Opioid and Chemokine Regulation of Cortical Synaptodendritic Damage in HIV-Associated Neurocognitive Disorders

Abstract

Human immunodeficiency virus (HIV)-associated neurocognitive disorders (HAND) persist despite effective antiretroviral therapies (ART). Evidence suggests that modern HAND is driven by subtle synaptodendritic damage in select brain regions, as ART-treated patients do not display overt neuronal death in postmortem brain studies. HAND symptoms are also aggravated by drug abuse, particularly with injection opioids. Opioid use produces region-specific synaptodendritic damage in similar brain regions, suggesting a convergent mechanism that may enhance HAND progression in opioid-using patients. Importantly, studies indicate that synaptodendritic damage and cognitive impairment in HAND may be reversible. Activation of the homeostatic chemokine receptor CXCR4 by its natural ligand CXCL12 positively regulates neuronal survival and dendritic spine density in cortical neurons, reducing functional deficits. However, the molecular mechanisms that underlie CXCR4, as well as opioid-mediated regulation of dendritic spines are not completely defined. Here, we will consolidate studies that describe the region-specific synaptodendritic damage in the cerebral cortex of patients and animal models of HAND, describe the pathways by which opioids may contribute to cortical synaptodendritic damage, and discuss the prospects of using the CXCR4 signaling pathway to identify new approaches to reverse dendritic spine deficits. Additionally, we will discuss novel research questions that have emerged from recent studies of CXCR4 and μ-opioid actions in the cortex. Understanding the pathways that underlie synaptodendritic damage and rescue are necessary for developing novel, effective therapeutics for this growing patient population.

1. Introduction

Human immunodeficiency virus (HIV)-associated neurocognitive disorders (HAND) are a group of neurological disorders resulting from HIV infection of select central nervous system (CNS) cells and chronic low-level CNS inflammation, often in combination with other comorbid factors (Saylor et al., 2016). Progression to the most severe forms of HAND has been sharply decreased by antiretroviral therapies (ART) (Heaton et al., 2010), which transformed HIV infection into a chronic illness by effectively controlling viral replication. However, the overall prevalence of HAND in patients on ART has remained roughly similar to that of the pre-ART era, and HAND prevalence is expected to rise as ART-treated patients age (Tozzi et al., 2007; Heaton et al., 2011; Saylor et al., 2016). The majority of ART-treated patients with HAND present with asymptomatic or mild symptoms, showing deficits in cognitive domains including working memory, executive function, learning, information processing speed, and fine motor coordination (Eggers et al., 2017; Sacktor, 2018). Asymptomatic forms of HAND, which are determined by neurocognitive testing, remain important as they increase the risk of cognitive decline and disease progression in aging patients (Grant et al., 2014). Unfortunately, all clinical trials testing adjuvant therapies for HAND have not shown clinical efficacy (McGuire et al., 2014), leaving HAND as a major therapeutic hurdle in the ART era. Failure of these trials suggests that our understanding of HAND is insufficient, and continued research is necessary to uncover the pathophysiological mechanisms underlying HAND, and to identify novel molecular targets for future drug development.

ART-treated patients with HAND generally do not display overt neuronal death or neurodegeneration (Gelman, 2015). Instead, subtle synaptodendritic injury in specific brain areas more likely underlies the neurocognitive symptoms of HAND in treated patients (Masliah et al., 1997; Everall et al., 1999; Ellis et al., 2007). Although most studies examined brain tissue from patients with poorly controlled infection, advanced neuroimaging studies in virally suppressed patients suggest that brain inflammation and synaptic deficits persist (Harezlak et al., 2011; Ances and Hammoud, 2014; Strain et al., 2017; Sanford et al., 2018). Nevertheless, it is crucial that future studies examine brain tissue from ART-treated patients with well-controlled infection to examine region-specific and subtle synaptodendritic injury. Synaptic injury involves dendritic spines, which are protrusions that often host the postsynaptic compartment of an excitatory synapse (Nimchinsky et al., 2002). Dendritic spines are key mediators of learning and memory processes, and dendritic spine deficits in brain regions including the prefrontal cortex (PFC) often predict impairment in specific cognitive domains (Morrison and Baxter, 2012). In patients with HAND, dendritic spine deficits in critical brain regions may be driven by multiple factors, including chronic neuroinflammation (Spudich, 2016), low-level expression of neurotoxic HIV proteins (Ellis et al., 2007; Hu, 2016; Smith et al., 2018), aging of ART-treated patients (Levine et al., 2016; Cole et al., 2017; Saloner et al., 2019), and even neurotoxic side-effects of some ART (Dalwadi et al., 2018; Stern et al., 2018). Cognitive symptoms are also exacerbated by illicit drug abuse, including injection opioids (Byrd et al., 2011). Sharing injection equipment is a major vector for the spread of HIV, and infected individuals are much more likely than the general population to inject opioid drugs (Degenhardt et al., 2013; Reddon et al., 2019). In line with the hypothesis that HAND manifests due to subtle damage to the dendritic arbor, opioids also aggravate synaptodendritic injury from HIV proteins or inflammatory molecules in vivo and in vitro (Fitting et al., 2010; Fitting et al., 2014). Therefore, understanding how opioid use contributes to synaptodendritic injury in HAND is a necessary step in the process of developing effective therapies for these patients.

Recent work from our lab and others suggests that cortical dendritic spine deficits and cognitive impairment are reversible (Nightingale et al., 2014; Saylor et al., 2016), likely through activation of the chemokine receptor CXCR4 by its natural ligand CXCL12 (Pitcher et al., 2014; Capsoni et al., 2017; Festa et al., 2018). Chemokine signaling is often considered for its roles in the immune system and cellular chemotaxis, but CXCR4 signaling additionally promotes neuroprotective and homeostatic processes in the CNS (Li and Ransohoff, 2008). These processes include survival signaling (Meucci et al., 1998; Khan et al., 2008; Nicolai et al., 2010), neuron-glia communication (Guyon, 2014), modulation of neurotransmission (Guyon and Nahon, 2007), inhibitory synapse formation (Wu et al., 2017), and regulation of cortical dendritic spines (Pitcher et al., 2014). Several studies show that CXCR4 signaling in the CNS is disrupted by HIV infection and μ-opioid agonists, which may partially contribute to synaptodendritic injury (Festa and Meucci, 2012; Festa et al., 2015). Here, we will discuss published and emerging evidence of region-specific synaptodendritic damage in HAND patients and animal models, as well as how opioid and CXCR4 signaling affects dendritic spines in specific cortical regions.

2. Dendritic Spines

Dendritic spines are characterized by their morphology (Figure 1), which reflects the maturity of the spine, and its integration into the local neuronal circuit (Nimchinsky et al., 2002; Berry and Nedivi, 2017). Spines may arise directly from the dendritic shaft (Lai and Ip, 2013) or from long and thin processes called filopodia (Yuste and Bonhoeffer, 2004). Filopodia lack much of the actin cytoskeleton and protein expression needed to become a stable, mature spine, but they may form synapses with other neurons and stabilize over time (Zuo et al., 2005). After a filopodia forms a synapse, it may develop into a thin spine, which is the first mature type of spine. Thin spines are characterized by a long neck region and a small head region with a post-synaptic density containing NMDA receptors and few AMPA receptors (Ganeshina et al., 2004). Although thin spines are more stable than filopodia, they remain somewhat transient and may retract in the absence of neuronal activity or in response to other stimuli (Holtmaat et al., 2005). Thin spines are thought to be involved in short-term cognitive processing and flexibility, as learning tasks often remodel thin spine organization and density in various brain areas (Dumitriu et al., 2010; Motley et al., 2018), and thin spine density correlates with synaptic plasticity and learning, particularly in the aged brain (Hao et al., 2006). A thin spine may further mature into a mushroom spine, which is the most stable type of dendritic spine (Zuo et al., 2005). Mushroom spines have a smaller neck region than a thin spine but have a larger head region and post-synaptic density containing AMPA receptors (Matsuzaki et al., 2001). The long-term stability of mushroom spines suggests that they are involved in long-term cognitive processing, including memory and recall (Bourne and Harris, 2007). Lastly, stubby spines are considered immature spines that lack a neck region (Harris et al., 1992). As the neck of a dendritic spine effectively isolates post-synaptic calcium influx from the dendritic shaft (Noguchi et al., 2005), stubby spines may lack the calcium buffering capacity of other mature spine types. Therefore, any disease state that increases stubby spine density may result in excessive calcium influx into the dendrite, inappropriate activation of dendritic calcium signaling pathways, and potentially excitotoxicity.

Figure 1.

Dendritic spine morphologies

A. Grayscale micrograph of a dendrite from the rat layer 2/3 medial prefrontal cortex that was stained with DiI, a lipophilic fluorescent dye that labels dendritic spines (Seabold et al., 2010). B. The spines on this dendrite can be characterized into different morphological classes using Neurolucida 360 software (Rodriguez et al., 2008): Dendrite (red), filopodia (white), thin spines (cyan), mushroom spines (magenta), and stubby spines (yellow).

3. Region-specific synaptic injury in HAND patients and animal models

Evidence suggests that the PFC, as well as other frontal cortex areas are altered in patients with HAND (Guha et al., 2016; Sanford et al., 2017). The PFC integrates signaling from many different brain areas and plays important roles in cognitive processes including learning and memory, executive function, cognitive flexibility, and selective attention (Euston et al., 2012; Lew et al., 2018). HAND patients may display decreased resting state functional connectivity (Ann et al., 2016; McIntosh et al., 2017), thinning of PFC areas (Thompson et al., 2005), increased expression of interferon and proinflammatory cytokine genes (Sanna et al., 2017), and abnormalities in neurotransmitter systems (Gelman et al., 2012; Buzhdygan et al., 2016). All of these may correlate with synaptodendritic injury in the PFC and HAND progression. Imaging studies of postmortem brain tissue from HIV-infected patients have also shown synaptic injury in other brain regions that underlie learning and memory processes, including the striatum and hippocampus (Sa et al., 2004; Moore et al., 2006), suggesting that specific areas of the brain are more vulnerable to injury during HIV infection. In both HAND patients and animal models of HAND, dendritic spine density in frontal cortex regions is often significantly reduced (Everall et al., 1999; Festa et al., 2015; McLaurin et al., 2019).

Much of the work on understanding how HIV infection contributes to dendritic spine deficits has utilized in vitro or animal models, including rodent models. Among these, the HIV-transgenic rat (HIV-Tg) has recently received increased attention as it recapitulates pathological features of virally suppressed patients (Li et al., 2013; Repunte-Canonigo et al., 2014; Festa et al., 2015; Vigorito et al., 2015). The HIV-Tg rat constitutively expresses all HIV proteins except gag and pol, which are required to produce infectious viral particles (Reid et al., 2001). Therefore, the non-infectious HIV-Tg rat somewhat models the pathology of an ART-treated patient, where viral replication is not detectable. Our group and others have shown that HIV-Tg rats have reduced dendritic spine density in layer 2/3 neurons of the medial PFC (mPFC) (Festa et al., 2015; McLaurin et al., 2019). Spine deficits mostly occur on proximal dendrites and sex differences in learning tasks were reported (McLaurin et al., 2019). Other studies demonstrated similar alterations in specific brain regions and associated behavioral outcomes (Wayman et al., 2016; Casas et al., 2018; Ohene-Nyako et al., 2018). Interestingly, our group has also observed changes in the morphology of mPFC dendritic spines, which underlie cognitive impairment (Festa et al., 2018; Festa et al., manuscript under revision). Increased density of immature spines in HIV-Tg rats suggests that their PFC neurons are more vulnerable to excessive calcium signaling and excitotoxicity. In line with this, controlling calcium flux in HIV-Tg rat mPFC neurons through modulation of NMDA and L-type calcium channels can be neuroprotective (Khodr et al., 2018). Additionally, dendritic spine changes in HIV-Tg rats are correlated with nitrosative stress and increased phosphorylated tau, suggesting that microtubule dysfunction may also affect spine density (Cho et al., 2017).

Since dendritic spine alterations in mPFC neurons are associated with measurable decline of cognitive flexibility, spine deficits in other brain areas associated with specific behaviors may also produce corresponding behavioral impairment. The motor symptoms of HAND have also been reduced by ART (Heaton et al., 2011; DeVaughn et al., 2015), but patients may still display subtle impairment including reduced psychomotor speed, fine motor control (Wilson et al., 2013), and mild extrapyramidal symptoms (Tierney et al., 2018). Dopaminergic dysfunction is likely an aspect of motor impairment in HAND (Gelman et al., 2012), but it is possible that spine deficits in motor areas of the cortex also contribute to motor impairment (Kida and Mitsushima, 2018) or are a consequence of dopamine dysfunction. Indeed, imaging studies in human patients have shown that cortical areas that control motor function have reduced grey matter area and reduced overall tissue volume, suggesting that these regions are altered or damaged during HIV infection (Thompson et al., 2005; Wilson et al., 2015). Our studies show that two distinct animal models of HAND present with dendritic spine deficits in layer 2/3 neurons of the motor cortex (Figure 2). In HIV-Tg rats, motor cortex overall dendritic spine density was decreased, and the percentage of immature filopodia spines was increased (Figure 2A). Sholl analysis of these neurons showed fewer intersections at 100 μM from the soma (Figure 2A), suggesting that subtle dendritic simplification also occurs in the HIV-Tg rat motor cortex. These results were almost mirrored in rats treated with the HIV envelope protein gp120_IIIB by intracerebroventricular injection, which were previously examined in (Festa et al., 2015). Rats exposed to gp120_IIIB showed decreased overall spine density, reduced percentage of thin spines, and increased percentage of filopodia (Figure 2C). Sholl analysis of these neurons showed fewer intersections at 80, 100, and 120 μM from the soma (Figure 2C), again suggesting dendritic simplification. Moreover, HIV-Tg rats did not show any differences in dendritic spine density, morphology, or Sholl analysis in layer 2/3 neurons of the somatosensory cortex (Figure 2B), and gp-120 injected rats only showed a reduced percentage of mushroom spines in this region (Figure 2D). These results provide further evidence that specific areas of the cortex are more vulnerable to HIV-induced synaptic damage, which may underlie specific cognitive or behavioral symptoms.

Figure 2. Synaptodendritic damage in motor and somatosensory cortices in animal models of HAND.

Studies in A and B were completed using adult (4–5 months-old) male F344 HIV-1 transgenic rats and wild-type F344 controls (n=4 rats per group). The preparation of brain tissue, dendritic spine staining (Seabold et al., 2010), and the analysis of layer 2/3 cortical neurons (Festa et al., 2015) were performed as previously described. A. HIV-Tg rat motor cortex. The overall dendritic spine density in motor cortex layer 2/3 neurons of HIV-Tg rats was significantly reduced compared to wild-type controls (t[14] = 4.183, p=0.0009). Dendritic spine morphology analysis showed that the percentage of immature filopodia are significantly increased in these same neurons (filopodia t[434] = 6.094, p<0.0001). Further, Sholl analysis showed the HIV-Tg rats have significantly fewer intersections at 100 μM (p=0.0172) from the soma in this region (distance F[9,100]= 41.98, p<0.0001; group F[1,100]= 7.590, p=0.007). B. HIV-Tg rat somatosensory cortex. There were no significant changes in dendritic spine density (t[14] = 1.044, p=0.3142), morphology, or Sholl analysis (distance F[9,100] = 66.68, p<0.0001; group F[1, 100] = 2.93, p=0.0901) in HIV-Tg rats compared to wild-type.

Studies in C and D were completed using adult (4–5 months-old) male Sprague Dawley rats. Each rat was stereotaxically implanted with a cannula targeting the lateral ventricle, infused with either gp120_IIIB (50 ng/μL in 0.1% BSA, n=5) or vehicle (n=6) once daily for 7 days, and sacrificed 28 days after the final infusion, as previously reported (Festa et al., 2015). The preparation of brain tissue, dendritic spine staining, and analysis of layer 2/3 cortical neurons were also performed as previously reported (Festa et al., 2015). C. gp120-infused rat motor cortex. The overall dendritic spine density in motor cortex layer 2/3 neurons of gp-120 infused rats was significantly decreased (t[20] = 4.896, p<0.0001). Dendritic spine morphology analysis showed that the percentage of thin spines was significantly decreased (thin t[84] = 4.027, p=0.0009), and the percentage of filopodia was significantly increased in these same neurons (filopodia t[84] = 3.856, p=0.0013). Further, Sholl analysis showed that gp-120 infused rats had significantly fewer dendritic intersections at 80 μm (p=0.0006), 100 μm (p=0.0029), and 120 μm (p=0.0197) from the soma in this region (distance F[9, 200] = 23.98, p<0.0001; group F[1, 200] = 50.64, p<0.0001). D. gp120 infused rat somatosensory cortex. There were no significant differences in overall dendritic spine density in layer 2/3 somatosensory cortex neurons of either group (t[20] = 1.046, p=0.3082), but spine morphology analysis showed that the percentage of mushroom spines were significantly decreased in gp-120 infused rats (mushroom t[126] = 4.059, p=0.0006). Further, Sholl analysis did not show any changes of dendritic intersections in this region (distance F[9, 200] = 66.38, p<0.0001; group F[1, 200] = 3.978, p= 0.0475).

Data in all figures represented as mean ± SEM. *p<0.05, **p<0.01, ***p<0.001. Spine density data was analyzed with two-tailed Students t-test; each point in HIV-Tg studies is one dendrite, and each point in gp120 studies is the average spine density of one neuron (two points per animal). Spine morphology data was analyzed with multiple t-tests and Holm Sidak correction. Sholl analysis was analyzed with two-way ANOVA and Sidak’s multiple comparisons test. All animals in these studies were housed in Association for Assessment and Accreditation of Laboratory Animal Care-accredited facilities in accordance with the National Institutes of Health guidelines and institutional approval by the Drexel University Institutional Animal Care and Use Committee.

4. Opioids contributions to cortical synaptodendritic damage

Synaptodendritic injury and dendritic spine deficits in HAND may be aggravated by comorbid use and abuse of opioid drugs. This can result in further progression of HAND symptoms, which was observed in a study of opioid-using HIV patients from the CNS HIV Antiretroviral Therapy Effects Research (CHARTER) group (Byrd et al., 2011). However, the mechanisms by which opioids contribute to synaptic damage in HAND are not clearly defined, and more work is required to elucidate opioids’ actions on individual neurons, as well as on neuronal circuit activity and connectivity. Since the CHARTER study, several groups have identified potential mechanisms by which μ-opioid agonists contribute to synaptic damage of various brain regions using in vivo models, as well as primary culture models. These mechanisms range from direct effects of opioid signaling in specific neuronal populations to downstream effects of indirect opioid signaling in glial cells (Dutta and Roy, 2012). Since the expression of μ-opioid receptors (μORs) varies in different brain regions (Mansour et al., 1994; Ruzicka et al., 1995), μ-opioid agonists may also potentiate the region-specific synaptodendritic damage observed in HAND. This section will highlight recent studies of the functions of μORs in cortical tissue and how these findings may relate to HAND pathology.

Expression of μORs in cortical neurons may differ depending on the cortical region and species examined. Studies of human PFC tissue suggest that μOR transcripts and protein are expressed in pyramidal neurons and non-pyramidal neurons in layers 2–5 (Peckys and Landwehrmeyer, 1999; Schmidt et al., 2003), although neuronal subtype distinctions were made by morphological examination and not by immunostaining against common neuronal markers. However, studies from rat cortex suggest that μORs are mainly expressed in GABAergic inhibitory neurons, as two separate groups did not observe μOR staining or transcript expression in excitatory neurons (Taki et al., 2000; Ferezou et al., 2007). Another study from a different group suggested that μORs are expressed in excitatory synapses of mouse cortical neurons and rat hippocampal neurons, as μOR staining co-localized with NMDA and AMPA receptor staining (Liao et al., 2005). Therefore, GABAergic interneurons may be the major target of μ-opioids in the rat cortex, but excitatory neurons may also contribute to opioid signaling pathways if μORs are expressed on excitatory synapses. The different results of these studies may be due to several factors, including differences between primary cultured neurons and cortical tissue, antibody quality and specificity between species, examination of different cortical regions, and incomplete identification of the multitude of μOR splice variants (Pasternak, 2014; Xu et al., 2017). Of note, μORs are also expressed by rodent cortical astrocytes (El-Hage et al., 2005; Burbassi et al., 2010) and microglia (Maduna et al., 2018), suggesting that μ-opioids have the capacity to modulate several cortical cell types. Furthermore, astrocyte μOR expression varies depending on cortical area or brain region examined (Ruzicka et al., 1995).

Just as HIV infection may produce brain region-specific synaptodendritic damage, evidence suggests that morphine, a prototypical μ-opioid agonist, also affects selected brain regions. One group found that daily, long-term morphine self-administration or experimenter-administration in rats reduced dendritic spine density in various brain regions out to at least one month after morphine cessation (Robinson et al., 2002). Both treatment paradigms reduced spine density in the mPFC, the primary somatosensory cortex, and the visual cortex, while only self-administration reduced spine density in the hippocampal CA1 and dentate gyrus. Notably, self-administration produced more pronounced spine deficits in all examined regions except for the somatosensory cortex, which showed less pronounced spine deficits (Robinson et al., 2002). Morphine can also have long-term effects on dendritic spine density in primary neurons. Using mouse cortical and rat hippocampal neurons, one group showed that morphine activation of μORs destabilized existing spines, resulting in reduced dendritic spine density (Liao et al., 2005). This occurred after long-term morphine treatment over 3–6 days, was rescued by the μOR antagonist CTOP, and was not observed in μOR knockout animals. Interestingly, blocking neuronal activity with tetrodotoxin did not prevent morphine’s ability to reduce spine numbers, suggesting that μOR signaling at the cellular level regulates dendritic spine stability (Liao et al., 2005). Our group has shown similar results in vivo and in vitro, where morphine caused a large reduction of dendritic spine density in primary rat cortical neurons after 24 hour exposure, and morphine similarly reduced dendritic spine density in layer 2/3 PFC neurons of rats implanted with morphine pellets for 96 hours (Pitcher et al., 2014; Nash et al., 2019).

5. μOR-dependent pathways that regulate dendritic spine density

As dendritic spines are activity-dependent structures, one may expect that the well documented activity-dampening effects of μ-opioid agonists are involved in the reduction of dendritic spines. However, recent research suggests that several interesting and perhaps unexpected pathways downstream of μORs in cortical neurons modulate dendritic spines. These recently discovered pathways, which will be described in the following section, may function in combination with altered neuronal activity to modulate dendritic spines in the cortex. Additionally, morphine’s actions on dendritic spines are reported to be enhanced in the presence of HIV proteins, suggesting a potential synergy of HIV infection and opioid use (Fitting et al., 2014). Below, we discuss recent work describing studies that illuminate morphine, or μOR-mediated pathways involved in modulating cortical dendritic spine density and stability.

5.1. Molecular mediators

A group of studies has demonstrated that morphine and other μ-opioid agonists modulate dendritic spines through aberrant calcium signaling. Although these studies are mostly in hippocampal neurons, the pathways that they identify may also apply to cortical neurons and therefore should be carefully considered. One study demonstrated that prolonged morphine exposure decreased the density and stability of hippocampal dendritic spines, which was dependent on the activity of Ca²⁺/calmodulin-dependent protein kinase II α (CaMKIIα) and calcineurin (Miller et al., 2012). Blocking the morphine-mediated activation of calcineurin preserved AMPA receptor expression in dendritic spines and AMPA receptor-mediated miniature excitatory postsynaptic currents (Miller et al., 2012). As AMPA receptors are highly expressed in mature spines (Matsuzaki et al., 2001), the loss of these receptors suggests that mature spines are affected by prolonged morphine use. Furthermore, long-term morphine exposure inhibited CaMKIIα activity and caused its translocation out of dendritic spines. This was associated with dendritic spine loss, as CaMKIIα could no longer activate synaptically localized Rac1, a small guanosine triphosphatase and regulator of dendritic spine structure (Xie et al., 2007). Further, neurons transfected with a dominant negative Rac1, as well as a constitutively active Rac1, blocked morphine’s ability to reduce dendritic spine density (Miller et al., 2012), highlighting the importance of Rac1 signaling to dendritic spine homeostasis.

Morphine also activates an additional signaling pathway that reduces cortical dendritic spine density that may not be directly associated with dampened neuronal activity. Our group has shown that cortical neurons exposed to morphine for 6–24 hours upregulated ferritin heavy chain (FHC), a subunit of the iron storage protein ferritin (Sengupta et al., 2009). Interestingly, FHC inhibits activation of the homeostatic chemokine receptor CXCR4 by its natural ligand CXCL12, as well as activation of its downstream effectors including ERK1/2 and Akt (Li et al., 2006; Sengupta et al., 2009; Aversa et al., 2017). As CXCL12/CXCR4 signaling increases dendritic spine density in primary cortical neurons, spine deficits occur when this pathway is blocked by morphine-mediated upregulation of FHC (Sengupta et al., 2009; Pitcher et al., 2014). FHC was previously known for its role in iron storage and oxidation (Arosio et al., 2015), but the novel interaction with the CXCR4 signaling complex suggests that the functional roles of FHC may not be restricted to cellular iron metabolism. Our mechanistic studies showed that the μOR antagonist CTAP and FHC knockdown both prevented morphine mediated CXCR4 inhibition (Sengupta et al., 2009), demonstrating morphine was not acting through other opioid receptors or toll-like receptor 4. Additionally, neuronal FHC is upregulated by interleukin-1β (IL-1β) secreted from glial cells exposed to HIV proteins such as gp120 (Festa et al., 2015), demonstrating morphine and HIV neurotoxins may inhibit CXCR4 signaling through a convergent mechanism. Furthermore, our evidence suggests that these pathways are present in vivo, as clinical samples of HAND patients with and without poly-drug use (including opioids), as well as simian immunodeficiency virus infected macaques treated with morphine, showed increased expression of FHC and reduced phosphorylation of the CXCR4 c-terminus in PFC neurons (Pitcher et al., 2014). Strikingly, neuronal FHC expression levels correlated with the severity of HAND symptoms in human patients (Pitcher et al., 2014), suggesting that morphine regulation of CXCR4 contributed to cognitive deficits via a reduction of cortical dendritic spine density.

Interestingly, we have shown that morphine post-transcriptionally regulates FHC levels in cortical neurons by modulating neuronal iron metabolism. Briefly, morphine causes iron efflux from endolysosomes to the cytoplasm prior to FHC upregulation, and chelation of endolysosomal iron with deferoxamine blocks morphine mediated FHC upregulation (Nash et al., 2019). Chelation of extracellular iron did not prevent morphine induced FHC upregulation, suggesting that morphine only modulates intracellular iron stores to upregulate FHC. The mechanism by which cytoplasmic labile iron upregulates neuronal FHC could involve the post-transcriptional iron regulatory protein (IRP) system (Wilkinson and Pantopoulos, 2014). The IRP system controls translation of mRNAs involved in iron metabolism based on free iron levels. On 5’ untranslated regions of FHC transcripts, IRPs tightly bind to iron response elements (IREs) in conditions of low labile iron and prevent ribosomal translation. When labile iron levels increase, free iron interacts with IRPs and releases them from IREs on the FHC transcript, resulting in translation of FHC. If this pathway is activated, it may also impact the expression of other proteins that are regulated by iron.

5.2. Neuronal networks

Cortical γ-aminobutyric acid (GABA)ergic interneurons appear as major targets of μ-opioid agonists, but it is unclear if μOR activation specifically on interneurons contributes to dendritic spine deficits on local excitatory neurons. However, recent studies show that μOR signaling in the PFC dysregulates circuits involved in motivational behaviors, which may result from a network-level effect mediated by altered GABAergic transmission (Baldo, 2016). Therefore, it is possible that the actions of μ-opioid agonists on cortical interneurons may contribute to HAND through network dysregulation in addition to synaptodendritic damage (Figure 3). As morphine inhibits the release of GABA (Qu et al., 2015; Yokota et al., 2016) this can lead to an increased activity of excitatory neurons driving spine compensatory changes or sublethal damage. GABAergic tone may also be affected by μOR activation on interneurons that upregulates FHC and inhibits CXCR4 function in the same cell. GABAergic interneuron CXCR4 signaling plays important roles in regulation of inhibitory neurotransmission and synapse formation in several regions of the brain. In the mouse mPFC, CXCR4 is expressed on inhibitory synapses from layer 5 parvalbumin interneurons, which form inhibitory synapses onto layer 5 pyramidal neurons expressing CXCL12 on their cell bodies (Wu et al., 2017). Additionally, CXCR4 signaling on interneurons from distinct subcortical areas promotes synaptic release of GABA (Guyon et al., 2006; Heinisch and Kirby, 2010). If CXCR4 signaling also promotes GABA release from cortical interneurons, then μ-opioid agonist induced inhibition of CXCR4 in these cells may disrupt GABAergic transmission in addition to inhibitory synapse formation in the cortex. Furthermore, several GABA_B agonists and antagonists, including GABA itself, act allosterically on CXCR4 and alter normal responses to CXCL12 signaling (Guyon et al., 2013), and this also may be disrupted by μ-opioids.

Figure 3. Potential network-level interactions of μOR and CXCR4 leading to dendritic spine deficits.

Although μOR activation could lead to CXCR4 inhibition and dendritic spine deficits all in the same excitatory pyramidal neuron (PN), μOR regulation of CXCR4 in GABAergic interneurons may also regulate dendritic spine density in PNs by altering the excitatory/inhibitory balance of the local circuit through distinct mechanisms. Some possibilities include A. Reduced GABAergic innervation of PNs; B. an overall decrease in neuronal activity of GABAergic interneurons and PNs; or C. Reduced activity of GABAergic disinhibition circuits.

Although cortical interneurons only make up roughly 15% of the total cortical neuron population in rats (Meyer et al., 2011), they are organized as subpopulations that perform diverse and specialized functions (Tremblay et al., 2016; Lim et al., 2018). These subpopulations are reported to express μORs at different levels and may therefore be more or less susceptible to exogenous μ-opioid agonist exposure. One study shows that cortical interneurons expressing vasoactive intestinal peptide (VIP), corticotropin releasing factor, and choline acetyltransferase showed the highest μOR expression, while inhibitory neurons expressing parvalbumin (PV), somatostatin (SST) and neuropeptide Y (NPY) displayed lower expression levels (Taki et al., 2000). In another study, μOR transcripts were mostly found in cortical interneurons that expressed VIP, cholecystokinin (CKK), and calretinin (CR), in addition to layer I interneurons that expressed NPY (Ferezou et al., 2007). As these studies suggest that all the major cortical interneuron subpopulations express μORs to some extent, μ-opioid agonists may produce a broad suppression of GABAergic tone in the cortex. Alternatively, μ-opioid agonists could dose-dependently inhibit GABA release from interneuron subpopulations with higher μOR expression. VIP expressing interneurons showed high expression of μOR transcript and protein in both studies, suggesting that these neurons are major targets of μ-opioid agonists. VIP interneurons are involved in disinhibition circuits, as they form synapses on nearby interneurons that innervate excitatory neurons (Pi et al., 2013; Karnani et al., 2016). Therefore, inhibiting VIP interneurons with μ-opioid agonists may result in increased GABAergic tone on local excitatory neurons. However, μ-opioid agonist induced inhibition of PV, SST, CCK, and other interneurons that directly synapse onto local pyramidal neurons may decrease overall GABAergic tone in the cortex. In line with this, one study showed that the specific μOR agonist DAMGO inhibited sodium currents in non-pyramidal PFC neurons, and had no effect on sodium currents in pyramidal excitatory neurons (Witkowski and Szulczyk, 2006). This reduction of GABAergic tone may contribute to excitotoxicity and synaptodendritic damage, especially in patients with HAND that may be more vulnerable to μ-opioid agonist induced synaptic damage (Byrd et al., 2011). Several recent studies have examined the contributions of interneuron subpopulations to cognitive function, demonstrating the importance of these neurons in learning and memory processes (Batista-Brito et al., 2017; Ognjanovski et al., 2017; Eavri et al., 2018; Lucas and Clem, 2018). As the consequences of CXCR4 signaling vary between different brain areas and types of neurons, region-specific and focused studies are necessary to determine how μ-opioid agonists regulate GABAergic transmission, if μ-opioid agonist induced inhibition of CXCR4 signaling is involved, and if the actions of μ-opioid agonists on interneurons contribute to cognitive impairment in HAND.

6. CXCL12/CXCR4 regulation of cortical dendritic spines and synapses

Although chemokine receptor signaling is typically associated with functional responses and chemotaxis of immune cells, CXCL12/CXCR4 signaling has additional functions in the CNS, as reviewed in (Nash and Meucci, 2014). These include homeostatic and neuromodulatory functions, including regulation of axonal pathfinding and outgrowth (Ohshima et al., 2008), neuronal branching (Pujol et al., 2005), neurogenesis and neuronal precursor differentiation (Peng et al., 2007; Hattori and Miyata, 2018), neuron-glia communication (Guyon, 2014), and neuronal excitability (Guyon and Nahon, 2007). In addition to these functions, our group has shown that CXCL12/CXCR4 signaling increases dendritic spine density of cortical neurons in vitro and in vivo (Pitcher et al., 2014). In cultured cortical neurons, CXCL12 treatment for 1 hour increased spine density in the absence of glia, suggesting that direct activation of CXCR4 on cortical neurons regulates dendritic spines.

Several reports have shown that CXCR4 and its natural ligand CXCL12 are expressed in various cells of the cortex during development and adulthood. CXCR4 is expressed in primary cortical neurons from several species and coupled to functional outcomes, including calcium signaling and activation of cellular kinases (Meucci et al., 1998; Bajetto et al., 1999; Klein et al., 1999). In the adult brain, CXCR4 and CXCL12 are constitutively expressed in several regions, including the cerebral cortex and hippocampus (Meucci et al., 1998; Banisadr et al., 2002; Pitcher et al., 2014). Although some information regarding CXCR4 expression on neuronal subpopulations has been reported (Banisadr et al., 2002; Wu et al., 2017), a complete characterization of CXCR4 expression on subpopulations of cortical neurons has not been published. ACKR3, another chemokine receptor that binds CXCL12, is also expressed in adult neurons of the cortex (Shimizu et al., 2011; Banisadr et al., 2016)

Our previous studies suggest that CXCL12/CXCR4 signaling increases spine density in vitro and in rat layer 2/3 mPFC neurons (Pitcher et al., 2014; Festa et al., 2015), but CXCL12’s effects on dendritic spines are not limited to mPFC layer 2/3 neurons. We also observed similar trends in rat layer 5 mPFC neurons, including increased overall dendritic spine density and a specific increase in thin spine density in the prelimbic cortex (unpublished observations). This finding is of added importance because layer 5 neuronal connectivity between prelimbic and infralimbic cortices of rat mPFC primes the brain to set up new associations, which are critical for set-shifting tasks and general working memory (Mukherjee and Caroni, 2018). Reciprocal connectivity between layer 5 neurons of the prelimbic and infralimbic cortices is also involved in similar adaptive behaviors, including conditioned place preference (Pena-Bravo et al., 2017) and extinction of fear memory (Marek et al., 2018). Fear memory extinction also increases layer 5 dendritic spine density (Lai et al., 2012). Generally, layer 5 prelimbic cortex activity is involved in applying previously learned associations (Hok et al., 2005), while layer 5 infralimbic cortex activity is involved in setting up alternative associations (Suto et al., 2016). Therefore, it is possible that dendritic alterations in these sub-regions, which may reduce their activity and connectivity, may drive cognitive deficits in working memory tasks. Furthermore, a CXCL12-mediated rescue of thin dendritic spines in layer 5 neurons of these regions in HIV-Tg rats may underlie improved performance in the extradimensional shift portion of set shifting studies (Festa et al., 2015; Festa et al., 2018).

CXCL12 also regulates inhibitory synapse formation on layer 5 pyramidal neurons of the mPFC. A report demonstrated that CXCL12 is expressed on the cell bodies of mouse layer 5 PFC neurons and promotes somatic synapse formation from PV interneurons, as PV synapses express CXCR4 and ACKR3 (Wu et al., 2017). Conditional knockout of CXCL12 from layer 5 mPFC neurons in this system resulted in reduced PV interneuron terminal innervation and reduced inhibitory post-synaptic currents on these neurons. This could be due to reduced Rac1 activation in PV interneurons, as another group showed that Rac1 conditional knockout in mouse hippocampal neurons reduced spontaneous inhibitory postsynaptic currents and PV terminals in the CA1 pyramidal layer (Pennucci et al., 2016). Taken together, the evidence suggests that CXCL12/CXCR4 signaling promotes inhibition of layer 5 pyramidal neurons of the mPFC while simultaneously increasing dendritic spine density in this area. Therefore, CXCL12 may also improve cognitive function through regulation of the excitatory/inhibitory balance of mPFC neurons (Ferguson and Gao, 2018).

7. CXCR4 signaling pathways that regulate dendritic spine density

As cognitive impairment and dendritic spine deficits in HAND may be reversible through CXCL12/CXCR4 signaling, it is critical to understand the underlying signaling pathways that produce these outcomes (Figure 4). Dendritic spine homeostasis is influenced by neuronal activity, actin stabilization, and expression of synaptic proteins (Bertling and Hotulainen, 2017), and evidence suggests that all of these processes are regulated, at least in part, by the CXCL12/CXCR4 signaling axis. Although more studies are needed to fully elucidate the signaling pathways downstream of CXCR4 that regulate dendritic spines, the most current knowledge is discussed in this section. The pathways discussed here may also be therapeutically useful for neurological disorders beyond HAND, as dendritic spine deficits drive cognitive impairment in Alzheimer’s disease, Parkinson’s disease, Huntington’s disease, and other dementias (Herms and Dorostkar, 2016), as well as schizophrenia (Moyer et al., 2015).

Figure 4.

μOR regulation of CXCR4 working model

Morphine or other μ-opioids binding to μORs on cortical neurons activates a Gαi-protein pathway that leads to de-acidification of endolysosomes, and efflux of endolysosomal iron to the cytoplasm. This increases labile iron levels in the neuron, resulting in increased production of the iron storage proteins ferritin heavy chain (FHC) and ferritin light chain (FLC) through a post-transcriptional mechanism. In addition to its iron storage functions, FHC also associates with the homeostatic chemokine receptor CXCR4 and blocks its downstream signaling. Normally, CXCR4 activation by its natural ligand CXCL12 increases dendritic spine density in cortical neurons, which may occur through activation of several pathways. First, CXCR4 activation leads to HDAC4 translocation to the nucleus, which regulates several genes involved in synaptic activity. Specific knockdown of HDAC4 prevents CXCL12 from increasing dendritic spine density, suggesting that CXCL12/CXCR4 signaling regulates synaptic protein expression. Second, CXCL12 may activate the Rac1/PAK pathway in cortical neurons, resulting in phosphorylation of the actin severing protein cofilin. Phosphorylated cofilin is unable to break down the actin cytoskeleton, which may result in stabilization of actin-rich dendritic spines.

7.1. Molecular mediators

The Rac1/PAK pathway regulates the stability of actin, which is the major cytoskeletal component of dendritic spines (Hotulainen and Hoogenraad, 2010; Basu and Lamprecht, 2018). Rac1 is activated by CXCR4 signaling (Lopez-Haber et al., 2016; Mills et al., 2018; Bajanca et al., 2019), which begins a phosphorylation cascade of second messengers, including p21-activated kinase (PAK), LIM kinase 1 (LIMK1), and the actin severing protein cofilin (Delorme et al., 2007). Phosphorylation of cofilin on S3 blocks its ability to disassemble actin filaments (Arber et al., 1998; Yang et al., 1998), which may preserve the structural integrity of dendritic spines. If Rac1 or any of its downstream mediators emerge as suitable drug targets for HAND, modulation of these proteins should be careful and deliberate. Two reports show that over-activation of Rac1 impaired neuritogenesis and synaptic maturation in cortical neurons, which contributed to cognitive impairment (Pyronneau et al., 2017; Zamboni et al., 2018). Therefore, a balance between activation and inhibition of the Rac1 pathway may be necessary to achieve an optimal therapeutic effect.

Dendritic spine maturation depends on synaptic activity and protein expression, which may both be regulated by CXCL12/CXCR4 signaling. Our studies in cultured cortical neurons showed that CXCL12 promoted synaptic activation of NMDA receptors while inhibiting extrasynaptic NMDA activation (Nicolai et al., 2010). This process occurred through a CXCR4 mediated repression of the NMDA NR2B subunit, which is highly expressed in extrasynaptic NMDA receptors (Tovar and Westbrook, 2002; Miwa et al., 2008; Yang et al., 2017). Our work suggests that repression of NR2B may be due to translocation of the class IIa histone deacetylase 4 (HDAC4) to the nucleus (Nicolai et al., 2010; Pitcher et al., 2014). HDAC4 alone has little to no catalytic activity (Lahm et al., 2007) but binds select transcription factors, including MEF2, and recruits other HDACs that deacetylate histones and repress gene expression (Wu et al., 2016). In neurons, HDAC4 translocates between the cytoplasm and nucleus in response to various stimuli and is hypothesized to have specific functions in each subcellular compartment (Fitzsimons, 2015). After nuclear translocation, HDAC4 suppresses a set of genes that is involved in synaptic activation (Sando et al., 2012). A mutated HDAC4 (S3A) with constitutive nuclear translocation severely reduces dendritic branching and complexity while preserving dendritic spines of hippocampal neurons (Litke et al., 2018). CXCL12 induces nuclear translocation of HDAC4 for at least 1 hour after treatment (Pitcher et al., 2014), but beyond this acute period, HDAC nuclear localization is unclear. It is possible that this pathway is a transient, neuroprotective response that represses synaptic activity in response to excitotoxicity. Additionally, HDAC4-mediated repression of MEF2 may enhance dendritic spine density and memory formation, as MEF2 restricts dendritic spine growth in primary hippocampal neurons and disrupts learning and memory processes (Cole et al., 2012). In line with this study, HDAC4 knockdown also produces memory deficits, suggesting that its role in the nuclear compartment is neuroprotective under certain circumstances and/or that it has additional memory-promoting functions in the cytoplasm (Kim et al., 2012). HDAC4 phosphorylation and nuclear export depends on neuronal activity and calcium influx, and cytoplasmic HDAC4 is expressed in some dendritic spines (Darcy et al., 2010). Cytoplasmic HDAC4 could contribute to memory formation and stabilization of dendritic spines through small ubiquitin-like modifier (SUMO)ylation of synaptic proteins (Craig and Henley, 2012). Several studies have reported the involvement of HDAC4 in SUMOylation of the transcription factor MEF2 (Gregoire and Yang, 2005; Zhao et al., 2005), suggesting that synaptic proteins may also be targeted. Synaptically localized HDAC4 may SUMOylate Rac1, which is typically required for Rac1 activation and proper cell motility (Castillo-Lluva et al., 2010). As CXCR4 increases dendritic spine density, HDAC4 may also be involved in Rac1 activation.

7.2. Neuronal networks

CXCR4 signaling may also produce circuit-level effects that alter cortical dendritic spine density by differential modulation of inhibitory and excitatory neurons. For example, CXCR4 activation on nearby interneurons may enhance GABA release, as reported in non-cortical areas (Heinisch and Kirby, 2010; Guyon et al., 2013), acutely dampening excitatory neuronal activity. At the same time, CXCR4 activation of the Rac1/PAK/cofilin pathway on excitatory neurons may stabilize transient dendritic spines that would normally retract in response to reduced neuronal activity. Interestingly, both of these pathways may also be inhibited by GABA spillover from the synapse, as GABA and other compounds that interact with GABA receptors are reported to be negative allosteric modulators of CXCR4 (Guyon et al., 2013). For interneurons, GABA spillover may begin a feedback mechanism that prevents further GABA release by inhibiting interneuron CXCR4 signaling. If this is the case, then CXCR4 activation status may follow the oscillatory patterns that are driven by cortical interneuron activity (Cardin, 2018). On excitatory neurons, CXCR4 signaling may stabilize dendritic spines and counteract neuronal activity-dependent spine retraction. During periods of increased neuronal activity, CXCR4 stabilized dendritic spines may then integrate into the synaptic circuit and mature. The long-term dysregulation of this hypothesized feedback mechanism by opioids or HIV neurotoxins may also explain some of the reported neurocognitive symptoms in HAND with opioid use. In HIV infected patients, the continued presence of even low levels of X4 gp120 may dysregulate this normal feedback between CXCR4 and GABA systems through altered CXCR4 activation and membrane trafficking. Indeed, one report shows that gp120 increases GABAergic currents in hippocampal slices by modulating α5-containing GABA_A receptors (Green and Thayer, 2019). The same group also reported that IL-1β release in response to gp120 enhances the same tonic GABA current, suggesting that sustained low-level inflammation may also enhance GABA release through this pathway. As removal of microglia from the culture blocked these effects, it is possible that the chronic inflammatory environment is a pre-requisite for gp120-induced dysregulation of CXCR4 signals, dendritic spine deficits, and cognitive deficits. Furthermore, inhibition of CXCR4 by μ-opioid agonists is likely to worsen dendritic spine deficits and potentially dysregulate the functions and/or inhibitory oscillations of cortical interneurons. For excitatory neurons, morphine inhibits the CXCR4-mediated stabilization of transient dendritic spines, suggesting that fewer dendritic spines will be preserved during periods of enhanced GABAergic tone. However, morphine-mediated inhibition of CXCR4 on inhibitory neurons may also disrupt their activity and subsequent GABA neurotransmission. Therefore, the pathological outcomes of μ-opioid agonist use are likely affected by many factors, including the doses and duration of use, receptor expression on neuronal subpopulations, and the stage of impairment in a patient with HAND.

8. Conclusions and future research

HIV infection and opioid use may converge to produce sub-lethal synaptic injury and cognitive impairment in HAND patients, and studies that exploit the dendritic spine promoting CXCL12/CXCR4 signaling axis show that this synaptic injury may be reversible. As reversing dendritic spine deficits in specific brain regions, including the PFC, results in improved cognitive performance in HIV-Tg rats, novel therapeutics that protect against the reduction of dendritic spines or restore spines that have been lost may lead to functional recovery of patients with HAND, as well as other neurocognitive disorders. However, direct pharmacological targeting of CXCR4 may not be a realistic strategy due to the widespread expression and functions of the receptor. Therefore, future research must aim to more fully understand the signaling pathways downstream of CXCR4 that are involved in the regulation of dendritic spines and how these pathways vary depending on neuronal sub-type and brain region. Furthermore, additional research must dissect the mechanism of CXCR4 regulation by μ-opioid receptors in a cell-type and region-specific manner because preventing this pathway may be another approach to enhance CXCR4 signaling and rescue dendritic spine deficits in opioid using HAND patients. In order to identify new drug targets and develop effective therapeutics, future work must identify the mechanisms driving synaptic injury and whether they exist as complex cellular signaling cascades or as region-specific dysregulation of neuronal network activity.

• Region-specific dendritic spine deficits may drive impairment in HAND

• CXCL12/CXCR4 signaling stabilizes dendritic spines

• μ-opioid use contributes to spine loss in various ways, including CXCR4 inhibition

Abbreviations

HAND

HIV-associated neurocognitive disorders

ART

antiretroviral therapy

μOR

μ opioid receptor

PFC

prefrontal cortex

HIV-Tg

HIV-transgenic rat

FHC

ferritin heavy chain

FLC

ferritin light chain

IRP

iron regulatory protein

IRE

iron responsive element

VIP

vasoactive intestinal peptide

PV

parvalbumin

HDAC4

histone deacetylase 4

MEF2

myocyte enhancer factor-2