Introduction

Human immunodeficiency virus-1 (HIV-1) can not only destroy the immune system and lead to acquired immunodeficiency syndrome (AIDS), but the virus can also induce neurological disease that culminates in frank dementia. The worldwide development of HIV-related disease is alarming, with more than 36 million existing infections, and about 20 million deaths.1 AIDS opportunistic infections may affect the central nervous system (CNS), but HIV infection itself can also induce a number of neurological syndromes.2 Interestingly, anemia in HIV-1 infection seems to be an early predictor for a high risk of neuropsychological impairment.3 Neuropathological conditions directly triggered by HIV-1 include peripheral neuropathies, vacuolar myelopathy and a syndrome of cognitive and motor dysfunction that has been designated HIV-associated dementia (HAD).2, 4, 5, 6 A mild form of HAD is termed minor cognitive/motor disorder (MCMD).2, 4, 6, 7

The mechanism of HAD and MCMD remains poorly understood, but the discovery in the brain of cellular binding sites for HIV-1, the chemokine receptors, and recent progress in neural stem cell biology are providing new and hitherto unexpected insights. It is widely believed that HIV entry into the CNS occurs via infected monocytes.8, 9, 10 Interestingly, in the brain, HIV-1 productively infects only macrophages and microglia, but injury and apoptotic death occur in neurons.11, 12 Activation of monocytic cells (macrophages and microglia) through infection, viral proteins or inflammatory mediators and their subsequent release of toxins apparently lead to neuronal and astrocytic dysfunction and thus seem to drive the pathogenesis of HAD.4, 13, 14, 15 However, viral proteins might also directly contribute to neuronal injury. Some of the neurotoxic factors excessively stimulate neurons, thus leading to excitotoxicity with subsequent breakdown in neurons of vital cellular functions in a manner shared with other neurodegenerative diseases.4, 15 Advances in understanding the molecular mechanisms of the disease-defining events provide hope for improved therapeutic intervention.16, 17

Epidemiology of HAD before and in the Era of HAART

In the early 1990s, the prevalence of HAD was estimated to be as high as 20–30% of those individuals with advanced HIV disease and low CD4 cell counts (<200 μl−1).3 The introduction of highly active antiretroviral therapy (HAART) has increased the life expectancy of people infected with HIV-1 and resulted in an at least temporary decrease in the incidence of HAD to as low as 10.5%.18, 19, 20 In fact, a case studied by our group demonstrated near-complete reversal of clinical signs and symptoms of HAD that has been sustained for more than 7 years.21 However, while improvements in control of peripheral viral replication and the treatment of opportunistic infections continue to extend survival times, HAART failed to provide complete protection from HAD or to reverse the disease in most cases.22, 23 This might at least in part be due to poor penetration into the CNS of HIV protease inhibitors and several of the nucleoside analogs.24 Therefore, an early CNS infection might evolve independently over time into a protected brain reservoir. In fact, distinct viral drug resistance patterns in the plasma and cerebrospinal fluid (CSF) compartments have recently been reported.25 Consequently, as people live longer with HIV-1 and AIDS, the prevalence of dementia might be rising and in recent years the incidence of HAD as an AIDS-defining illness has actually increased.22, 26, 27 Furthermore, the proportion of new cases of HAD displaying a CD4 cell count greater than 200 μl−1 is growing,18 and MCMD may be more prevalent than frank dementia in the HAART era.7 However, HAD might currently be the most common cause of dementia worldwide among people aged 40 or less, and it remains a significant independent risk factor for death due to AIDS.7 Thus, a better understanding of the pathogenesis of HAD, including viral and host factors, is needed in order to identify additional therapeutic targets for the prevention and treatment of this neurodegenerative disease.

From HIV Entry into the Brain to Development of MCMD/HAD

Soon after infection in the periphery HIV penetrates into the CNS where the virus primarily resides in microglia and macrophages.8, 9 Viral load in brain can be measured by quantitative PCR, and the highest concentrations of virus are detected in those subcortical structures most often affected in patients with HAD.28, 29 However, infection of macrophages and microglia alone does not seem to initiate neurodegeneration, and it has therefore been proposed that additional factors associated with advanced HIV infection in the periphery, thus outside the CNS, provide important triggers for events leading to dementia.9 An elevated number of circulating monocytes that express CD16 and CD69 could constitute one such factor. These activated cells tend to adhere to and transmigrate through the normal endothelium of the brain microvasculature and might then initiate processes deleterious to neurons.9

The blood–brain barrier (BBB) also plays a crucial role in HIV infection of the CNS.30, 31, 32 Microglia and astrocytes produce chemokines – cell migration/chemotaxis-inducing cytokines – such as monocyte chemoattractant protein (MCP)-1, which appear to regulate migration of peripheral blood mononuclear cells through the BBB.32 In fact, a mutant MCP-1 allele that causes increased infiltration of mononuclear phagocytes into tissues has recently been implicated in an increased risk of HAD.33 Histological studies in specimens from HIV-1-infected humans and SIV-infected rhesus macaques found that lymphocytes and monocytes enter the brain.34, 35 The pathophysiological relevance of CNS-invading lymphocytes in HAD is not clearly established.35, 36 However, infiltrating lymphocytes and activated microglia in brains with HIV-1 encephalitis showed strong immunoreactivity for interleukin (IL)-16, a natural ligand of CD4. Since this cytokine inhibits HIV-1 propagation, lymphocytes might contribute to an innate antiviral immune response in the CNS in addition to microglia.37 Cell migration also engages adhesion molecules, and increased expression of vascular cell adhesion molecule-1 (VCAM-1) has been implicated in mononuclear cell migration into the brain during HIV and SIV infection.30, 31, 38 As an alternative to entry via infected macrophages, it has been suggested that the inflammatory cytokine, tumor necrosis factor-alpha (TNF-α), promotes a paracellular route for HIV-1 across the BBB.39 Interestingly, alterations in the BBB occur even in the absence of intact virus in transgenic mice expressing the HIV envelope protein gp120 in a form that circulates in plasma.40 This finding suggests that circulating virus or envelope proteins may provoke BBB dysfunction during the viremic phase of primary infection. On the part of the host, a vicious cycle of immune dysregulation and BBB dysfunction might be required to achieve sufficient entry of infected or activated immune cells into the brain to cause neuronal injury.4, 41 On the side of the virus, variations of the envelope protein gp120 might also influence the timing and extent of events allowing viral entry into the CNS and leading to neuronal injury.42

Potential Links between Neuropathology of HIV Infection and Pathogenesis of HAD

The neuropathological hallmarks of HIV infection in the brain are termed HIV encephalitis and include widespread reactive astrocytosis, myelin pallor, microglial nodules, activated resident microglia, multinucleated giant cells, and infiltration predominantly by monocytoid cells, including blood-derived macrophages.43 Surprisingly, measures of cognitive function do not correlate well with numbers of HIV-infected cells, multinucleated giant cells or viral antigens in CNS tissue.44, 45 In contrast, increased numbers of microglia,44 elevated TNF-α mRNA in microglia and astrocytes,46 evidence of excitotoxins,47, 48, 49 decreased synaptic and dendritic density,45, 50 and selective neuronal loss51, 52 constitute the pathologic features most closely associated with the clinical signs of HAD. Furthermore, signs of neuronal apoptosis have been linked to HAD,53, 54, 55 although this finding is not clearly associated with viral burden53 or a history of dementia.56 The localization of apoptotic neurons is correlated with evidence of structural atrophy and closely associated with signs of microglial activation, especially within subcortical deep gray structures,56 which may show a predilection for atrophy in HAD.

The neuropathology observed in post-mortem specimens from HAD patients in combination with extensive studies using both in vitro and animal models of HIV-induced neurodegeneration has led to a fairly complex model for the pathogenesis of HAD. The available information strongly suggests that the pathogenesis of HAD might be most effectively explained when viewed as similar to the multihit model of oncogenesis. Figure 1 shows a model of potential intercellular interactions and alterations of normal cell functions that can lead to neuronal injury and death in the setting of HIV infection.4 Macrophages and microglia can be infected by HIV-1, but they can also be activated by factors released from infected cells. These factors include cytokines and shed viral proteins such as gp120. Variations of the HIV-1 envelope protein gp120, in particular in its V1, V2 and V3 loop sequences, have been implicated in modulating the activation of macrophages and microglia.42 Factors released by activated microglia affect all cell types in the CNS, resulting in upregulation of cytokines, chemokines and endothelial adhesion molecules.4, 9, 15 Some of these factors may directly or indirectly contribute to neuronal damage and apoptosis. Directly neurotoxic factors released from activated microglia include excitatory amino acids (EAAs) and related substances, such as quinolinate, cysteine and a not completely characterized amine compound named ‘Ntox’.13, 49, 57, 58, 59, 60, 61 EAAs induce neuronal apoptosis through a process known as excitotoxicity. This detrimental process engenders excessive Ca2+ influx and free radical (nitric oxide (NO) and superoxide anion) formation by overstimulation of glutamate receptors.58, 62 Certain HIV proteins, such as gp120 and Tat, have also been reported to be directly neurotoxic, although high concentrations of viral protein may be needed, or neurons may have to be cultured in isolation to see these direct effects.63, 64 It is important to note that toxic viral proteins among factors released from microglia and glutamate set free by astrocytes may act in concert to promote neurodegeneration, even in the absence of extensive viral invasion of the CNS.

Figure 1
figure 1

Current model of neuronal injury and death induced by HIV-1 infection. Immune-activated and HIV-infected brain macrophages (MΦ)/microglia release potentially neurotoxic substances. These substances include quinolinic acid and other excitatory amino acids (EAAs) such as glutamate and L-cysteine, arachidonic acid, platelet-activating factor (PAF), NTox, free radicals, TNF-α, and probably others. These factors from macrophages and also possibly from reactive astrocytes contribute to neuronal injury, dendritic and synaptic damage, and apoptosis as well as to astrocytosis. Entry of HIV-1 into monocytoid cells occurs via gp120 binding, and therefore it is not surprising that gp120 (or a fragment thereof) is capable of activating uninfected macrophages to release similar factors to those secreted in response to productive HIV infection. Macrophages express CCR5 and CXCR4 chemokine receptors on their surface in addition to CD4, and gp120 binds via these receptors. Some populations of neurons and astrocytes have been reported to also possess CXCR4 and CCR5 receptors on their surface, raising the possibility of direct interaction with gp120. Macrophages/microglia and astrocytes have mutual feedback loops (bidirectional arrow). Cytokines participate in this cellular network in several ways. For example, HIV infection or gp120 stimulation of macrophages enhances their production of TNF-α and IL-1β (arrow). The TNF-α and IL-1β produced by macrophages stimulate astrocytosis. Arachidonate released from macrophages impairs astrocyte clearing of the neurotransmitter glutamate and thus contributes to excitotoxicity. In conjunction with cytokines, the α-chemokine stromal cell-derived factor (SDF)-1 stimulates reactive astrocytes to release glutamate in addition to the free radical nitric oxide [NO], which in turn may react with superoxide (O2•−) to form the neurotoxic molecule peroxynitrite (ONOO). NO might also activate extracellular matrix metalloproteinases (MMPs), which can then proteolytically affect neurons, and also cleave membrane-anchored fractalkine.131, 198 Neuronal injury is primarily mediated by overactivation of NMDARs with resultant excessive influx of Ca2+. This in turn leads to overactivation of a variety of potentially harmful signaling systems, the formation of free radicals and release of additional neurotransmitter glutamate. Glutamate subsequently overstimulates NMDARs on neighboring neurons, resulting in further injury. This final common pathway of neurotoxic action can be blocked by NMDAR antagonists. For certain neurons, depending on their exact repertoire of ionic channels, this form of damage can also be ameliorated to some degree by calcium channel antagonists or non-NMDAR antagonists. Additionally, agonists of β-chemokine receptors, which are present in the CNS on neurons, astrocytes and microglia, can confer partial protection against neuronal apoptosis induced by HIV/gp120 or NMDA. The figure is modified from Kaul et al.4

Chemokine Receptors in HIV-1 Infection and HAD

Chemokine receptors are seven transmembrane-spanning domain, G-protein-coupled receptors, and as such trigger intracellular signaling events. While chemokines and their receptors were originally shown to mediate leukocyte trafficking and to contribute intimately to the organization of inflammatory responses of the immune system, they are now known to contribute to far more physiological and pathological processes.41, 65, 66 The additional functions include the intricate control of organogenesis, including hematopoiesis, angiogenesis and development of heart and brain.67, 68, 69, 70 Furthermore, chemokines and their receptors are essential for maintenance, maturation and migration of hematopoietic and neural stem cells.66, 71 However, the most prominent pathological function of certain chemokine receptors seems to be the mediation of HIV-1 infection.70, 72, 73

Infection of macrophages and lymphocytes by HIV-1 can occur after binding of the viral envelope protein gp120 to one of several possible chemokine receptors in conjunction with CD4. Generally, T cells are infected via the α-chemokine receptor CXCR4 and/or the β-chemokine receptor CCR5. In contrast, macrophages and microglia are primarily infected via the β-chemokine receptor CCR5 or CCR3, but the α-chemokine receptor CXCR4 may also be involved.74, 75, 76, 77 The HIV coreceptors CCR5 and CXCR4, among other chemokine receptors, are also present on neurons and astrocytes,78, 79 although these cells are not thought to harbor productive infection. Several in vitro studies strongly suggest that CXCR4 is directly involved in HIV-associated neuronal damage while CCR5 may additionally serve a protective role.63, 80, 81

In cerebrocortical neurons and neuronal cell lines from humans and rodents, picomolar concentrations of HIV-1 gp120, as well as intact virus, can induce neuronal death via CXCR4 receptors.76, 77, 80, 81, 82 In mixed neuronal/glial cerebrocortical cultures that mimic the cellular composition of the intact brain, this apoptotic death appears to be mediated predominantly via the release of microglial toxins rather than by direct neuronal damage.77, 81, 82 However, nanomolar concentrations of SDF-1α/β interacting with CXCR4 can induce apoptotic death of neurons in the absence of microglia, suggesting a possible direct interaction with neurons while interaction with astrocytes can also occur.81, 83, 84 In contrast to these findings, it has been reported that somewhat higher concentrations of SDF-1α provide neuroprotection from X4-preferring gp120-induced damage of isolated hippocampal neurons.63

Using mixed neuronal/glial cerebrocortical cultures from rat and mouse, we have further investigated the role of chemokine receptors in the neurotoxicity of gp120. We found that gp120 from CXCR4 (X4)-preferring as well as CCR5 (R5)-preferring and dual tropic HIV-1 strains all were able to trigger neuronal death. While gp120 from one out of two X4-preferring HIV-1 strains no longer showed neurotoxicity in CXCR4-deficient cerebrocortical cultures, dual tropic gp120SF2 showed, surprisingly, even greater neurotoxicity in CCR5 knockout cultures compared to wild-type or CXCR4-deficient cultures.85 These findings are consistent with a primarily neurotoxic effect of CXCR4 activation by gp120. In contrast, activation of CCR5 might at least in part be neuroprotective depending on the HIV-1 strain from which a given gp120 originated. Furthermore, we observed earlier that the CCR5 ligands macrophage inflammatory protein (MIP)-1β and regulated upon activation-normal T-cell expressed and secreted (RANTES) protect neurons against gp120-induced toxicity.81

Since in vitro inhibition of microglial activation is sufficient to prevent neuronal death after gp120 exposure, it seems likely that stimulation of CXCR4 in macrophages/microglia is a prerequisite for the neurotoxicity of gp120.76, 81 In contrast, SDF-1 might directly activate CXCR4 in astrocytes and neurons to trigger neuronal death, for example, by reversing glutamate uptake in astrocytes.4, 80, 81, 84 SDF-1 is produced by astrocytes, macrophages, neurons and Schwann cells.83, 86, 87, 88 An increase in SDF-1 mRNA has been detected in HIV encephalitis79 and protein expression of SDF-1 also appears to be elevated in the brains of HIV patients.89 To what degree the increased expression of SDF-1 aggravates neuronal damage by HIV-1 remains to be shown. We had reported previously that intact SDF-1 can be toxic to mature neurons in a CXCR4-dependent manner, at least in culture.81, 83, 85 Additionally, it was recently reported that cleavage of SDF-1 by MMPs may contribute to neuronal injury and thus HAD via a non-CXCR4-mediated mechanism.90 Importantly, increased expression and activation of MMPs, including MMP-2 and MMP-9, were detected in HIV-infected macrophages and also in post-mortem brain specimens from AIDS patients compared with uninfected controls.91 As elegantly shown by Power and colleagues, MMP-2 released from HIV-infected macrophages is able to proteolytically remove four amino acids from the N-terminus of SDF-1. This truncated form of SDF-1 no longer binds CXCR4 and is an even more powerful neurotoxin than full-length SDF-1.90

Effect of Chemokines and HIV/gp120 on Neural Stem and Progenitor Cells

CXCR4 is expressed on neurons, microglia, astrocytes and endothelia in the brain.83, 92, 93 However, this chemokine receptor and its ligand SDF-1 are also major components in many physiological processes involving hematopoietic and neural stem cells.36, 66, 94 This indicates that HIV-1 could also directly interfere with biological functions of neural stem and progenitor cells.

Neural stem cells and later progenitor cells are widely thought to provide a reservoir to replace neurons or glia under conditions of brain injury or disease.95 Neurogenesis can be stimulated by ischemic and excitotoxic brain injuries, physical exercise, diet, learning or an enriched environment,41, 96 and may decrease with aging.97 Functionally, it is suggested that this neurogenesis contributes to long-term synaptic plasticity and cognitive processes and is also involved in pathological processes such as depression.98, 99 Three important steps are involved in neurogenesis and regulation of neural progenitor cells (NPCs): directed migration, proliferation and differentiation.100 Chemokines such as SDF-1 and its receptor CXCR4 appear to play an important role in the process. CXCR4 is highly expressed during development in the cerebellum, hippocampus and neocortex, and this expression persists into adulthood.67, 69, 101, 102, 103 SDF-1 transcripts are predominantly expressed by oligodendrocytes, astrocytes and neurons in the neocortex, hippocampus and cerebellum87, 103 and by meningeal cells.102 In vitro, the production of SDF-1 by purified astrocyte cultures is associated with a macrophage–astrocyte interaction.83 Whether or not SDF-1/CXCR4 interaction is involved in all of the above-mentioned three steps of neurogenesis is not known; however, it has been documented that SDF-1/CXCR4 signaling regulates migration of NPCs in the cerebellum, dentate gyrus and cortex.67, 69, 102, 103 Furthermore, it has recently been reported that the recruitment of CXCR4-positive progenitor cells into regenerating tissue is mediated by a hypoxic gradient and expression of SDF-1 that is induced by the transcription factor hypoxia-inducible factor-1 (HIF-1).104 Thus, several lines of evidence indicate a significant involvement of the SDF-1/CXCR4 interaction in tissue damage and repair.

In order to investigate this possibility, we utilized cultures of primary mouse and human neural progenitor cells obtained during the fetal period. These cells stain positively for the neural stem cell marker nestin and readily undergo cell division. After several rounds of proliferation, the progenitors exit the cell cycle and express neuronal markers such as βIII-tubulin (TuJ1). Our immunocytochemical studies showed that the progenitors are positive for CXCR4 and CCR5. Treatment with HIV-1/gp120 reduced the number of progenitors and differentiating neurons. Accounting for these observations, we found that gp120 inhibited proliferation of neural progenitor cells without producing apoptosis. The resulting decrease in neural stem cell proliferation engendered by gp120 also means that there are fewer progenitor cells present to differentiate in neurons, thus impairing neurogenesis (S Okamoto, S McKercher, M Kaul and SA Lipton, unpublished). Recently, these findings were complemented and extended by others using commercially generated human neural progenitor cells.105, 106 In those experiments, chemokines promoted the quiescence and survival of human neural progenitor cells via stimulation of CXCR4 and CCR3 via a mechanism that involves downregulation of extracellularly regulated kinase-1 and -2 (ERK-1/2) and simultaneous upregulation of Reelin.105 Exposure to HIV-1 appeared to induce quiescence of neural progenitors, also through engagement of CXCR4 and CCR3. The coat protein HIV-1/gp120 reportedly downregulated ERK-1/2 but had no effect on the neuronal glycoprotein Reelin.106 The effects of both the chemokines and HIV-1/gp120 were reversible and could be inhibited with recombinant apolipoprotein E3 (ApoE3), but not ApoE4. The finding that HIV-1/gp120 could indeed interfere with the normal function of neural progenitor cells raised the possibility that the virus contributes to the development of HAD not only by injuring and killing existing neurons but also by preventing potential repair mechanisms in the CNS (Figure 2).

Figure 2
figure 2

Interference of HIV-1/gp120 with the function of neural progenitor cells. Exposure to chemokines, SDF-1 and Eotaxin, or HIV-1/gp120 of mouse or human NPCs reduces proliferation and promotes quiescence. ApoE3 inhibits these effects on NPCs. NPCs express nestin and show decreased proliferation as judged by decreased BrdU incorporation. However, NPCs do not undergo apoptosis, as evidenced by lack of TUNEL staining and nuclear condensation under the same conditions105, 106 (S Okamoto, S McKercher, M Kaul and SA Lipton, unpublished)

Mechanisms of Neuronal Injury and Death in HAD

How HIV infection results in neuronal injury as well as neurocognitive and motor impairment continues to remain a controversial topic. While there is general agreement that HIV does not infect neurons, the primary cause of the neuronal damage remains in question. There is evidence to support multiple theories for neuronal injury by various viral proteins, including Tat, Nef, Vpr and the Env proteins gp120 and gp41.4 These findings have led to at least two different theories on how HIV results in neuronal injury in the brain. The theories can be described as the ‘direct injury’ hypothesis and the ‘indirect’ or ‘bystander effect’ hypothesis. These two theories are not mutually exclusive, and the available data support a role for both, although an indirect form of neurotoxicity seems to predominate.4, 9, 107

The theory that HIV proteins can directly injure neurons without requiring the intermediary function of non-neuronal cells (microglia and/or astrocytes) is supported by experiments showing that viral envelope proteins are toxic in serum-free primary neuronal cultures63 and in neuroblastoma cell lines.108 In these experimental paradigms, the impact of neurotoxic cytokines and EAAs secreted from non-neuronal cells is minimized because serum-free neuronal cultures contain few if any non-neuronal cells, and neuroblastoma lines do not contain cells of other phenotypes. The HIV coat protein gp120 interacts with several members of the chemokine receptor family (see above), and the direct form of HIV-induced neuronal injury may be mediated by chemokine receptor signaling. Indeed, experiments aimed at blocking chemokine receptor signaling can in some cases prevent HIV/gp120-induced neuronal apoptosis.83, 109 Additionally, nanomolar concentrations of gp120 have been reported to interact with the glycine binding site of the N-methyl-D-aspartate-type glutamate receptor (NMDAR),110 suggesting another mechanism by which HIV/gp120 may have a direct effect on neuronal cell death. The HIV-protein Tat (HIV/Tat) can be taken up into PC12 cells by a receptor-mediated mechanism64 and may also have a direct effect on neurons by potentiating the response to excitotoxic stimuli.111 Experiments using cultured hippocampal neurons revealed that the HIV-protein Vpr (HIV/Vpr) may be directly neurotoxic through formation of a cation-permeable channel.112 However, all of these in vitro findings must be interpreted in the context of the limitations of the experimental paradigm and concentration of HIV proteins employed. Most of the experimental results described above were obtained in the absence of non-neuronal cells and therefore a predominantly indirect effect would not be detected. In addition, the concentrations of HIV proteins employed were frequently well above the picomolar or lower range thought to be present in the brain or CSF from patients with HAD.

Apoptotic neurons do not colocalize with infected microglia in HAD patients,113 supporting the hypothesis that HIV infection causes neurodegeneration through the release of soluble factors. Therefore, the propensity for cell–cell interactions mandates that disease pathogenesis in vitro be approached in a ’mixed’ neuronal/glial primary culture system that recapitulates the type and proportion of cells normally found in the intact brain (Figure 1). Systems designed to study the effect of soluble factors released from microglia have included mixed cerebrocortical cultures from human fetal brain directly infected with HIV,113 severe combined immunodeficiency (SCID) mice inoculated with HIV-infected human monocytes,114 and mixed rodent cerebrocortical cultures exposed to picomolar concentrations of the envelope protein HIV/gp120.81, 115, 116

Using such in vitro models, we and others have found evidence for a predominantly indirect neurotoxic effect that occurs due to the response of non-neuronal cells to HIV infection or shed HIV proteins, as described previously. Much of the data supporting the theory of indirect neuronal injury stem from experiments designed to examine the toxicity of HIV envelope proteins or supernatants of infected macrophages.13, 117, 118 Picomolar concentrations of HIV/gp120 induce injury and apoptosis in primary rodent and human neurons.83, 113, 115, 117 In our hands, the predominant mode of HIV/gp120 neurotoxicity to cerebrocortical neurons requires the presence of macrophages/microglia.15, 81 Indeed, HIV-1-infected or gp120-stimulated mononuclear phagocytes release neurotoxins that stimulate the NMDAR, as described earlier. NMDAR antagonists can ameliorate neuronal cell death in vitro due to HIV-infected macrophages or purified recombinant gp120,77, 119, 120 and in vivo in gp120 transgenic mice.121

Excessive stimulation of the NMDAR induces several detrimental intracellular signals that contribute to neuronal cell injury and subsequent death by apoptosis or necrosis, depending on the intensity of the initial insult (Figure 3).62 If the initial excitotoxic insult is fulminant, for example, in the ischemic core of a stroke, the cells die early from loss of ionic homeostasis, resulting in acute swelling and lysis (necrosis). If the insult is more mild, as seen in several neurodegenerative disorders including HAD, neurons enter a delayed death pathway known as apoptosis.62 Neuronal apoptosis after excitotoxic insult involves Ca2+ overload, activation of p38 MAP kinase and p53, release of cytochrome c and other molecules such as apoptosis-inducing factor (AIF) from mitochondria, activation of caspases, free radical formation, lipid peroxidation and chromatin condensation.36, 82, 122, 123, 124 Activated caspase-3 and p53 are prominently detected in neurons of brains from HAD patients, and, in vitro, p53 is indispensable in neurons (and microglia) for HIV-1/gp120 to cause neurotoxicity.82, 125

Figure 3
figure 3

Cellular signaling in HAD – pathways engaged in neuronal injury and apoptosis. (Middle panel) Neuronal signaling in HAD: Overstimulation of the NMDAR is triggered by neurotoxins released from HIV-infected or immune-stimulated macrophages/microglia and by impaired clearance (or release) of glutamate that under normal conditions would have been taken up by astrocytes. Consequently, excessive Ca2+ influx into neurons triggers activation of p38 MAPK and p53, mitochondrial Ca2+ overload and cytochrome c (cyt c) release, free radical generation (nitric oxide [NO] and reactive oxygen species [ROS]), caspase activation, and ultimately apoptosis. NMDARs are physically tethered to neuronal nitric oxide synthase (nNOS), facilitating its activation. NO passing out of the cell may activate MMPs and trigger an extracellular proteolytic pathway to neuronal injury. Inside the cell, the Bcl-2 family members Bad, Bax and Bid promote apoptosis mediated by glutamate, ROS and TNF-α/TRAIL, respectively. Bcl-2 prevents apoptosis, apparently by attenuating cytochrome c release and ROS production. Activation of the p38 MAPK pathway by a Ca2+-mediated mechanism and by oxidative stress may lead to phosphorylation/activation of transcription factors involved in apoptosis, such as p53. Stimulation of the α-chemokine receptor CXCR4 can also induce several pathways in neurons, including activation of p38 MAPK, which leads to apoptosis. In contrast, activation of the β-chemokine receptor CCR5 initiates an as yet uncharacterized neuroprotective pathway that interferes with toxicity triggered by HIV/gp120 or excessive stimulation of NMDARs. The chemokine fractalkine (Fkn) is released from neurons subsequent to excitotoxic injury, and may represent feedback signaling onto non-neuronal cells. (Left panel) Microglial/macrophage signaling in HAD: HIV/gp120 (gp) interacts with chemokine receptors CXCR4 or CCR5 in conjunction with CD4 to stimulate or infect (if the entire virus is present) microglia and macrophages. Natural ligands of CXCR4 (i.e., the α-chemokine SDF-1) and CCR5 (i.e., the β-chemokines MIP-1β and RANTES) interfere with HIV/gp120 binding and signaling. However, only the β-chemokines can prevent the neurotoxic effect of activated microglia and macrophages. Neurons release the δ-chemokine fractalkine (Fkn), which activates microglia. Hence, Fkn may mediate communication between neurons and glia. The HIV envelope protein gp120 triggers a signaling pathway that involves p38 MAPK, a pivotal factor in immune stimulation of macrophages that in turn activates the transcription factor MEF2C, and directly or indirectly also p53. HIV/gp120 induces the release of neurotoxic substances, including EAAs, arachidonic acid and related molecules such as PAF, which engenders neuronal glutamate release. Furthermore, the HIV-1/gp120 also induces production of TRAIL and release of inflammatory cytokines, such as TNF-α. Inflammatory cytokines can activate adjacent microglia/macrophages and astrocytes, and thus indirectly contribute to brain injury. (Right panel) Astrocyte signaling in HAD: Astrocytes express the HIV coreceptors CXCR4 and CCR5 in addition to other chemokine receptors, but lack CD4. Therefore, astrocytic reactivity may be influenced primarily by natural ligands of chemokine receptors. However, via these chemokine receptors, astrocytes may possibly also be stimulated by a CD4-independent effect of HIV/gp120. Astrocytes are activated by inflammatory cytokines, including TNF-α, IL-1β and interferon-γ (IFN-γ). Exposure to arachidonic acid released from macrophages/microglia and cytokine activation results in impaired glutamate uptake, increased glutamate release and induction of iNOS, leading to release of potentially neurotoxic NO. TNF-α, released from macrophages/microglia, and SDF-1 stimulate astrocytes to release glutamate. TNF-α also promotes expression of the astrocytic fractalkine (Fkn) receptor CX3CR1. Stimulation of CX3CR1 on astrocytes induces release of a soluble factor that triggers microglial proliferation. The figure is modified from Kaul and Lipton199

The scaffolding protein PSD-95 (postsynaptic density-95) links the principal subunit of the NMDAR (NR1) with nNOS, a Ca2+-activated enzyme, and thus brings nNOS into close proximity to Ca2+ via the NMDAR-operated ion channel.126 Excessive intracellular Ca2+ overstimulates nNOS and protein kinase cascades with consequent generation of deleterious levels of free radicals, including reactive oxygen species (ROS) and NO.127 NO can react with ROS to form cytotoxic peroxynitrite (ONOO).127 However, in alternative redox states, NO can activate p21ras128 and inhibit caspases129 via S-nitrosylation (transfer of the NO group to critical cysteine thiols), thereby attenuating apoptosis in cerebrocortical neurons. Oxidative processes and cell stress are also reflected by changes to the cellular lipid metabolism, and an increase in ceramide, sphingomyelin and hydroxynonenal has been implicated in the neurotoxic pathways associated with HAD.130

In addition to the intracellular effects of NO and oxidative stress, we have recently identified an extracellular proteolytic pathway to neuronal injury mediated by these effectors. In this pathway, S-nitrosylation (transfer of NO to a critical cysteine thiol group) and subsequent oxidation serve to activate MMP-9 and possibly other MMPs.131 Proteolytically active MMP-9 induces and promotes neuronal death presumably by disrupting the cellular mechanism(s) that allow essential attachment to the extracellular matrix and neighboring neurons.

Furthermore, we have found that neurons exposed to HIV/gp120 and grown in mixed cerebrocortical cultures containing astrocytes and microglia demonstrate release of mitochondrial cytochrome c, caspase activation, chromatin condensation and apoptosis, which is blocked by inhibition of the p38 MAP kinase.81, 125

In addition to chemokines and EAAs, HIV-infected or gp120-activated microglia also release inflammatory cytokines, including TNF-α and IL-1β.46, 132 Among other actions, both of these cytokines stimulate release of L-cysteine from macrophages, and pharmacologic blockade of IL-1β or antibody neutralization of TNF-α prevents this release.60 Under physiological or pathophysiological conditions, L-cysteine can stimulate NMDARs and lead to neuronal apoptosis.60 TNF-α is capable of stimulating apoptosis in human neurons,133, 134 but an indirect route of injury cannot be excluded. Expression of TNF-α and its receptor are elevated in brains from patients with HAD.46 Experiments aimed at addressing the question of interactions between neurotoxins associated with HAD have revealed that TNF-α and HIV/Tat synergize to promote neuronal death, and this effect is prevented by antioxidants.135 It remains possible that TNF-α can activate caspases within neurons via TNF-α receptor-1 (TNFR1), since TNFR1 is found on at least some neurons, and it can trigger caspase-8 activation. Indeed, we have found that antibody neutralization of TNF-α or inhibition of caspase-8 prevents the neurotoxicity of HIV/gp120 in cultured cerebrocortical neurons,125 and caspase-8 activity can directly or indirectly activate caspase-3, leading to apoptosis. These findings suggest that inflammatory cytokines, including TNF-α and IL-1β, may have important synergistic roles in HIV-associated neuropathology.4, 60, 134, 136

Transgenic (tg) mice expressing HIV-1/gp120 in their CNS manifest neuropathological features that are similar to the findings in brains of AIDS patients, including reactive astrocytosis, increased number and activation of microglia, reduction of synapto-dendritic complexity, loss of large pyramidal neurons,137 and induction of MMP-2.138 In addition, these gp120 tg mice display significant behavioral deficits, such as extended escape latency, and reduced swimming velocity and spatial retention.139 In gp120 tg mice, neuronal damage is ameliorated by the NMDAR antagonist memantine.121 Memantine-treated gp120 tg and non-tg control mice retain a density of presynaptic terminals and dendrites that is similar to untreated non-tg/wild-type controls but significantly higher than in untreated gp120 tg animals.121 This finding confirms the hypothesis that the HIV-1 surface glycoprotein is sufficient to initiate excitotoxic neuronal injury and death. It also shows that an antagonist of NMDAR overstimulation can ameliorate HIV-associated neuronal damage in vivo.140

Macrophages and Neuronal Injury in HAD

Macrophages play a pivotal role, although somewhat paradoxical, in the pathobiology of HAD.4, 141, 142 Under steady-state conditions, mononuclear phagocytes, macrophages and microglia act as scavengers and sentinel cells, nonspecifically eliminating foreign material, and secreting trophic factors critical for maintenance of homeostasis within the CNS microenvironment.83, 143, 144, 145, 146 These protective functions, however, can evolve into destructive ones. A number of neurotrophins are secreted by macrophages.97 These factors include, but are not limited to, brain-derived neurotrophic factor (BDNF),147 insulin-like growth factor (IGF)-2,148 β-nerve growth factor (βNGF),149 transforming growth factor beta (TGF-β),150 neurotrophin-3 (NT3)151 and glial-derived neurotrophic factor (GDNF).152 Clearly, a dysregulation of macrophage neurotrophic factors by viral infection and/or immune activation may occur during disease. This dysregulation may be as important as the production of neurotoxins for eliciting neuronal damage. Additionally, some neurotrophic factors are regulated by cytokines. For example, TNF-α (a candidate HIV-1-induced neurotoxin) produced by immune competent microglia can play a neurotrophic role by inducing biologically active TGF-β.150 TGF-β is a protective cytokine for mammalian neurons, particularly in protection against glutamate neurotoxicity, hypoxia and gp120-mediated neural injury.153 This cytokine also affects long-term synaptic facilitation.141

HIV establishes a latent and persistent infection within macrophages.8 The majority of HIV within the CNS appears to be localized within perivascular and blood-derived parenchymal brain macrophages and microglia.8 Astrocytes, oligodendrocytes and brain endothelial cells are rarely infected, if at all.154 As a result of viral infection and resultant immune activation, macrophages produce and release a variety of neurotoxins within the brain.4, 155, 156 These products comprise not only viral proteins, such as gp120,115 gp41157 and Tat,158 but also host cell-encoded products including platelet-activating factor (PAF),159 glutamate,49 arachidonic acid and its metabolites,160 proinflammatory cytokines, such as IL-1β, TNF-α, TNF-related apoptosis-inducing ligand (TRAIL),136, 161 quinolinic acid,48, 162 NTox47 and NO157 among others. In this manner, macrophages, which were once pillars of the immune system, are now responsible for tissue damage, although it is still unclear how macrophages evolve from producing neurotrophins to producing neurotoxins. Perhaps, HIV-1 infection and immune activation induce a transition between neurotrophic and neurotoxic activities. In any case, it seems that activation of p53 in microglia plays a crucial role for neurotoxicity to occur upon exposure of the cells to HIV-1/gp120.82

TRAIL

Recent data suggest that specific subsets of peripherally activated monocytes may preferentially enter the brain and cause disease.9, 163 The neurotoxicity of these subsets may be enhanced not only by changes in functional properties but also by the upregulation of specific cell-surface factors. One such factor is TRAIL: in a model using NOD-SCID mice and HIV-infected human peripheral blood mononuclear cells, it was shown recently that addition of lipopolysaccharide (LPS) causes the infected human cells to infiltrate the murine brain and to cause neuronal apoptosis. This effect was not only specific for macrophage-tropic HIV-1 but was also prevented by a neutralizing anti-TRAIL antibody.164 These findings strongly suggested a role for TRAIL in the induction of neuronal death by infected human macrophages. However, even though TRAIL has been reported to induce apoptosis in brain cells,165 it remains to be shown whether or not killing of neurons occurs as a consequence of a direct or indirect interaction.

TRAIL is a type II integral membrane protein. It is a member of the TNF superfamily and is closely related to FAS ligand.166 TRAIL interacts with at least five unique receptors found on a variety of cell types. TRAIL receptor 1 and 2 (TRAIL-R1 and TRAIL-R2) have death domains and induce cellular apoptosis following ligand binding.167 TRAIL-R3 and TRAIL-R4, however, do not possess these domains and instead act as decoy receptors.168 The fifth soluble TRAIL receptor is osteoprotegerin.169

TRAIL and Apoptosis Signaling

TRAIL receptor-mediated signaling events leading to apoptosis can be divided into two distinct pathways, involving either mitochondria (intrinsic) or death receptors (extrinsic)170 (for review, see Green171). The mitochondrial pathway is initiated through various stress signals that damage mitochondria. Bcl-2 family proteins, including antiapoptotic members, that is, Bcl-2 and Bcl-XL, and proapoptotic members, that is, Bax and Bak, play a critical role in this pathway.171 The BH3-only Bcl-2 family proteins, such as Bid, Bad, Bim and PUMA, serve as sentinels to these stress signals. They are activated through various means, including transcriptional activation, post-translational modification, proteolytic cleavage, etc., during apoptosis.172

In the death receptor (extrinsic) pathway, it has been suggested that during the activation of death receptors DR4 (TRAIL R1) and DR5 (TRAIL R2) by TRAIL, the receptor undergoes oligomerization upon activation, at which time the adapter protein Fas-associated death domain (FADD) is recruited. The receptor–FADD complex then recruits procaspase-8, which together form the death-inducing signaling complex (DISC) where procaspase-8 is activated.173 Depending on the cell type, active caspase-8 can directly lead to the activation of downstream effector caspases, including caspase-3, -6 and -7.174

While the death receptor (extrinsic) pathway and mitochondrial (intrinsic) pathway for apoptosis are capable of operating independently, accumulating evidence suggests that crosstalk between the two pathways exists in cells.171 Recently, Deng et al.170 demonstrated that mitochondrial events are required for TRAIL-mediated apoptosis using human colon cancer cells. They discovered that the reason for this requirement is the presence of negative regulation of the caspase cascade by XIAP, a widely expressed inhibitor of apoptosis protein (IAP) member. Binding of mitochondrially released Smac (also known as DIABLO) to XIAP antagonizes the caspase–XIAP interaction, thereby promoting apoptosis.175 It remains to be shown to what degree these pathways to cell death are operative in the brains of AIDS patients, but we observed in vitro that HIV-1/gp120 activated both the extrinsic and intrinsic pathways to neuronal apoptosis.125

Potential Strategies for Prevention or Therapy of HAD

Presently, an effective pharmacotherapy for HAD is not available. Previous approaches to cope with HAD reflect the challenging complexity inherent in the treatment of patients with AIDS (reviewed by Melton et al.176 and Clifford177). Previous and current therapeutic approaches include various anti-retroviral compounds, alone or in combination, such as Zidovudine, Didanosine, Zalcitabine, and Stavudine. Of these, however, only Zidovudine has been shown to cross the BBB to some extent. Zidovudine has a beneficial effect on HAD but the effect is not long lasting. The other antiretroviral drugs may not penetrate the brain sufficiently to eradicate the virus in the CNS. Thus, an adjunctive treatment besides antiretroviral drugs is needed.

Based on the evolving pathogenesis of HAD described above, several potential therapeutic strategies to attenuate neuronal damage are worth exploring. Among others, agents warranting consideration include NMDAR blockers, cytokines, chemokines, chemokine and cytokine receptor antagonists, p38 MAPK inhibitors, caspase inhibitors and antioxidants (free radical scavengers or other inhibitors of excessive NO or ROS).

NMDAR antagonists have been shown to attenuate neuronal damage due to either HIV-infected macrophages or HIV/gp120, both in vitro and in vivo. The open-channel blocker, memantine, prevents excessive NMDAR activity while sparing physiological function.119, 178 Also, unlike other NMDAR antagonists tested in clinical trials to date, memantine has proven both safe and effective in a number of phase III clinical trials for Alzheimer's disease and vascular dementia. The results of a large, multicenter NIH-sponsored clinical trial using this agent in patients with HAD has suggested some benefit, and improved second-generation drugs are currently under development. Previous, small clinical trials of a voltage-activated calcium channel blocker, nimodipine, and a PAF inhibitor suggested some therapeutic benefit but were not conclusive.16, 179, 180 An additional clinical trial using the antioxidant drug selegiline is aimed at combating the effects of excitotoxicity by minimizing the impact of free radicals.181

Mood changes reaching the level of disorders are one of many problems associated with HIV-1 disease. Sodium valproate (VPA), which functions as a mood stabilizer, might be valuable as a part of the therapeutic armamentarium for HAD. Therapeutic concentrations of VPA (0.6 mM) resulted in (1) significant increases in both nuclear and cytoplasmic β-catenin protein levels; (2) decreases in the level of protein α-kinase C and epsilon isozymes182 and (3) downregulation of myristoylated alanine-rich C-kinase substrate (MARCKS)183 through inositol-independent mechanisms.184 VPA-mediated neuroprotection involves diminished activity of GSK-3β via the inhibition of phosphorylation of β-catenin (Ser33, 37) and tau (Ser202 and Thr181),185 as well as the overall increase in total β-catenin protein levels (Figure 3). Hyperphosphorylation of β-Catenin and tau directly affects neuronal apoptosis and dysfunction.186 β-catenin levels are markedly reduced in some neurodegenerative diseases, and decreased β-catenin signaling seems to increase neuronal vulnerability to apoptosis. Thus, inhibition of GSK-3β may serve to offset the β-catenin destabilization, thereby reducing the vulnerability of affected neurons to apoptosis. In our studies using a model of HIV encephalitis (HIVE) in SCID mice, we found that hyperphosphorylation of β-catenin occurs in the basal ganglia concurrently with gliosis and neuronal degeneration.185 Similarly, specific phosphorylated isoforms of tau have been associated with neurodegenerative disorders, including AD187 and, more recently, also HAD.185, 188 In our model, highly phosphorylated tau at Ser202 and Thr181 is consistently associated with neuronal injury in SCID mice with the neuropathologic features of HIVE. Both tau and β-catenin may represent important physiologic targets of GSK-3β contributing to neuronal loss and neuronal damage in the context of HAD.185 The results support the hypothesis that downstream targets for pathologically activated GSK-3β, including β-catenin and tau, might be a major event in the pathogenesis of HIVE or HAD. Furthermore, our data raise the possibility that VPA inhibits hyperphosphorylation of β-catenin and tau through the regulation of GSK-3β, thus promoting neuronal survival. In connection with the same potentially protective mechanism, lithium has been suggested as a treatment for HAD because it similarly affects the phosphoinositol-3 kinase (PI3K)/Akt (protein kinase B)/GSK-3β pathway.189

Previously, we have shown that the cytokine erythropoietin (EPO) may not only be effective in treating anemia but also in protecting neurons, since it prevents NMDAR-mediated and HIV-1/gp120-induced neuronal death in mixed cerebrocortical cultures.190 Since EPO is already clinically approved for the treatment of anemia, human trials of EPO as a neuroprotectant from HIV-associated dementia may be expedited.191 Additionally, EPO plus IGF-1 act synergistically as neuroprotectants by activating the PI3K/Akt pathway;192 so the use of these two cytokines in conjunction has been advocated for clinical trials.191

Chemokine receptors allow HIV-1 to enter cells and as such are major potential therapeutic targets in the fight against AIDS.75, 193 Antagonists of CXCR4 and CCR5 inhibit HIV-1 entry and are being assessed in clinical trials.75, 193 However, the benefit of inhibitors of chemokine receptors for HIV-associated neurological complications, although likely, remains to be shown.9 Interestingly, as alluded to above, certain chemokines have been shown to protect neurons, even though the virus does not productively infect neurons. In particular, β-chemokines (acting on CCR5 receptors) and fractalkine prevent gp120-induced neuronal apoptosis in vitro,81, 109, 194 and, similarly, some β-chemokines can ameliorate NMDAR-mediated neurotoxicity.194 Additionally, HIV-infected patients with higher CSF concentrations of the β-chemokines MIP-1α/β and RANTES performed better on neuropsychological measures than those with low or undetectable levels.195 These findings support the hypothesis that selected β-chemokines may represent a potential treatment modality for HAD.

Neuronal apoptosis appears to be one of the hallmarks of neurodegenerative diseases including HAD.53 Since caspases carry out the apoptotic program, caspase inhibitors may be helpful in preventing detrimental neuronal loss.196 As detailed above, caspases have been implicated in HIV-related neuronal damage. However, caspase inhibitors are not currently available in a form deliverable to the CNS or targeted to degenerating neurons. With further advances in the caspase field, such drugs may be developed. Care must be exerted to avoid inhibitors that promote oncogenic processes or interrupt physiologic circuits.

Finally, p38 MAPK inhibitors have been shown to reduce or abrogate neuronal apoptosis due to excitotoxicity, HIV/gp120 exposure or α-chemokine (SDF-1) toxicity.81, 197 The pharmaceutical industry is currently developing p38 inhibitors for a variety of inflammatory- and stress-related conditions, such as arthritis, and this may expedite trials for CNS indications such as HAD.

The most recent experimental evidence regarding HAD indicates that synergy between excitatory and inflammatory pathways to neuronal injury and death may, at least in part, be common to other CNS disorders including stroke, spinal cord injury and Alzheimer's disease. It seems likely therefore that the development of new therapeutic strategies for HAD will impact several other neurodegenerative diseases and possibly vice versa.