Pathogenesis of HIV‐associated sensory neuropathy: evidence from in vivo and in vitro experimental models

HIV‐associated sensory neuropathy (HIV‐SN) is a frequent neurological complication of HIV infection and its treatment with some antiretroviral drugs. We review the pathogenesis of the viral‐ and drug‐induced causes of the neuropathy, and its primary symptom, pain, based on evidence from in vivo and in vitro models of HIV‐SN. Viral coat proteins mediate nerve fibre damage and hypernociception through direct and indirect mechanisms. Direct interactions between viral proteins and nerve fibres dominate axonal pathology, while somal pathology is dominated by indirect mechanisms that occur secondary to virus‐mediated activation of glia and macrophage infiltration into the dorsal root ganglia. The treatment‐induced neuropathy and resulting hypernociception arise primarily from drug‐induced mitochondrial dysfunction, but the sequence of events initiated by the mitochondrial dysfunction that leads to the nerve fibre damage and dysfunction are still unclear. Overall, the models that have been developed to study the pathogenesis of HIV‐SN, and hypernociception associated with the neuropathy, are reasonable models and have provided useful insights into the pathogenesis of HIV‐SN. As new models are developed they may ultimately lead to identification of therapeutic targets for the prevention or treatment of this common neurological complication of HIV infection.


Introduction
Between 30% and 60% of individuals infected with the human immunodeficiency virus (HIV) develop HIV-associated sensory neuropathy (HIV-SN), a peripheral sensory neuropathy that is frequently painful, and is characterised by length-dependent degeneration of myelinated and unmyelinated nerve and improved virologic control (Smyth et al., 2007;Ellis et al., 2010;Maritz et al., 2010;Phillips et al., 2010).
In the last decade, significant progress has been made in developing and characterising in vivo and in vitro experimental models of HIV-SN. Here we review studies that have investigated primary lesions to the peripheral nervous system by HIV or antiretroviral drugs. We discuss whether the in vivo models adequately reflect the pathological processes that occur in HIV-SN, and how data from in vivo and in vitro studies have advanced our understanding of the pathophysiology of HIV-SN and its primary symptom, pain. We do not provide a detailed description of the animal models of HIV-SN that have been developed since they have recently been reviewed (Bhangoo et al., 2010). The review is divided into two parts: part 1 describes experimental models of HIV-mediated neuropathy, part 2 describes experimental models of antiretroviral drug-induced neuropathy and the neurotoxic interaction that occurs between the virus and antiretroviral drugs.

Virus-Mediated Neurotoxicity in the Peripheral Nervous System
HIV-SN is characterised by distal axonal degeneration and reduced nerve fibre density along the peripheral nerve trunk (de la Monte et al., 1988;Chaunu et al., 1989;Rizzuto et al., 1995), neuronal loss in the dorsal root ganglia (DRG) (Rance et al., 1988), and dyingback of peripheral and central terminals of peripheral nerve fibres (Rance et al., 1988;Holland et al., 1997;Polydefkis et al., 2002). Even in the absence of clinical features of HIV-SN, post-mortem neuro-anatomical studies have demonstrated immune cell infiltration and inflammatory mediator release in peripheral nerve trunks and DRG of individuals with advanced disease (de la Monte et al., 1988;Mah et al., 1988;Yoshioka et al., 1994;Rizzuto et al., 1995;Nagano et al., 1996;Bradley et al., 1998;Jones et al., 2005). Neuronal damage in HIV-infected individuals is unlikely to result from active infection of neurones, especially because neurones do not express CD4 receptors, which are required for viral entry into a cell (Dubois-Dalcq et al., 1995;Jones et al., 2005;Acharjee et al., 2010). Rather, the infiltration of activated immune cells into peripheral nerves indicates that peripheral nerve damage may occur indirectly through inappropriate activation of immunocompetent cells by the virus or viral proteins (Tyor et al., 1995;Keswani et al., 2002). Alternatively, the damage may be mediated by direct viral-mediated neurotoxicity, similar to that observed in the central nervous system, but such evidence is scarce in the peripheral nervous system (Ahmed et al., 2009;Acharjee et al., 2010). As active viral infection of nerves does not appear to be important in the neurotoxicity of HIV, research on the neuropathology of HIV-SN has focused on experimental models employing soluble viral gene products rather than active virus. However, there have been several studies which have employed recombinant HIV-1 clones generated from virus derived from peripheral nerves or brain tissue of HIV-infected individuals (Zhang et al., 2003;Jones et al., 2005), or models employing lentivirus infections that result in immunodeficiency in felids (Zhu et al., 2007) and non-human primates (Laast et al., 2007;Lehmann et al., 2011).
The primary viral gene product used in vivo and in vitro to investigate the pathogenesis of HIVmediated neurotoxicity is HIV-1 gp120, the coat protein that mediates the binding to and transmission of HIV into permissive cells through interactions with the CD4 protein and the chemokine co-receptors, CCR5 and CXCR4. Even in its soluble form, as occurs when gp120 is shed from viral particles or released from infected cells, the molecule is able to activate sensitive cells (Conti et al., 2004). Additionally, HIV-1 viral protein R (Vpr), an accessory protein that regulates HIV-1 infectivity through a variety of mechanisms and that has been identified in macrophage-like cells in the DRGs of HIV-infected individuals, has been used experimentally to assess the neurotoxicity of viral gene products in vitro (Romani and Engelbrecht, 2009;Acharjee et al., 2010).

Pathogenesis of HIV-SN (Fig. 1)
Acute perineural application of gp120 to rat sciatic nerves caused acute axonal swelling and increased tumour necrosis factor (TNF)-α expression in the nerve trunk at the site of application (Herzberg and Sagen, 2001), in the DRG of the exposed nerve, and in glial cells in the dorsal horn of the spinal cord at the level the exposed nerve enters the spinal cord (Zheng et al., 2011b). In addition, macrophage infiltration into the nerve at the site of gp120 application has been demonstrated within 1 week of gp120 application (Wallace et al., 2007a). When the macrophage accumulation in the nerve trunk was at its greatest, about 2 weeks after gp120 application, there was associated dying-back of distal nerve fibre endings in the epidermis, and macrophage infiltration, activation of satellite cells, and inflammatory mediator production occurred in the DRG (Wallace et al., 2007a;2007b). The aforementioned changes were preceded by expression of ATF3, a marker of cellular stress and neuronal injury, in DRG neurones (Wallace et al., 2007a). Thus, perineural application of gp120 produced a localised inflammatory reaction similar to that seen in peripheral nerve trunks of individuals infected with HIV (de la Monte Figure 1. Pathogenesis of HIV-induced peripheral nerve fibre damage. In dorsal root ganglia, activation of Schwann cells and recruitment of macrophages following gp120-mediated activation of chemokine receptors CXCR4 and/or CCR5 causes the release of inflammatory mediators by these cells, and ultimately causes apoptosis. In the nerve trunk, activation of apoptotic pathways by gp120 contributes to axonal degeneration and intraepidermal nerve fibre die-back. Axonal Schwann cells provide a barrier to gp120-mediated axonal damage. The details of pathways shown with dashed lines are uncertain. Mah et al., 1988;Nagano et al., 1996), and these localised changes were associated with pathological changes characteristic of the distal nerve fibre ending and DRG changes that occur in HIV-SN (Yoshioka et al., 1994;Nagano et al., 1996;Bradley et al., 1998;Skopelitis et al., 2007;Zhou et al., 2007). However, unlike individuals infected with HIV, acute perineural application of gp120 in rats did not cause significant demyelination and axonal degeneration at the site of gp120 application, or apoptosis in DRG cells (de la Monte et al., 1988;Chaunu et al., 1989;Rizzuto et al., 1995;Wallace et al., 2007a). These differences from the clinical state may reflect acute and localised exposure to gp120 in the model compared to chronic and generalised exposure to gp120 that occurs in HIV-infected individuals. Indeed, in a primate model of HIV-SN in which macaque monkeys were infected with simian immunodeficiency virus (SIV), the infection resulted in neuronal loss in the trigeminal nerve ganglia of the monkeys (Laast et al., 2007) and reduced conduction velocity in C-fibres (Laast et al., 2011) that correlated with accumulation of infected macrophages in the DRG. Similarly, in a feline model of HIV-SN in which cats were infected with feline immunodeficiency virus (FIV), infection was associated with reduced axonal density and dying-back of distal nerve fibre endings in the sural nerve (Zhu et al., 2007).
Studies using cultured primary rat and human DRG neurones have shown that the neurotoxic effects of gp120 may be directly mediated by activation of chemokine receptors on the surface of neurones, or indirectly through the activation of Schwann cells and macrophages. Exposing co-cultures of rat DRG neurones and Schwann cells to gp120 caused dose-dependent neurite degeneration and neuronal apoptosis via caspase-3 and mitochondrial-dependent neuronal apoptosis involving inositol triphosphatedependent calcium release from endoplasmic reticulum (Keswani et al., 2003b;Hoke et al., 2009). Activation of pro-apoptotic c-Jun N-terminal kinasedependent pathways also may have contributed to the neurite retraction and apoptosis induced by gp120 (Bodner et al., 2004). The toxicity observed by Keswani et al. (2003b) required the presence of Schwann cells, and based on the available evidence, they proposed that binding of gp120 to CXCR4 receptors on Schwann cells stimulated the release of the chemokine CCL5, which activated CCR5 receptors on DRG neuronal cells. This neuronal activation induced TNF-α release from the neuronal cells which, through an autocrine and/or paracrine mechanism, activated neuronally expressed TNFR1 receptors, leading to cell death (Keswani et al., 2003b;Melli et al., 2006). This proposed mechanism recently gained support from in vivo experiments, which showed increased expression of TNF-α in the DRG of exposed nerves following perineural application of gp120 in rats (Zheng et al., 2011b). And, consistent with the findings of Keswani and colleagues, exposing mixed primary neuronal/glial/macrophage cultures from transgenic rats expressing human CD4/CCR5 receptors to recombinant HIV-1 clones containing the C2V3 region of the HIV env gene derived from virus isolated from peripheral nerves of HIV-infected individuals activated Schwann cells and induced neurite retraction (Jones et al., 2005).
Activation of the non-specific immune response probably also contributes to the indirect neurotoxicity of the virus. In vivo trigeminal nerve ganglionitis in SIV-infected macaques was associated with infiltrating mononuclear cells, evidence of neuronophagia, and neuronal loss and replacement by satellite cells and immune cells (Laast et al., 2007), while in vitro, gp120 induced neuronal cell lysis in DRG cultures through activation of complement pathways (Apostolski et al., 1994), and several studies have reported that inflammatory mediators released by HIVinfected macrophages may contribute to the indirect neurotoxicity of the virus (Zhang et al., 2003;Jones et al., 2005;Hahn et al., 2008). Hahn and colleagues (2008) showed that supernatant from cultured human monocytic cells infected with HIV inhibited neurite growth in cultured human foetal DRGs even when neuronally expressed CCR5 and CXCR4 receptors were neutralised. Inhibition of neurite growth was associated with depolarisation of the mitochondrial membrane and generation of reactive oxygen species. Unfortunately, the substances in the supernatant that mediated the neurotoxicity were not determined, but presumably inflammatory substances such as interleukin(IL)-1β and TNF-α, which are involved in neuronal death in the central nervous system of HIVinfected individuals (Kaul and Lipton, 2006), and which have been found in the DRG and peripheral nerve trunks of people infected with HIV (Yoshioka et al., 1994;Nagano et al., 1996;Jones et al., 2005), may have been responsible. Indeed, some recombinant HIV-1 clones derived from virus extracted from peripheral nerves of HIV-infected individuals increased the expression of IL-1β and TNF-α mRNA in cultured rat DRG, with the most likely source being macrophages (Jones et al., 2005), and perineural application of gp120 in rats increased expression of TNF-α in the DRG (Zheng et al., 2011b). The ability of recombinant HIV-1 virions to induce production of inflammatory mediators by macrophages has been shown to be dependent on gp120 activating CXCR4 and CCR5 receptors on these cells (Zhang et al., 2003). Mitochondrial dysfunction similar to that reported by Hahn and colleagues (2008) in vitro was recently reported by Lehmann and colleagues (2011) in axonal mitochondria isolated from macaque monkeys infected with SIV. Although they did not identify the mechanism by which SIV infection induced mitochondrial dysfunction, the data from Hahn and colleagues and Lehmann and colleagues indicate that mitochondrial dysfunction may contribute to the virus-induced neuropathy.
In addition to indirect neurotoxicity, there is experimental evidence for direct neurotoxicity of the virus on peripheral nerve fibres, especially along axons, which may also contribute to the neuronal degeneration and dying-back of epidermal nerve fibres observed in affected patients. Using compartmentalised culture systems, it has been demonstrated that selective application of gp120 to the neurites of cultured rat DRG neurones, in the absence of Schwann cells, resulted in direct gp120mediated neurite degeneration (Melli et al., 2006). However, when Schwann cells were present, the magnitude of gp120-mediated axonal degeneration was reduced, probably by shielding the axon from gp120. The direct axonal toxicity of gp120 was mediated through an axonal caspase-3-dependent mechanism, which was initiated following activation of neuronal CXCR4 and CCR5 receptors. Similar caspase-3-dependent axonopathy, which occurs independently of somal apoptotic pathways, has been reported following traumatic nerve injury in rats (Buki et al., 2000). Recombinant HIV-1 virions also caused cell death in human neuroblastoma cells from the LAN-2 cell line through a gp120-dependent effect mediated through activation of CXCR4 and CCR5 receptors (Zhang et al., 2003).
Thus, similar to what has been observed in models of HIV-induced neurotoxicity in the central nervous system (Kaul and Lipton, 2006), gp120 mediated activation of glia and macrophages in the peripheral nervous system can produce indirect neurotoxicity through the release of inflammatory mediators. These in vitro effects fit with the clinical findings of macrophage infiltration and inflammatory cytokine release in peripheral nerve trunks and DRG of HIVinfected individuals (Yoshioka et al., 1994;Nagano et al., 1996;Bradley et al., 1998;Skopelitis et al., 2007;Zhou et al., 2007). In addition, the virus, and in particular the gp120 protein, may cause direct axonopathy. This direct effect may contribute to the dying-back of epidermal nerve fibres that is observed in HIV-infected individuals.
Neuropathic pain-like behaviours in animal models of HIV-SN (Fig. 2) Pain is the primary symptom of HIV-SN (Smyth et al., 2007;Wadley et al., 2011), and consistent with this feature of the disease, perineural exposure of rat sciatic nerve to gp120 resulted in thermal (Herzberg and Sagen, 2001) and mechanical hypernociception; where hypernociception is defined as the enhanced nociceptive sensitivity (Herzberg and Sagen, 2001;Oh et al., 2001;Wallace et al., 2007a;2007b;Zheng et al., Figure 2. HIV-induced hypernociception. Activation of chemokine receptors CXCR4 and/or CCR5 by gp120 causes membrane depolarisation, sensitisation of TRPV1 and bradykinin (BK) receptors, and release of Substance P, all of which contribute to the development of acute hypernociception. gp120 also causes macrophage infiltration along the nerve trunk and the release of pro-hypernociceptive cytokine TNF-α. Sustained exposure to gp120 leads to release of pro-hypernociceptive molecules CCL2 and TNF-α in the dorsal root ganglia (dashed line). Ultimately, hyperactivity in the affected primary afferents results in gliosis in the dorsal horn of the spinal cord, which contributes to the maintenance of the hypernociceptive state. Viral protein R (Vpr) may also contribute to the development of hypernociception through membrane depolarization and release of interferon-γ in the dorsal root ganglion (dashed lines). However, the mechanisms behind these processes are largely unknown. The details of pathways shown with dashed lines are uncertain. 2011b). Oh and colleagues (2001) showed that injection of gp120 or chemokines that activate CXCR4 and CCR5 receptors into the foot pad of rats produced significant mechanical hypernociception. At the cellular level, these same ligands caused Substance P release, and induced direct depolarisation of cultured rat DRG, which were putative nociceptive DRG cells (Oh et al., 2001). Thus, binding of gp120 to its chemokine receptors causes activation of nociceptive neurones, and this activity may have contributed to the hypernociception observed in animals exposed to gp120 and the pain reported by HIV-infected individuals. Similar increases in neuronal excitability also occurred when the viral accessory protein, Vpr, was applied to cultured rat and human DRG cells, indicating that viral gene products other than gp120 may also contribute to the development of hypernociception, but it is unknown what cell types were being assessed (Acharjee et al., 2010).
Exposure of nerve fibres to gp120 through intraplantar injection or direct application to the sciatic nerve trunk probably only results in acute exposure to gp120 before it is denatured or cleared by the immune system. Yet, pain-like behaviours described by several investigators lasted several weeks, indicating that more long-term changes in the processing of nociceptive signals had occurred (Herzberg and Sagen, 2001;Wallace et al., 2007a;2007b;Zheng et al., 2011b). The sustained thermal (Herzberg and Sagen, 2001) and mechanical (Herzberg and Sagen, 2001;Wallace et al., 2007a;2007b;Zheng et al., 2011b) hypernociception observed in rats correlated loosely with increased TNF-α expression at the site of gp120 application (Herzberg and Sagen, 2001), in the DRG of the exposed nerve and in the dorsal horn of the spinal cord at the level of entry of the exposed nerve (Zheng et al., 2011b), and macrophage accumulation at the site of gp120 application (Wallace et al., 2007a;2007b). Furthermore, hypernociception to cold and mechanical stimuli also developed in immunodeficient transgenic mice expressing Vpr in monocytoid cells (Acharjee et al., 2010). This hypernociception was associated with increased expression of interferon-γ , but not IL-6, in DRG and sciatic nerves of the mice. Thus inflammatory changes caused by at least two viral gene products of HIV have been shown to induce inflammatory changes in peripheral nerves associated with hypernociception.
In addition to changes in the peripheral nervous system, exposing the sciatic nerve of rats to gp120 resulted in activation of microglia and astrocytes in the dorsal horn of the spinal cord (Herzberg and Sagen, 2001;Wallace et al., 2007a;2007b;Zheng et al., 2011b). The time course of this gliosis coincided with the development of peak hypernociception (Wallace et al., 2007a), and its maintenance for up to 30 days post-gp120 exposure (Herzberg and Sagen, 2001). Zheng and colleagues recently demonstrated that this gliosis is associated with expression of TNF-α by astrocytes and microglia, and that neutralising TNF-α attenuated gp120-mediated mechanical hypernociception in rats, as did neutralising TNF-α in the DRG (Zheng et al., 2011b). Thus, increased expression of TNF-α in the central and peripheral nervous system appears to have a pivotal role in the maintenance of the hypernociceptive state in animals exposed to gp120. Another possible contributor to hypernociception is the pronociceptive chemokine CCL2, which is increased in the DRG of the exposed nerve root (Wallace et al., 2007b;White et al., 2007). Thus, the pain experienced by individuals with HIV-SN may arise from direct stimulation of nociceptive neurones by gp120 or Vpr, and be augmented and sustained through inflammatory reactions in peripheral nerves, DRG, and spinal cord dorsal horns, resulting in peripheral and central sensitisation of nociceptive pathways. The changes described in animals exposed to gp120 are consistent with those described in other rodent models of painful peripheral neuropathy (Latremoliere and Woolf, 2009;Thacker et al., 2007).
In summary, in vivo and in vitro evidence indicates that HIV may cause nerve pathology via direct and indirect mechanisms, with direct mechanisms dominating the axonal pathology and indirect mechanisms dominating the somal pathology. This dual effect of the virus on peripheral nerve fibres, in terms of direct and indirect mechanisms of action, and somal vs. axonal sites of action, may be additive or synergistic, enhancing the overall nerve pathology caused by HIV infection, leading to the high incidence of neuropathy in the infected population. However, data from in vivo and in vitro studies need to be considered in the context of the pharmacological doses of gp120 used in these studies (Klasse and Moore, 2004), the acute onset of the toxicity in the models compared to the slower progression to clinically apparent neuropathy in HIV-infected individuals, the potentially confounding effects of species differences in receptor affinity for viral gene products, and differences in receptor tropism depending on the source of the gp120 used.
To overcome these problems further, use of more sophisticated experimental models, such as those models that employ recombinant HIV-1 clones derived from nervous system tissues of infected patients, and transgenic animal expressing viral gene products or myelogenous cells expressing human CD4, CXCR4, or CCR5 receptors, needs to occur.

Antiretroviral Drug-Mediated Neurotoxicity in the Peripheral Nervous System
The use of NRTIs such as zalcitabine, didanosine, and stavudine to treat HIV infection is strongly associated with the development of HIV-SN (Dalakas, 2001;Dalakas et al., 2001;Cherry et al., 2009;Ellis et al., 2010). Of the three drugs, stavudine is the only one currently in widespread use (WHO et al., 2010). Experimental evidence for the pathological processes by which NRTIs may cause neuropathy is growing and indicates a pivotal role for mitochondrial dysfunction. However, elucidation of the specific neuropathological processes that ultimately cause nerve fibre dysfuncion is confounded in HIV-infected individuals by concomitant viral-induced nerve damage. Therefore, delineation of the mechanisms that underlie the drug-induced component of the neuropathy has required the development of in vitro and in vivo models of pure NRTI-induced neurotoxicity.

Toxicity of NRTIs in vitro
Zalcitabine, didanosine, and stavudine induce dose-dependent inhibition of neurite outgrowth and mitochondrial DNA synthesis in neuronal cell line PC-12 cells (Cui et al., 1997), neurite degeneration in disassociated embryonic and neonatal rat DRG neurones (Keswani et al., 2003a;Bodner et al., 2004), and reduced soma size and neurite length in feline DRG neurones (Zhu et al., 2007). The reduced soma size and neurite length in cultured feline DRG neurones were associated with reduced expression of mRNA for brain-derived neurotrophic factor (BDNF) and its receptor, TrkB. Accordingly, these effects could be reversed by administration of BDNF, indicating that the impaired neuronal growth caused by NRTIs may be secondary to reduced expression and responsiveness to neurotrophic factors (Zhu et al., 2007). In addition to impairing neuronal growth, didanosine induced c-Jun N-terminal kinase-dependent apoptosis of the neonatal DRG cells (Bodner et al., 2004), while zalcitabine reduced PC-12 cell proliferation (Cui et al., 1997) and resulted in a loss of mitochondrial membrane integrity and calcium-dependent necrosis in embryonic rat DRG neurones (Keswani et al., 2003a;. Thus, NRTIinduced cell death may be mediated by apoptotic and necrotic mechanisms. The reduced neuronal growth and increased cell death induced by some NRTIs probably involves disrupted mitochondrial function due to drug-dependent inhibition of mitochondrial DNA polymerase and depletion of mitochondrial DNA (Keilbaugh et al., 1997;Kakuda, 2000). As such, the in vitro data support the findings of Dalakas and colleagues (2001) who reported that the majority of HIV-positive individuals who developed a neuropathy after starting zalcitabine therapy had enlarged, vacuolated mitochondria, and depleted mitochondrial DNA in peripheral nerves and Schwann cells. However, direct disruption of mitochondrial proteins by the drugs may have also contributed to the mitochondrial dysfunction observed in neuronal tissues in vitro and in vivo (Skuta et al., 1999). In the following section, we describe the evidence for, and mechanism of, NRTI-induced neurotoxicity in vivo.

Neuropathological features of NRTI-induced neuropathy in vivo
In the first animal model of NRTI-induced neuropathology, Anderson and colleagues (1992) showed that chronic oral administration of zalcitabine to rabbits resulted in reduced sural nerve conduction velocities and amplitudes. These changes were associated with dose and duration-dependent neuropathology, including myelin splitting, intramyelinic oedema, demyelination/remyelination, and axonal loss. The ultrastructural features of the nerve pathology induced by zalcitabine in rabbits led the authors to conclude that the primary event involved damage to Schwann cells, with secondary axonopathy (Feldman et al., 1992). Similarly, decreases in conduction velocity (Chen and Levine, 2007), myelin hypertrophy and disruption of Remak bundles (Bhangoo et al., 2007), and morphological changes in neuronal cell mitochondria (van Steenwinckel et al., 2008) have been reported to occur in rats within a week of a single parenteral dose of zalcitabine. Thus, even acute exposure to zalcitabine increases risk of neuropathy, which is consistent with the epidemiological data showing that any previous exposure to zalcitabine, didanosine or stavudine increases risk of developing HIV-SN (Smyth et al., 2007;Cherry et al., 2009). However, this acute exposure was not sufficient to cause significant demyelination, remyelination, axonal degeneration (Bhangoo et al., 2007;van Steenwinckel et al., 2008), or decreases in intraepidermal nerve fibre (IENF) density in hind paw glabrous skin of rats (Siau et al., 2006), indicating that these pathological features may require sustained exposure to the drug. Indeed, intraperitoneal injections of zalcitabine three times a week for 3 weeks did decrease IENF density in the plantar skin of the hind paw of rats (Wallace et al., 2007b), which is consistent with the time course of HIV-SN developing in individuals starting antiretroviral therapy (Husstedt et al., 2001). Development of in vivo models using didanosine and stavudine has proven difficult. Whereas chronic, twice daily oral administration of didanosine to rats induced myelin splitting and intramyelin oedema in peripheral nerve fibres (Schmued et al., 1996;Patterson et al., 2000), Zhu and colleagues (2007) failed to induce nerve fibre loss, electrophysiological changes, or dying-back of IENFs in cats administered oral didanosine daily for 6 weeks. Moreover, Warner and colleagues (1995) failed to observe morphological or electrophysiological evidence of peripheral nerve toxicity in rabbits given once daily oral didanosine for 24 weeks, even though the drug was given at near-lethal doses and plasma levels confirmed systemic drug exposure. This study also failed to demonstrate any neurotoxic effects of chronic once daily oral administration of high-dose stavudine. This lack of overt neuropathology in animals administered stavudine has also been reported in rats administered the drug daily for 6 weeks at doses previously shown to induce the development of hypernociception to a noxious mechanical stimulus (Weber et al., 2007;Makweya et al., 2008). Moreover, no signs of neuropathy were reported in monkeys administered the drug on a daily basis for 1 year (Kaul et al., 1999), although the details of how neuropathy was assessed in the monkeys was not provided. The variability in response to didanosine and the lack of neurotoxic effect of stavudine on peripheral nerves in previously reported animal models has meant that zalcitabine-based animal models have formed the basis of investigations into the mechanisms of NRTI-induced neuropathology and neuropathic pain-like behaviours, despite the drug no longer being used clinically.
Pathogenesis of NRTI-induced neuropathy (Fig. 3) The mechanisms underlying neuropathy in animals given zalcitabine have been studied primarily in the context of the neuropathic pain-like behaviours in rodents. In a series of experiments, Levine and colleagues Chen and Levine, 2007) demonstrated that a single intravenous injection, or chronic daily oral dosing, of zalcitabine in rats resulted in dose-dependent mechanical and thermal hypersensitivity of the hind paws, which in the case of intravenous injection, developed rapidly and was sustained for at least 20 days. The hypernociception was associated with a reduction in C-fibre conduction velocity and a change in the distribution of the interspike interval, but not with a decrease in firing threshold or an increase in response to sustained stimulation with threshold or supra-threshold stimuli Chen and Levine, 2007). The behavioural and electrophysiological changes Figure 3. Drug-induced hypernociception. The nucleotide reverse transcriptase inhibitor zalcitabine induces mitochondrial dysfunction associated with altered intracellular calcium homoeostasis and cytochrome c release, which initiates apoptotic pathways. These apoptotic pathways can lead to hypernociception through an imprecisely known mechanism. Activation of apoptotic pathways in the axon may contribute to drug-induced die-back of intraepidermal nerve fibres. In addition, zalcitabine causes activation of Schwann cells and macrophage infiltration into the dorsal root ganglia. Activated Schwann cells express the chemokine CXCL12, which probably mediates hypernociception through activation of CXCR4 receptors expressed on neurones. The dorsal root ganglion neurones also express the pro-nociceptive molecules CCR2 and TNF-α. An astrocytosis develops in the dorsal horn of the spinal cord following exposure to zalcitabine, which contributes to the development of the hypernociceptive state through expression of TNF-α. The details of pathways shown with dashed lines are uncertain. appear to have been related to increased calcium signalling in the nerve fibres  and activation of pro-apoptotic caspase pathways ; data which support a hypothesis of drug-induced mitochondrial damage leading to altered calcium homeostatis and activation of apoptotic pathways. Yet, the same group of investigators also showed that disrupting the mitochondrial electron transport chain and preventing mitochondrial phosphorylation attenuated zalcitabine-induced hypernociception (Joseph and Levine, 2006). These findings are counterintuitive given the authors' earlier findings, and the standard explanation of disrupted mitochondrial function causing the neuropathy in rodents and humans. Indeed, Lehmann and colleagues (2011) recently showed increased mitochondrial DNA mutation deletions and decreased mitochondrial protein levels, including the catalytic subunit of cytochrome c oxidase I, in sural nerve biopsies of HIV-infected individuals who had been exposed to neurotoxic antiretroviral regimens and who had developed HIV-SN. Thus data from affected HIV-infected individuals complement a model of mitochondrial dysfunction.
Recently, Wallace and colleagues (2007b) demonstrated that zalcitabine, injected intraperitoneally three times a week for 3 weeks, induced mechanical, but not thermal, hypersensitivity and anxiety-like behaviour in rats. Overt pathological changes in peripheral nerves, including dying-back of IENFs and macrophage infiltration into the DRG of exposed nerves were also seen. The infiltration of macrophages was accompanied by increased expression of the pro-nociceptive chemokine CCL2, which could have contributed to the animals' hypernociceptive state. Another possible contributor to hypernociception that develops after exposure to zalcitabine is TNF-α. Zheng and colleagues (2011c) showed that a single intraperitoneal injection of zalcitabine in rats that induced sustained mechanical hypernociception was associated with increased expression of the pro-nociceptive cytokine TNF-α by DRG neurones. Unfortunately, neither group of investigators determined whether neutralising TNF-α or CCL2 in the DRG attenuated the hypernociception, so the contribution of these molecules to the hypernociception remains speculative. Indeed, with regards to CCL2, Bhangoo and colleagues (2007) reported that a single intraperitoneal injection of zalcitabine in rats, which induced mechanical hypersensitivity, failed to induce upregulation of CCR2, the receptor for CCL2, in dissociated DRG neurones, or to increase the responsiveness of these cells to application of the chemokine. Rather, Bhangoo and colleagues reported enhanced mRNA expression of the chemokine CXCR4 in DRG neurones and glia, and its ligand, CXCL12, in glia. Blocking CXCR4 blocked the development of the zalcitabine-induced mechanical hypersensitivity, demonstrating the involvement of CXCR4 activation in the development of the hypernociception. Thus, chemokines may be involved in the development of NRTI-induced neuropathy, but the nature and extent of that involvement, and whether there are differences across animal models, is unclear.
The behavioural and neuropathological changes induced by sustained exposure to zalcitabine were also associated with modest increases in dorsal horn astrocyte and microglial activation (Wallace et al., 2007b). However, blocking microglial activation did not attenuate zalcitabine-induced mechanical hypersensitivity, indicating that the microgliosis was not significantly contributing to the hypersensitivity in this model of neuropathic pain. A more recent study also shows that spinal microgliosis does not appear to be responsible for the development of zalcitabine-induced SN (Zheng et al., 2011a). Yet, a single intraperitoneal injection of zalcitabine did result in release of TNF-α by activated astrocytes, and neutralisation of this TNF-α attenuated zalcitabineinduced hypernociception (Zheng et al., 2011c). These conflicting data on the role of spinal glia in mediating the hypernociception induced by zalcitabine may reflect differences in experimental design (e.g., chronic vs. single administration of zalcitabine), or they may reflect the selective importance of activated astrocytes rather than microglia in mediating the central sensitisation of nociceptive pathways. Whatever the case may be, there is accumulating evidence for changes in the dorsal horn contributing to NRTIinduced hypernociception. In a recent study, Renn and colleagues (2011) reported increases in BDNF concentrations in the dorsal horn of mice after a single intravenous injection of stavudine, and that this rise in dorsal horn BDNF was correlated with mechanical hypersensitivity and enhanced activity in dorsal horn wide dynamic range neurones. And, this hypersensitivity was reduced in BDNF heterozygous mice (BDNF +/− ) compared to homozygous (BDNF +/+ ) mice, and in animals depleted of BDNF by intrathecal administration of trkB chimeric protein. Wallace and colleagues (2007b) demonstrated that chronic intraperitoneal administration of zalcitabine induced sustained mechanical hypernociception and some evidence of nerve fibre pathology in rats, but in contrast to their gp120-model of HIV-SN (Wallace et al., 2007a), this treatment did not induce macrophage infiltration of the nerve trunk or the expression of markers of neuronal stress, such as ATF3 and c-Jun, in DRG neurones. Thus, the overt neurotoxicity of the drug was limited compared to that induced by gp120 alone. However, in a model that more closely mimics the clinical situation, where patients exposed to antiretroviral drugs typically also have an underlying HIV infection, the authors reported that the algesic effects of acute perineural exposure to gp120 in rats was enhanced by administration of zalcitabine, and that this enhanced mechanical hypersensitivity was associated with increased expression of CCL2 in the DRG and increased microgliosis in the dorsal horn of the spinal cord. Other studies have also shown positive neurotoxic interactions between NRTIs and viral components. In transgenic mice expressing gp120 under a GFAP promoter, constitutive expression of gp120 within the nervous system did not induce any significant neuropathology, but administration of didanosine in the drinking water of the mice, at a dose that in itself was not neurotoxic, resulted in the development of hypersensitivity to noxious heat. This neurotoxicity was associated with loss of unmyelinated axons in peripheral nerves and dyingback of IENFs; all features of the pathology are seen in humans with HIV-SN . In a feline model of HIV-SN using FIV, administering didanosine to cats infected with FIV enhanced the virus-induced loss of IENFs. Moreover, addition of didanosine to ex vivo preparations of DRG cells from the cats enhanced the reductions in neurite length and soma size, and the reductions in mitochondrial cytochrome c oxidase subunit I and BDNF mRNA induced by FIV (Zhu et al., 2007). The enhanced toxicity of NRTIs, even those like didanosine that do not produce reliable animal models of pure NTRI-induced neuropathy, when given in combination with virus or viral antigens, may explain the continued problem of HIV-SN in individuals on antiretroviral therapy as the use of zalcitabine was stopped. It remains to be determined whether this positive virus-drug interaction will maintain the rate of HIV-SN even with the substitution of stavudine for reverse transcriptase inhibitors like tenofovir, which exhibit very low levels of mitochondrial toxicity in vitro (Lee et al., 2003).

Models of combined NRTI and virus-induced neuropathy
In summary, animal models of pure NRTI-induced neuropathy have yielded mixed results, with only zalcitabine producing reliable models. Even then, the different zalcitabine models have identified slightly different mechanisms for the nerve pathology, especially with regards to the involvement of the chemokine CCL2. But even within models, the nature of the mitochondrial dysfunction induced by zalcitabine, and its contribution to the development of the neuropathy, is equivocal. For example, zalcitabine-induced hypernociception was associated with mitochondrial dysfunction sufficient to initiate mitochondrialdependent caspase signalling, but maintenance of the hypernociception was dependent on sustained functional mitochondria, implying adequate mitochondrial function. Similar findings were reported in another model of drug-induced neuropathy, vincristine-induced neuropathy, suggesting possible overlap in the underlying mechanisms of these two drug-induced neuropathies Joseph et al., 2006). Thus, further investigation into either model may yield answers pertinent to both models. Models of combined NRTI and virus-induced neuropathy, however, have provided useful insight into the positive neurotoxic interactions between the two, and may help explain the continuing problem of HIV-SN in the era of combination antiretroviral therapy.

Conclusion
Models of virus and drug-induced HIV-SN have provided significant insight into the pathological processes that underlie the morphological, and to a lesser extent, the electrophysiological changes seen in patients with the neuropathy. The models of virus-induced neuropathy strongly support direct and indirect mechanisms of viral toxicity, with indirect damage involving inappropriate activation of immunocompetent cells and the subsequent release of inflammatory mediators in peripheral nerves. The in vivo models of the drug-induced neuropathy provide evidence supporting mitochondrial dysfunction as a key process in the development of neuropathological changes, but the sequelae of events initiated by the mitochondrial dysfunction that leads to the neuropathy is still unclear. In general, the in vivo models have construct validity, and the models of pure drug-induced neuropathy and combined drug and virus-induced neuropathy developed by Wallace and colleagues (2007b) appear to have predictive validity also. For example, congruent with efficacy in clinical trials for painful HIV-SN, amitriptyline did not reduce mechanical hypersensitivity in these rat models, but the mixed CB1/CB2 cannabinoid receptor agonist WIN 55,212-2 did (Phillips et al., 2010). Also, the models have started to address the interaction between the viral or druginduced neuropathy and co-morbid conditions that may affect the development and progression of the painful symptoms of the neuropathy. For example, depression has been linked to the presence of pain in individuals with HIV-SN (Lucey et al., 2011) and Wallace and colleagues demonstrated interaction between the presence of pain and anxiety-like behaviour in rat models of HIV-SN (Wallace et al., 2008). Yet, there remains significant scope for the models to also address co-morbid medical conditions such as diabetes mellitus, isoniazid therapy for tuberculosis infection, and nutritional deficiencies (e.g., vitamin B 12 ) that are common in the HIV-infected population and which may increase the risk of developing HIV-SN. Indeed, as we move forward in the development of animal models of HIV-SN we anticipate advances in models that address co-morbid conditions, but probably the most significant advances will come with development of models that better reflect viral exposure of the peripheral nervous system in infected individuals, such as the FIV and SIV models employed by Zhu and colleagues (2007) and Laast and colleagues (2007;, respectively, and the modern drugs and drug combinations that are used clinically to treat HIV. Ultimately, the goal of the models is to provide insight into potential therapeutic targets that may either prevent or attenuate the development of the neuropathy, or treat the symptoms of the neuropathy, of which pain is the primary symptom. To date, the models have not identified potential targets for either of these two goals, and neither have the models identified a clear link between the nature of the nerve damage and whether pain develops or not. However the continued development, study, and refinement of animal models of HIV-SN, such as recent investigations into changes in gene expression in the dorsal horn and DRG cells of animals exposed to systemic NRTIs and gp120 (Dorsey et al., 2009;Maratou et al., 2009;Renn et al., 2011) hold promise for the identification of new and efficacious therapeutic targets for this debilitating neuropathy that is common in the HIV-positive population, and whose symptoms are resistant to conventional treatments.