Schwann cell derived IL-1α and TNF-α serve as very early mediators in the response of axons to injury. They recruit macrophages and initiate molecular and cellular events in Wallarian degeneration such as the production of additional cytokines and NGF (81).

Generation of IL-1β is brought about by activation of the Nod-like receptor family pyrin domain containing 3 (NLRP3) inflammasone (33, 154–156). NLRP3 is activated following the release of DAMPs and their interaction with pattern recognition receptors (PRRs) (157), such as Toll-like receptors (TLRs) (158, 159) and with purinergic P2X7 receptors (159). IL-1β is released as a pro-protein and processed into its mature bioactive form by caspase-1 (160) or by metalloproteases 2 and 9 (MMP2 and MMP9) (161). Release of IL-1β from macrophages, dendritic cells and neutrophils, may be brought about via the formation of gasdermin D pores in the cell membrane (160, 162, 163). Alternatively IL-1β release may involve its exocytosis via panexin channels (164).

Metalloproteases also cleave the membrane bound form of TNF-α into the mature 17-kDa form (165). This and their ability to also cleave IL-1β and to produce pain when administered intrathecally has led to the suggestion that MMP antagonists may be useful in pain management (161). There are however no reports of use of metalloprotease inhibitors in pain management in the clinic. This may be due, in part, to the observation that activated MMP2 and MMP9 cleave the mature form of NGF into biologically inactive products (166). The effect of MMP blockade here would be to preserve the presence of pro-inflammatory NGF. This possible proinflammatory action of MMP blockers would tend to restrain any anti-inflammatory/analgesic action.

Peripheral nerve injury, via the actions of primary mediators, leads to ectopic spontaneous activity in primary afferents that is crucial for the onset and persistence of neuropathic pain in humans and signs of such pain in rodent models (19, 30, 35, 197–206). Thus, suppression of aberrant peripheral nerve activity in animal models in vivo by either optogenetic or pharmacological methodologies (205, 207) leads to attenuation of hyperalgesia and abatement of injury-induced allodynia.

In general, peripheral nerve injury, by the action of primary mediators, decreases K⁺ channel function and increases that of voltage-gated Na⁺ and Ca²⁺ channels, TRP channels and HCN channels in DRG neurons (208–213). Injury-induced changes in ion channels can also provoke bursting activity in sensory neurons (214) that may relate to release of ATP and its interaction with P2X3 receptors at the site of injury and the initiation synchronous oscillations in primary afferents (206). In addition, altered excitability may be a consequence of mitochondrial dysfunction and chronic energy deficit (215).

Peripheral activity after injury may affect the whole somatosensory system. It may provoke enduring low frequency cortical oscillations and synaptic remodeling in S1 somatosensory cortex as well as for inducing animals' pain-like behaviors (206). This is supported by the observation that enhancing the synchrony of DRG neuronal activity causes synaptic changes in S1 and pain-like behaviors similar to those seen after spared nerve injury (SNI).

An overview of the actions of IL-1β, TNF-α, Wnt ligands, chemokines and other primary mediators on peripheral neurons is presented in the succeeding sections.

Acute application of IL-1β increases the excitability of DRG neurons by relieving slow inactivation of tetrodotoxin (TTX)-resistant voltage-gated sodium channels (135). IL-1β levels peak at 1 d after injury and remain elevated for ∼7 d (139) and investigations of its longer term actions following 5 d–6 d exposure reveal different effects on different neuronal subpopulations (136). These are observed at remarkably low concentrations (216).

The long term effects of IL-1β on small IB₄ -positive neurons (most of which are non-peptidergic, low threshold mechanoceptors) include a reversible increase in action potential (AP) amplitude as a result of increased tetrodotoxin (TTX)-sensitive Na⁺ current and an irreversible increase in AP duration as result of decreased Ca²⁺- sensitive K⁺ conductance (138).

The effects of IL-1β on medium sized neurons, which are the cell bodies of Aδ fibres, are dominated by decreases in K⁺ currents (137). Although the precise ionic mechanisms differ, IL-1β increases the excitability of both small-diameter IB₄-positive neurons and medium-diameter neurons. By contrast, large neurons which are the cell bodies of fast conducting Aβ fibres and IB₄-negative neurons, which are predominantly peptidergic nociceptors, are little affected (136).

Macrophage and Schwann cell derived TNF-α is upregulated at the site of injury following CCI (85) and its peripheral application promotes ectopic activity in nociceptors in vivo (217). This effect is enhanced after SNL injury (173). Microinjection of TNF-α lowers mechanical pain threshold in nerve-injured animals in a similar fashion to IL-1β. Most actions of TNF-α in DRG involve modifications of Na⁺ channel function (218) rather than effects on K⁺ channels (210). For example, TNF-α upregulates Na_v1.7 (219) as well as slow persistent TTX-resistant Na⁺ channel currents (149).

Intraplantar injection of Wnt3a promotes mechanical hypersensitivity and thermal hyperalgesia in uninjured animals. It also upregulates the ionotropic ATP receptor P2X3 as well as TRPA1 receptor channels. P2X3 receptors may be activated by the passive release of ATP from damaged cells leading to increased sensory neuron excitability (153). Wnt3a also stimulates production of TNF-α and IL-18, thereby augmenting the overall inflammatory response

Several chemokines excite DRG neurons (97, 220).

CCL-2 signals through CCR-2 to increase nociceptor excitability (97, 128, 221, 222). Its effectiveness is increased after DRG compression (CCDRG) (223). CCL-2 is expressed by DRG neurons where it is packaged into large dense-core vesicles. Release of vesicles can be induced by depolarization in a Ca²⁺-dependent manner (224). This autocrine function could thereby amplify injury-induced excitatory processes evoked in DRG.

CXCL-12 signalling through its cognate receptor, CXCR-4 increases excitability of Na_v1.8-positive DRG neurons and this plays a role in the generation of mechanical allodynia as well as small-fiber degeneration in a mouse model of peripheral diabetic neuropathy (101). CXCL-12 and CXCR-4 are upregulated after CCDRG. In addition, intrathecal injection of a CXCL-12 antagonist or a CXCL-12 neutralizing antibody reverse allodynia after SNI or CCDRG (103, 104, 130). These findings suggest that peripheral CXCL-12/CXCR-4 signaling contributes to pain after damage to the DRG per se (104).

In addition to secreted proteins, chemokines, cytokines and growth factors, several small molecules produced at the site of injury act as primary mediators. These include prostaglandin E2, bradykinin, serotonin (146) and histamine (92); all of which increase the excitability of DRG neurons (131, 147). Actions of both serotonin and PGE2 involve augmentation of TTX-resistant I_Na in nociceptors (225).

As already mentioned, manipulation of the actions of cytokines, chemokines or other primary mediators has so far failed to bring forth any promising therapeutic approaches. On the other hand, the crucial role of primary afferent hyperexcitability and spontaneous activity in pain etiology (30, 199, 202, 205) draws attention to the potential use of ion channels as therapeutic targets (59, 208–210, 226).

DRG neurons express a variety of K⁺ channel subtypes including delayed rectifiers (K_v1.1, 1.2), A-channels (K_v1.4, 3.3, 3.4, 4.1, 4.2, and 4.3), KCNQ or M-channels (K_v7.2, 7.3, 7.4, and 7.5), ATP-sensitive K⁺ channels (K_IR6.2), Ca²⁺-activated K⁺ channels (K_Ca1.1, 2.1, 2.2, 2.3, and 3.1), Na⁺-activated K⁺ channels (K_Ca4.1and 4.2) and two pore domain leak channels (K2p; TWIK related channels). These channel subtypes are preferentially and differentially expressed in various neuronal subpopulations and attempts to restore K⁺ channel function have involved the use of channel activators (210). Although K_v7 activators are quite effective in rodent models (227, 228),, the anticonvulsant, retigabine failed to reach its efficacy endpoint in a trial for post herpetic neuralgia (17).. Nevertheless, as will be outlined below, better phenotypical stratification of patents into clusters on the basis of quantitative measurements of their pathophysiology may reveal clinical efficacy of drugs that failed to demonstrate effectiveness in large groups of patients (8). In the case of K⁺ channel activators, over 200 new molecules are currently under investigation (227).

Mechanisms that control K⁺ channel expression and function may present additional therapeutic targets. For example, the expression of K_v7.2, K_v1.4 and K_Ca1.1 is controlled by the histone methyltranferase G9a (229). Pharmacological inhibition of G9a attenuates neuropathic pain in rodent models (230, 231). Although there is considerable interest in developing histone methyltransferase inhibitors in cancer treatment (232), none have been examined for treatment of neuropathic pain.

A variety of Na⁺ channel blockers show promise as therapeutic agents; inhibition of Na_v1.7, 1.8 or 1.9 seems particularly effective (208, 233, 234). Because it is not found to any great extent in non-neuronal vital tissue such as heart or skeletal muscle, Na_v1.7 represents an especially attractive target for therapeutic manipulation (59, 208). Indeed, some level of success has been realized in phase II clinical trials for trigeminal and diabetic neuralgia with the Na_v1.7 blocker, vixotrigine (235, 236) but phase III trials remain at the planning stage (235).

Expression of Na_v1.8 in DRG neurons is controlled by NGF (237) and the NGF binding antibody tanezumab is effective in various human pain states (238). Small molecule, peripherally-acting TrkA inhibitors have also been identified (239–241).

DRG neurons express high voltage-activated (HVA) Ca²⁺ channels; Ca_v2.2 (N-type) as Ca_v2.1 (P/Q-type) and Ca_v1.2 (L-type) (242). Low voltage-activated (LVA) channels (T-type) are also present, notably Ca_v3.2 and 3.3 (243–245).

Because Ca_v2.1 (P/Q type) and Ca_v2.2 (N-type) Ca²⁺ channels contain a synaptic protein interaction site (246) they are closely associated with the synaptic vesicles that govern neurotransmitter release. In view of this, the role of Ca²⁺ channels in controlling neuronal excitability and reports of upregulation of both HVA and LVA Ca²⁺ channels by injury (247–249), Ca²⁺ channels emerge as an important therapeutic target for pain management (208, 209, 226, 250–252). This potential has been realized by the use of the of N-type Ca²⁺ channel blocker ziconotide as a last resort for pain that is refractory to all other treatments (253). The drawback is that ziconotide needs to be delivered directly to the spinal cord via the intrathecal route (254). In view of this, there is strong interest in developing orally effective N-channel blockers (209, 226, 251, 253) and although several promising agents have appeared in the last five years, the ubiquitous distribution of N-type channels throughout the nervous system means that side effects of such agents may present a serious barrier to drug development.

The function of N-type Ca²⁺ channels is modulated by G_i/o coupled agonists (255, 256) but the clinical efficacy of the α2-adrenoceptor agonist, clonidine is limited to subsets of patients within the postherpetic neuralgia, complex regional pain syndrome or diabetic neuropathy cohorts (257). Nevertheless, this documented efficacy of clonidine has led to an extensive in silico modelling study. Compounds with nanomolar affinities for the α2a-adrenoceptors and limited ability to recruit arrestin β have been identified and tested in animal models where they behave as non-sedating, orally effective agents that attenuate signs of neuropathic, inflammatory and acute pain (258). The potent α2-adrenoceptor agonist, xylazine has been available for over 30 years, but its use has been restricted to pain management in veterinary medicine as it promotes severe hypotension and dangerous bradycardia in humans (259). It also has documented abuse potential (260).

In addition, the therapeutically important gabapentinoids (16, 261) modulate HVA Ca²⁺ channel function by binding to their α2δ–1 regulatory subunits (262). Gabapentinoids may antagonise the actions of the endogenous ligand thrombospondin (263). This means that perturbation of thrombospondin expression and/or function may present a novel therapeutic route to pain management. The α2δ–1 subunit plays a major role in Ca²⁺ channel trafficking, expression and function (22, 248, 264, 265) and deletion of the α2δ–1 gene delays development of mechanical hypersensitivity that follows peripheral nerve damage (262). α2δ–1 is also implicated in controlling the expression of Ca²⁺ permeable AMPA channels (266) and NMDA receptor channels (267). It is likely therefore that the therapeutic benefits of gabapentinoids involve interactions with several channel types.

LVA T-channels control nociceptor excitability (226, 268–270) and are involved in transmitter release from primary afferent terminals (271, 272). In some patients, gain of function mutations of Ca_v3.1 contribute to trigeminal neuralgia (273). Peripheral nerve injury (CCI or diabetic neuropathy model) increases function of Ca_v3.2, in rodent DRG neurons (249, 269) and specific knockdown of Ca_v3.2 induces marked analgesia in vivo (270).

Although several small molecule Ca_v3.2 blockers have shown promise in preclinical studies (274, 275) most have failed to exert a significant effects in cohorts of pain patients (208, 276). On the other hand, the high-affinity T-type channel blocker Z944 is especially effective in murine pain models and this may reflect selective blockade of Ca_v3.1, Ca_v3.2, and Ca_v3.3 (277) in peripheral, spinal and thalamic neurons (278, 279). Preliminary results of phase 1 and phase 2 trials with Z944 also appear promising (280).

Some neuropathic pain patients respond favorably to cannabinoids (281) and this may be ascribed to inhibition of Ca_v3.1 and/or Ca_v3.2 Ca²⁺ channels (282, 283) as well as inhibition of N-type Ca²⁺ channels (284), augmentation of BK type K⁺ channel currents (285) and stabilization of an inactivated state of Na_v1.8 channels (286). There has been considerable interest in NMP-7 and Compound 9 which affect Ca_v3.2 channels by interactions CB1 and/or CB2 receptors. Although these compounds seem highly effective in animal models, they do not appear to have been tested in the clinic (287–289).

Rather than direct channel block or inhibition by the action of G_i/o coupled agonists, there is considerable interest in modulating Ca_v3 channel activity by targeting the molecular mechanisms that regulate them.

For example, upregulation of the deubiquitinase, USP5 by IL-1β impairs Ca_v3.2 ubiquitination thereby protecting it from proteasomal degradation and prolonging its surface expression (140, 272, 290, 291). USP5 knockdown thus increases Ca_v3.2 ubiquitination, reduces its surface expression leading to reduction of Ca_v3.2 whole-cell currents. This in turn, leads to attenuation of mechanical hypersensitivity in murine models of both inflammatory and neuropathic pain. As shown in Figure 2A, Ca_v3.2/USP5 interactions are interrupted by a novel bioactive rhodanine compound (292), by the antiparasitic agent, suramin, and by a TAT-cUBP1-USP5 peptide. Each of these substances attenuate surface expression of Ca_v3.2 and show analgesic activity in neuropathic and inflammatory pain models (292–294). These observations may lead to the development of new therapeutic approaches (292).

Nerve injury or long-term exposure to IL-1β increases HCN channel function in DRG (137, 295). This increase drives spontaneous activity (296, 297) and increases the release of neurotransmitter from primary afferents terminals (298, 299). HCN channel blockers thus supress signs of neuropathic pain in rodent models (300, 301) and selective deletion of HCN2 in nociceptive neurons prevents the development of neuropathic and inflammatory pain (296). Because the HCN2 channel subtype is mainly expressed in neurons as opposed to other excitable tissues (302), HCN2 blockers abrogate DRG hyperexcitability without affecting the HCN1 channels that control cardiac rhythmicity (303). In the clinic, the non-selective HCN blocker, ivabradine which is approved for treatment of heart failure, has a beneficial effect in painful diabetic neuropathy but only a weak effect in other forms of neuropathic pain (304).

TRPV1 receptor channels in nociceptors are upregulated following SNL injury (305) and sensitized by the action of inflammatory mediators (211). Unfortunately, the clinical effectiveness of TRPV1 blockers is limited by the presence of undesirable side effects (306). By contrast, transdermal patches containing a high concentration of the TRPV1 agonist, capsaicin have a role pain management (16). They are applied for 60 min in combination with regional anesthesia. This high level of capsaicin destroys the terminals of TRPV1 expressing nociceptors. This may include those that have sprouted into areas previously occupied by low threshold mechanoceptors (29).

In animal models, combining local anesthetics with capsaicin is especially effective in attenuating signs of pain. Molecules such as lidocaine pass freely through the pore of activated TRPV1 channels and thereby gain access to their intracellular binding site on the Na⁺ channel. The local anesthetic thus directly and selectively targets TRPV1 expressing nociceptors (307).

In addition to generation and release of inflammatory primary mediators, peripheral nerve injury alters expression of neuropeptides and their cognate receptors in primary afferent neurons (308–311). Neuronal activity promotes the release of neuropeptides such as CGRP and substance P from peripheral nerve endings, DRG cell bodies (312, 313) and primary afferent terminals (314, 315). They modulate sensory neuron activity by excitatory actions in DRG (148, 316, 317) and by their participation in axon reflexes at peripheral nerve endings (318, 319). Although increased effect of CGRP and substance P thus likely contributes to increased excitability and spontaneous activity of peripheral nerve, substance P antagonists are not effective in pain management in the clinical setting (320).

Erenumab, a monoclonal antibody raised against CGRP is available for the management of migraine (321) and recent evidence support the use of CGRP antagonists in the management of trigeminal neuralgia (321, 322). CGRP antagonism, both in the clinic and in animal models is less effective in males than in females (323, 324).

Neuronal activity produces enduring changes in immune and glial cell function (18, 27, 68, 125, 325–327). This process has been termed neurogenic neuroinflammation (328, 329).

Injury-induced upregulation and release of neuropeptides is one of several mechanisms that effects transmission from neurons to glia and immune cells. For example, CGRP, substance P and vasoactive intestinal peptide (VIP) act on their cognate receptors on immune cells and vasculature to promote inflammation (318). CGRP regulates spinal microglial activation in a rodent model of neuropathic pain (330) and substance P regulates expression of IL-1β in keratinocytes (331).

Neuron-immune cell interactions can also be brought about by the synthesis and release of cytokines (110) and chemokines (224) from neurons per se.

By contrast with neurogenic neuroinflammation which is a consequence of injury, essentially the reverse effect; suppression of immune system activity by neuronal activity, characterises a well-defined immune reflex. This contributes to the resolution of inflammation following injury (327). The best characterized part of this reflex involves the vagal release of acetylcholine which acts on the nicotinic acetylcholine receptor subunit α7 (α7nAChR) on innate immune cells to supress cytokine generation and release (327). Activation of β2 adrenoceptors is also immunosuppressant and this is thought to involve downregulation of the TNF-α signaling pathway within the DRG. This may contribute to the efficacy of serotonin - noradrenaline re uptake inhibitors (SNRI's) in neuropathic pain (332). This is because invading sympathetic fibres following nerve injury (44) provide a source of noradrenaline to active immunosuppressant β2 adrenoceptors and noradrenaline abundance is increased by the action of the SNRI, duloxetine. It has also been reported that activation of β2 adrenoceptors on microglia attenuates signs of neuropathic pain in a mouse model (333).

Neuropathic pain generated by peripheral nerve injury may involve sprouting of nociceptors into denervated territories such as skeletal muscle and skin. Here they replace the initial map and configuration of low threshold sensory axons that do not regenerate. Genetic ablation of nociceptors fully abrogates this type of re-innervation allodynia. These results reveal the emergence of a component of neuropathic pain that is driven by structural plasticity of peripheral sensory nerves, abnormal terminal connectivity and malfunction of nociceptors during reinnervation (29).

Peripheral nerve injury provokes sprouting of perivascular sympathetic axons and appearance of ectopic excitatory α-adrenoceptors on the cell bodies of primary afferent neurons and on their terminals at the site of injury (44, 255). This sprouting may be driven by the neurotrophic action of LIF or NGF (145, 336, 337) and/or may be a consequence of spontaneous afferent activity (338). This is yet another means by which nerve injury increases primary afferent excitability (44, 255, 339–341), leading to signs of neuropathic pain in animal models (342) and to the development of complex regional pain syndromes in humans (343).

The best characterized secondary mediators include the cytokine CSF-1, the chemokines CCL-21 and CXCL-12 as well as Wnt5 and CGRP. Secondary mediators affect the properties of spinal microglia and astrocytes which in turn release tertiary mediators (68) (Table 1 and Figure 1). As will be described below, glial-derived tertiary mediators such as IL-1β and BDNF (117) act on neurons to bring about misprocessing of sensory information and increased activity and excitability leading to central sensitisation (359) (Table 1 and Figure 1).

Although microglia play a predominant role in central sensitization in males, invading macrophages and T-lymphocytes are predominant in females (360–362). Spinal signalling mechanisms invoked in males are therefore very different from those invoked in females (10, 352, 361). Lines of investigation initiated over 20 years ago have been directed towards understanding the numerous cellular and molecular processes that underlie this difference (68, 360, 362–369) and relevant and important differences will be outlined in the succeeding sections.

Unilateral CCI (chronic constriction injury) increases IL-6 mRNA and protein bilaterally in both neurons and satellite glial cells of the DRG (83). IL-6 promotes hyperalgesic priming in rodents (323) and conditional knockout of its cognate gp130 receptor in nociceptors abrogates pain in inflammatory and tumor-induced pain models (393). Although these results are consistent with a secondary mediator role for IL-6, other work suggests that it may have an anti-nociceptive action both in the periphery and at the spinal level following SNI (spared nerve injury) in rodents (349, 350) and may be capable of inducing a desensitized microglial phenotype (394).

IFN-γ alters spinal microglial function and induces tactile allodynia. Genetic ablation of the interferon receptor (IFN-γR) impairs nerve injury-evoked allodynia and prevents phenotypic modulation of spinal microglia (395). The P2X4 receptor is upregulated in IFN-γ stimulated - microglia and, as mentioned already, these purinergic receptors play a crucial role in the onset of neuropathic pain in males (118, 120, 375, 377, 378). IFN-γ also increases dorsal horn excitability (396, 397) and facilitates synaptic transmission between C-fibres and Lamina 1 neurons via a microglial-dependent mechanism (132). Although IFN-γ is found in DRG neurons (398) and the level of IFN-γ is increased in spinal cord following peripheral nerve injury (399) this may originate from invading T-lymphocytes. However, given the role of T-lymphocytes in females (361, 362), IFN-γ may be important in pain aetiology in women.

The lysosomal cysteine protease, cathepsin S is released from microglia by a P2X7-dependent mechanism (400). Cathepsin S, as well as the metalloproteinase ADAM₁₀ and TNF-α converting enzyme liberate the soluble form of the chemokine, fractalkine (FKN; CX3CL-1) from dorsal horn neurons (125, 126, 401) (Figure 2B).

The transmembrane form of FKN and its cognate receptor (CX3CR-1) are expressed constitutively in spinal cord neurons (402, 403). CX3CR-1 which is strongly expressed in dorsal horn microglia (125, 403, 404), is upregulated after nerve injury. In naïve animals, intrathecal injection of FKN produces mechanical allodynia and thermal hyperalgesia whereas injection of an antibody raised against CX3CR-1 attenuates signs of neuropathic pain in animal models (404). This is consistent with the observation that peripheral nerve injury fails to provoke allodynia in mice lacking CX3CR-1 (405).

Spinal nerve ligation (SNL) also increases the level of the soluble form of FKN in cerebrospinal fluid (401) and such release appears to be obligatory for the expression of neuropathic pain (127, 383, 406). Soluble FKN modulates microglial phenotype leading to the generation of tertiary mediators such as TNF and IL-1β (404, 407).

Antibodies raised against CX3CR-1 reduce nociceptive responses when administered as long as 5–7 days after CCI suggesting that the prolonged release of FKN contributes to the maintenance as opposed to the onset of neuropathic pain. This may also relate to the observation that SNL provokes de novo expression of FKN in dorsal horn astrocytes (403).

In addition to producing synaptic potentials in almost all CNS neurons, glutamate affects astrocytes, T-cells, endothelial cells, microglia and vascular cells by interaction with mGluRs (329, 408, 409). These actions are predominantly anti-inflammatory (410). For example, mGluR5 activation in spinal microglia inhibits the release of inflammatory mediators both in vitro (410) and in vivo (411). Also, activation of group I mGluRs in astrocytes leads to increased glutamate and potassium uptake (412). These actions may thus be associated with offset of neuroinflammation rather than its onset.

The secondary mediator CSF-1 interacts with CSF-1R on spinal microglia (353). This leads to increased expression of the tertiary mediator BDNF as a result of up regulation of the Bndf gene (413). As illustrated in Figure 3, the release of BDNF plays an indispensable role in the onset and maintenance of neuropathic pain in male but not in female rodents (22, 68, 113, 117, 119, 121–123, 414–420). BDNF acts primarily via TrkB to increase dorsal horn excitability (113).

Exposure of dorsal horn neurons to CSF-1 also increases the frequency and amplitude of sEPSC's (spontaneous excitatory postsynaptic currents) and this effect is abrogated by the BDNF binding protein TrkB-fc (113). These findings underline the importance of the sensory neuron- CSF-1 -microglia - BDNF signalling process in the aetiology of neuropathic pain (14, 22, 68, 110, 183, 421) (Figure 3A).

ATP activation of microglial P2X4 receptors leading to the release of BDNF is involved in the aetiology of neuropathic pain in males, but not in females. This is congruent with the absence of functional P2X receptors on microglia of female rodents (364). There is also evidence for a role of microglial metabotropic P2Y6, 11, 12, 13 and 14 receptors in the onset of neuropathic pain (68, 422–427). Primary afferent neurons are not the primary source of ATP following peripheral nerve injury. It may rather derive from neurons in the superficial dorsal horn itself (428) as well as from microglia themselves (429). BDNF release from neurons is vesicular and dependent on extracellular Ca²⁺ (118, 119, 375, 377, 378).

The action of ATP on microglia is not the sole mechanism for promoting BDNF release. Wnt proteins that are upregulated in the spinal cord in various pain models (50, 124, 429–431) activate “frizzled” receptors (432) on microglia to increase expression of BDNF and promote its release (420, 433). This phenomenon has been examined in models of HIV pain which involve exposure of sensory neurons to toxic viral coat proteins such as Vpr1 (49) or gp120 (433, 434). The latter promotes allodynia and increases glutamatergic neuronal activity leading to NMDA receptor activation and increases the level of intraneuronal Ca²⁺. This, in turn promotes Wnt protein synthesis and release (435, 436).

Phenotypic modulation of microglial function in rodent dorsal horn persists for more than 3 months after injury (437). Thus sequestration of BDNF with TrkBFc (438) or selective depletion of spinal microglia with the targeted immunotoxin Mac1-saporin almost completely reverses mechanical and thermal allodynia up to 3 months after injury. By contrast, intrathecal injection of a cocktail of antibodies against IL-1β, TNF-α, and IL-6 significantly attenuates tactile and cold allodynia at 2 weeks but not at 3 months after injury. These findings suggest that different mediators should be targeted in the short vs. long term management of neuropathic pain (437).

The tertiary mediator, IL-1β is produced and released from macrophages, astrocytes and microglia (18, 439, 440). Release of IL-1β from microglia is a consequence of activation of P2X7 receptors (164, 380, 441, 442) and may be provoked by the action of FKN (407). In agreement with this, it has been reported that the Ca_v1 channel blocker, cilnidipine which also blocks microglial P2X7 receptors, impairs IL-1β release and reverses SNL-induced mechanical hypersensitivity (142). It has also been suggested that P2X4 receptors interact intracellularly with P2X7 receptors to augment P2X7 receptor-mediated IL-1β release (442).

In addition to the extracellular actions of BDNF and IL-1β, cell-to-cell transport of material via exosomes or extracellular vesicles is now believed to contribute to the development of central sensitisation (183, 443–449). Extracellular vesicles are released from both microglia (450) and astrocytes (451) and are taken up by neurons (447). They may serve as a conduit for the transfer of microRNA's between cell types (452). For example, Na_v1.7 protein may be transported from primary afferents to the dendrites of lamina II neurons; a process which may be effected by transfer of exosomes (449).

Exposure of rat substantia gelatinosa neurons to BDNF increases excitatory synaptic drive to excitatory neurons and decreases excitatory drive to inhibitory neurons (122, 414). In mice, effects of BDNF are dominated by increased excitatory drive to excitatory neurons (113). Whilst resting potential, rheobase, input resistance and excitability are little affected in rat neurons (113, 122, 414), the altered synaptic activity increases spontaneous AP discharge in excitatory neurons whilst reducing it in inhibitory neurons (414).

Several observations show that these actions of BDNF are relevant to injury-CSF-1-microglia-BDNF evoked central sensitization (Figure 3A). Firstly BDNF-induced changes in synaptic transmission and its lack of effect on the intrinsic excitability the cell bodies of lamina II neurons very much parallel those invoked by peripheral nerve injury (122, 454–456). Secondly, Ca²⁺ responses evoked by neuronal depolarization are enhanced by BDNF and by conditioned medium from lipopolysaccharide-activated microglia. The effect of this conditioned medium is attenuated by sequestering BDNF with TrkBd5 (122). Thirdly, the secondary mediator CSF-1 increases synaptic excitation of excitatory lamina II neurons in mice and this effect is abrogated by sequestering BDNF with TrkBfc (113).

It should be noted however that mitochondrial dysfunction following peripheral nerve injury and the resultant high levels of superoxide may also contribute to increased excitatory synaptic strength in dorsal horn neurons and neuropathic mechanical hypersensitivity (457).

CCI of the sciatic nerve reduces expression of the potassium-chloride exporter (KCC2) in lamina 1 neurons of the dorsal horn (458, 459). The resulting intracellular accumulation of Cl⁻ reverses the Cl⁻ concentration gradient such that normally outward, inhibitory GABAergic synaptic currents mediated by Cl⁻ influx become inward excitatory currents mediated by Cl⁻ efflux (458–460). Knockdown of KCC2 in uninjured rats reduces pain thresholds and induces neuropathic pain-like behaviors. By contrast, rescue of KCC2 expression abrogates signs of neuropathic pain in nerve injured animals (461, 462). Taken together and as illustrated in Figure 4, these findings strongly implicate perturbation of the Cl⁻ gradient and the phenomenon of disinhibition in the pathophysiology of central sensitization (121, 458).

BDNF is responsible for downregulation of KCC2 protein levels in male rats (121, 419). Thus, administration of ATP-activated microglia reproduces the shift in anion gradient seen after nerve injury in the same way as BDNF. Also, blocking TrkB or using interfering RNA against BDNF reverses both injury-induced pain behaviors and the shift in Cl⁻ gradient (121). Changes in KCC2 expression in deep dorsal horn neurons are confined to nociceptive neurons that project via the spinothalamic tract whereas wide dynamic range (WDR) neurons that are activated by a variety of sensory modalities are unaffected (461). It has also been shown that neurons in lamina I are more susceptible to changes in Cl⁻ gradient than those in lamina II (459) and biophysical and modelling analysis shows this loss is especially effective in promoting increased neuronal firing (463). These are important observations as lamina I and deep dorsal horn nociceptive neurons are the primary site for relay of nociceptive information to the brain (464–466). Loss of GABAergic inhibition enables non-noxious Aβ fiber-mediated excitatory transmission to access and excite the pain transmitting neurons of the superficial spinal dorsal horn. Thus, as already mentioned, tactile activation of Aβ fibres is perceived as pain and this process plays a role in the establishment of mechanical allodynia (467–469).

Descending serotonergic inhibition of nociceptive processing from the nucleus raphe magnus becomes excitatory and proalgesic in rats subject to spared nerve injury (SNI). This change is also dependent on collapse of the Cl⁻ gradient following KCC2 hypofunction in the dorsal horn as the KCC2 enhancer CLP290 restores both 5-HT–mediated descending inhibition and analgesia (470).

KCC2 downregulation also contributes to pain hypersensitivity in females (363). Whereas this is mediated by release of BDNF from microglia in males, it involves activation and invasion of adaptive immune cells such as T-lymphocytes in females (362, 381) (Figure 3) as well as downregulation of KCC2 expression by CGRP (323).

BDNF enhances excitatory responses to NMDA in rat spinal cord in vitro (471). In male rodents, this potentiation is dependent on BDNF-mediated GABA disinhibition. By processes yet to be discovered, KCC2-dependent disinhibition promotes downregulation of the tyrosine phosphatase STEP₆₁. Loss of function of STEP₆₁ phosphatase then clears the way for phosphorylation of GluN2B subunits by the Src family kinase Fyn (472). As illustrated in Figure 4, synaptic NMDAR responses are therefore enhanced and neuronal excitability is increased. Decreased activity of STEP₆₁ is both necessary and sufficient to affect GluN2B function (473), This sequence of events is supported by the observation that blocking of KCC2-mediated disinhibition with acetazolamide (474) reverses the downregulation of STEP₆₁ and attenuates behavioural hypersensitivity generated by chronic inflammation.

In female rats however BDNF fails to downregulate KCC2 and STEP₆₁ and to upregulate pFyn, GluN2B and its phosphorylated form GluN2B. This means that BDNF fails to affect synaptic NMDAR responses in lamina I neurons of females. Ovariectomy recapitulates the male pathological pain neuronal phenotype in female rats, with BDNF driving coupling between disinhibition and NMDAR potentiation in lamina I neurons following the elimination of sex hormones (475).

This sex difference in spinal pain processing in rodents is conserved in humans. Thus ex vivo spinal treatment with BDNF downregulates KCC2 and STEP₆₁ and upregulates markers of facilitated excitation in superficial dorsal horn neurons from male but not female human organ donors (475).

In addition to the postsynaptic effects described above (121, 473, 475), BDNF activation of TrkB increases the function of presynaptic NMDA receptors on primary afferent terminals (476). This leads to the potentiation of glutamate release from primary afferents that is observed after SNL (477) and may account for the increased frequency of sEPSC's seen in some dorsal horn neurons in the presence of BDNF(414). Functional upregulation of GluN2B subunits of NMDA receptors (478) may also account for the observation that long term potentiation (LTP) of synaptic transmission of C-fibre responses is enhanced by BDNF (479).

The level of IL-1β is elevated in the cerebrospinal fluid (CSF) of patients with complex regional pain syndrome (480) and in spinal cords obtained post-mortem from patients with painful HIV related neuropathy (50). As already mentioned, activation of P2X7 receptors promotes release of IL-1β from microglia (142, 164, 380, 441) and this is amplified by the action of FKN (407).

Microglial derived IL-1β stimulates astrocytic production of TNF-α well as IL-1β itself (440, 481) thereby amplifying the overall IL-1β signal. IL-1β promotes internalization of the astrocytic glutamate transporter (EAAT2) thereby reducing the capacity of astrocytes to take up glutamate (482, 483). Loss of EAAT2 function thus augments excitatory synaptic transmission and induces hyperalgesia and increased sensitivity of dorsal horn neurons to primary afferent stimulation (484, 485). Activated astrocytes also release CSF-1 (187) thereby amplifying signaling via the CSF-1-microglia-BDNF cascade (Figure 3A). Astrocytes also release the NMDA receptor co-agonist D-serine (486) thereby further augmenting overall dorsal horn excitability.

In a similar fashion to BDNF, IL-1β increases glutamate release from primary afferents and augments excitatory synaptic transmission between primary afferent C-fibres and lamina 1 neurons. It also amplifies Ca²⁺ responses evoked by exposure of neurons to 20 mM K⁺ (143, 407, 483).

Like BDNF, IL-1β also does not affect the membrane potential or rheobase of lamina II neurons, suggesting that most of its effect on dorsal horn excitability can be ascribed to changes in synaptic transmission (143, 144). Exposure of rat spinal cord to IL-1β for 6–8 d increases the amplitude of spontaneous EPSC's (sEPSC) in putative excitatory ‘delay’ neurons, and decreases the frequency of spontaneous IPSC's (sIPSC). These actions are similar but not identical to those seen with BDNF or peripheral nerve injury (414, 454, 455). Acute application of IL-1β increases the amplitude of AMPA and NMDA currents dorsal horn neurons (487). Its effect on glutamate release can be ascribed to augmentation of presynaptic NMDA receptor function (483) where signaling between IL-1r and NMDA involves the sphingomyelinase/ceramide pathway (477, 483).

Taken together, all of these actions of IL-1β would be expected to increase dorsal horn excitability and to facilitate the transfer of nociceptive information.

TNF-α decreases the excitability of a subset of spinal GABAergic neurons by suppression of current through HCN channels (488). These effects diminish with time suggesting TNF-α may be primarily involved with the induction rather than the persistence of neuropathic pain (489). As might be expected, blockade of TNF receptor 1 attenuates signs of neuropathic pain in the CCI model but this only occurs in males and not in female rodents (150).

Although FKN action on microglia and potentiation of synaptic transmission in the dorsal horn involves IL-1β but not TNF-α (407), it does appear to be inolved in the generation of a phenomenon named “gliomic LTP” (151, 490). By contrast with classical LTP which is highly localized, “gliomic LTP” spreads extensivly throughout the dorsal horn by the action of TNF-α and of the NMDA co-agonist D-serine (490).

IL-17 is expressed in spinal astrocytes and its cognate receptor is expressed in neurons, especially by those expressing somatostatin (133). SNI-induced static and dynamic allodynia are prevented by intrathecal injection of IL-17 neutralizing antibody and attenuated in IL-17a mutant mice. IL-17 neutralizing antibodies supress LTP of C-fiber evoked field potentials in spinal cord and intrathecal injection of IL-17 or its overexpression in astrocytes produces mechanical allodynia and facilitates spinal LTP (134). IL-17 also supresses inhibitory transmission and enhances excitatory transmission in spinal lamina II_o (133). It may thus serve both as a primary and tertiary mediator (Table 1) but the mechanism of its release from astrocytes is yet to be determined.

Several detailed reviews address supra-spinal changes associated with neuropathic pain (491, 492).

Blood borne inflammatory mediators (38, 493) generated at the site of injury open tight junctions between capillary endothelial cells leading to increased permeability of the blood-brain barrier (192). This allows supra-spinal neurons to interact with blood cells and respond to the cytokines and chemokines they produce (195). Following peripheral injury, afferent information is modulated in various thalamic nuclei (494), somatosensory cortex (495), insular and anterior cingulate cortex (491, 496), nucleus accumbens, and amygdala (497–500). Ascending pathways also interact with the mesolimbic dopamine system (501),

Peripheral nerve injury changes the properties of microglia in the contralateral thalamus, sensory cortex and amygdala as might be expected from the known anatomical arrangement of ascending sensory fibres. Brain regions not directly involved in either sensory or affective aspects of pain such as the motor cortex, do not display altered microglial function (497). This selective modulation of microglia and immune cells in nociceptive pathways (497) may be a consequence of localized neurogenic neuroinflammation as a result of enduring intense activity (329).

Cortico-cortical or cortico-subcortical interactions contribute to the co-morbidies seen in some patients. For example, one form of long-term potentiation (LTP) in the anterior cingulate cortex (ACC) which is triggered by the activation of NMDA receptors and expressed by an increase in AMPA-receptor function, sustains the affective component of the pain state. Another form of LTP in the ACC, which is triggered by the activation of kainate receptors and expressed by an increase in glutamate release, may contribute to pain-related anxiety (491).

There are several parallels between injury-induced cellular changes in higher centres and those seen in the periphery or spinal cord. For example, peripheral neuropathy induces HCN channel dysfunction in medial prefrontal cortex (502) and thalamus (503, 504) and Na_v1.3 function is altered in thalamic neurons (505, 506). Both channel types are similarly affected by peripheral nerve injury (208). These findings are fortuitous in terms of drug action and identification of therapeutic targets; drugs developed to act peripherally may also exert beneficial effects as a result of similar central actions.

Cortical structures modulate nociception through descending control of spinal circuitry (507). This occurs by direct corticospinal projections as well as activation of structures in the brainstem such as the periaqueductal grey matter, locus coeruleus, raphe nuclei and rostroventral medulla (492). Descending inhibition of spinal nociceptive processing is mediated via 5HT₇ receptors and α₂ adrenoceptors whereas serotonergic activation of metabotropic 5HT₂ receptors and ionotropic 5HT₃ receptors facilitates transmission (508–512). This explains the effectiveness of noradrenaline-serotonin reuptake inhibitors (SNRI) in pain management (16) and the limited efficacy of selective serotonin reuptake inhibitors.

There is normally a balance between descending inhibition and excitation but after peripheral nerve injury the excitatory processes gain the upper hand (470, 513). These changes have been associated with the persistence as opposed to the onset of pain (514, 515).

Peripheral nerve injury impairs dopamine release in the reward circuitry associated with the mesolimbic system (497, 501). This may also relate to the changes in affect (anxiety, depression) experienced by neuropathic pain patients (516). Peripheral nerve injury selectively increases excitability of the nucleus accumbens indirect pathway spiny projection neurons and alters their synaptic connectivity. In addition, tactile allodynia can be reversed by inhibiting and exacerbated by exciting these neurons. This suggests that neurons in the nucleus accumbens not only participate in the central representation of pain, but that they may gate activity in ascending pathways associated with expression of pain in higher centres (517).

Classical rodent pain models such as SNI (spared nerve injury), CCI (chronic constriction injury), SNL (spinal nerve ligation) or CCDRG (chronic constriction of DRG) have revealed general principles that help to explain the aetilogy of neuropathic pain. These include the identification of various chemokines, cytokines, neuropeptides and growth factors as primary, secondary or tertiary mediators, the concept of neuroinflammation and bidirectional signalling between neurons and immune cells, alterations in synaptic transmission, ion channels and descending modulation, the roles of microglia and astrocytes, central sensitization and role of peripheral spontaneous activity (12, 15, 20, 22, 27, 47, 68, 183, 208, 210). These findings fall short of addressing the multiplicity of chronic pain presentations in the clinic (8, 47) as even in animal models, different types of nerve injury provoke distinct behavioral, physiological and cellular responses.

For example, mechanical allodynia produced by CCI is short-lived and recovery is seen in about 4 weeks whereas that produced by SNI persists for 7 weeks or more (61, 72). Similarly, changes in synaptic transmission in lamina II neurons are more robust after sciatic CCI than after complete sciatic nerve section (axotomy) (455). These findings relate to the observation that CCI promotes stronger and more long lasting upregulation of the inflammatory mediators IL-1β, TNF-α, IL-10, MCP-1/CCL-2 in nerve stumps than nerve crush (354), Recent work has also shown that glycine inputs onto radial neurons in spinal lamina II are reduced following partial nerve ligation (PNL) of the sciatic nerve, this finding was not seen in animals subject to CCI (63).

Whilst neuropathic pain associated with multiple sclerosis is characterized by loss of spinal neurons (371), this is not seen with CCI (518, 519). Although the NGF binding antibody tanezumab is effective in some pain patients (238), studies in animal models suggest that NGF itself may be effective in management of pain and neuropathy associated with HIV infection (520).

The nature of peripheral injury also dictates the precise spinal circuitry involved in the generation of mechanical allodynia (521). Thus nerve injuries generate allodynia by activation of excitatory neurons that express protein kinase C gamma (PKCγ) (522) whereas mechanical allodynia induced by inflammation involves excitatory neurons that are calretinin positive (523). Cholecystokinin (CCK) positive neurons are important in both situations. Punctate allodynia as produced by Von Frey filaments is distinct from dynamic allodynia that is produced by brushing a cotton swab across the hindpaw skin (521). A subset of CCK positive neurons are primarily involved in conveying dynamic rather than punctate allodynia.

Work using knockout mice has shown that deficiency of CCL19/21 attenuates nerve injury evoked pain but not the hyperalgesia observed in an animal model of multiple sclerosis (116).

This issue of injury-specific mechanisms is starting to be resolved as basic scientists have increasingly turned their attention to disease models rather than classical neuropathic pain models such as CCI and SNI. There are now reliable animal models for diabetic neuropathy (524), multiple sclerosis (34), phantom limb pain (30), chemotherapy induced pain (53, 129, 525), spinal cord injury (31) and trigeminal neuralgia (526).

In the situation of inflammatory as opposed to neuropathic pain, it has recently been reported that nociceptor-neuroimmune interactomes reveal cell type- and injury-specific pathways in three different inflammatory models (527). The availability of a similar database in the neuropathic pain field would be of great advantage to developing specific treatments.

Another major step forward from the basic science perspective is the ongoing improvement in pain assessment in animal models. Regardless of the type of nerve injury used, preclinical effectiveness of therapeutic interventions has classically been assessed in rodent models by examining drugs’ ability to attenuate withdrawal responses to stimuli that would normally be innocuous (47, 60, 61). This typically involves measurements of mechanical or thermal withdrawal thresholds to quantify hyperalgesia or allodynia. Such responses are difficult to quantify as they may be influenced by the subjective impressions of the investigator as well as the olfactory signals they emit. For example, male investigators promote analgesia in female mice (528). In addition, withdrawal responses to innocuous stimuli in injured animal may simply reflect activation of spinal reflexes (529, 530) rather than bona fide manifestations of pain. This may help to explain why classical rodent models have limited ability to predict clinical efficacy (17, 47, 529, 531). In view of this, non-invasive models for objective assessment of chronic pain have been developed. These involve assessment of hypersonic vocalisation, facial grimace score, quantification of social interaction, rearing and nest-building (47, 532–537) and the use of operant models in which the animal is required to make a decision based on the cortical processing of a noxious stimulus (538–540).

The use of operant and non-invasive protocols to effect translation between preclinical observations and development of effective therapeutic approaches may be further refined by combining findings from as many as 6 operant protocols (534).

Advances in technology now permit the use of human nerves in the laboratory (475). Because this has identified the cellular basis for differences in nociceptive processing between humans and rodents (541), the use of such models may be a way forward for identification of more relevant therapeutic targets. Human nociceptors are more heterogeneous than those in rodents and there are also differences in ion channel function and expression leading to differences in cellular excitability (19, 542, 543). Most human DRG neurons exhibit TRPV1 receptor channels but these are expressed exclusively in peptidergic nociceptors in rodents (544). A subpopulation of human DRG neurons display a relatively large constitutive Ca²⁺ channel current and although HVA Ca²⁺ current density is significantly smaller in human than in rodent DRG, the proportion of nifedipine-sensitive (Ca_v1.2) currents is much greater (543). Although this identifies dihydropyridines as a potential therapeutic approach to some types of neuropathic pain, their further development is limited by their propensity to produce postural hypotension (545).

Contemporary methodologies that allow the collection of data from human nerves include observation of nociceptor morphology in skin biopsy samples (546) and use of explant cultures of DRG neurons from aborted fetuses (49). Human DRG's have also been acutely isolated from organ donors or cadavers or from patients undergoing surgical treatment for spinal reconstruction (475, 543, 547).

The use of human induced pluripotent stem cells (hiPSC) differentiated into nociceptive sensory neurons may provide a means to address the limited availability of human DRG neurons (548–554). The use of hiPSC has the advantage of providing large numbers of human neurons, glia and immunocompetent cells (555). This in turn allows the application of high throughput technologies to screen small molecule therapeutic agents to modify nociceptor function (556).

In recent years, considerable attention has been paid to analysis of molecular mechanisms of pain in male vs. female rodents (10, 362, 381, 475). As already emphasized, microglia are not required for mechanical sensitivity to pain in female mice as they require activation of adaptive immune cells such as T-lymphocytes (362, 381). The difference may result from a lack of P2X4 receptors in the microglia of females (364, 376). Despite this, behavioral responses to nerve injury in female rats are similar to those seen in males and both involve downregulation of KCC2 and perturbation of Cl- gradients (363). Because BDNF is not necessary for the development of allodynia in females (362), the mediator released from adaptive immune cells remains to be determined. The possible involvement of IFN-γ has already been alluded to.

Numerous differences in pain mechanisms in males vs. females have emerged over recent years (10). For example loss of GABA_A receptors containing the α6 subunit plays a predominant role in female rodents (557). The relative importance of CGRP in females (323), the role of macrophage derived IL-23 (368), and the absence of functional P2X receptors on microglia of female rodents have already been alluded to (364).

This realization has obvious implications for the design of clinical trials (10); potential new therapies must be evaluated in women and men as separate subgroups of patients.

Given the drawbacks of classical therapeutic approaches; drug toxicity, off target effects, drug interactions and in some cases drug dependence, there is a movement in all fields of medicine towards genetic rather than pharmacological approaches to disease management. Pain management is no exception to this trend (572).

By way of demonstration of the principle, CRISPR (clustered regularly interspaced short palindromic repeats) technology has been used in a mouse model to prevent expression of Na_v1.7 by editing a regulatory sequence (573). This technology might have therapeutic potential in management of persistent pain states.in the clinic.