA sequencing data are available in the ArrayExpress database (http://www.ebi.ac.uk/arrayexpress) under accession number E-MTAB-9987.

Methods

All methods can be found in the accompanying Transparent methods supplemental file.