Introduction

Pharmaceutical advances stemming from immunologic and genetic research have greatly improved and extended the lives of people living with HIV-1 (PLWH). Despite this, the prevalence of neurocognitive deficits in PLWH, collectively termed HIV-associated neurocognitive disorders (HAND), remains high (Antinori et al., 2007; Heaton et al., 2011). Prior to the widespread use of combination antiretroviral therapy (cART), neurocognitive syndromes due to HIV-1 infection were often severe and largely considered the manifestation of HIV encephalitis (McArthur et al., 1993; Moore et al., 2006; Glass et al., 1995; Bell et al., 1998; Persidsky & Gendelman, 2003; Everall et al., 2005; Letendre et al., 2011; Boven et al., 2000; Conant et al., 1998; Eugenin et al., 2006; Kraft-Terry et al., 2009). However, HAND currently presents with milder symptoms in the vast majority of cART-treated cases (McArthur et al., 2005; Heaton et al., n.d.) and is not typically associated with HIV encephalitis (Everall et al., 2009). Instead, the neuropathogenesis of HAND is now believed to be largely the result of neurodegeneration driven by chronic neuroinflammation (Glass et al., 1995; Persidsky & Gendelman, 2003; Kraft-Terry et al., 2009; Everall et al., 2009).

Candidate gene studies have identified functional variants, largely within immune-related genes, that modify risk for HAND (as reviewed in (Levine et al., 2014a; Kallianpur & Levine, 2014)). Such findings make sense biologically, supporting the role of inflammation in HAND pathogenesis. For example, some chemokines affect neuronal signaling with subsequent disturbance of glial and neuronal functions (Zheng et al., 2001), while others serve to block the HIV-1 co-receptor, thus mitigating HIV-1 replication (Lane et al., 2003) and slowing disease progression (Gonzalez et al., 2001; Gonzalez et al., 2005), as well as reducing macrophage activation and chemotaxis of monocytes and other cells into the brain (Kaul & Lipton, 2005; Weiss et al., 1999). In addition to immune-related genes, candidate gene studies have implicated dopaminergic dysregulation (Levine et al., 2014a; Kumar et al., 2009; Kumar et al., 2011; Levine et al., 2012; Levine et al., 2014b), variation in mitochondrial function (Samuels et al., 2016; Hulgan et al., 2015), and cellular lipid and cholesterol transport in HAND pathogenesis, as recently reviewed in (Geffin & McCarthy, 2018). However, very few candidate gene markers have been replicated in subsequent studies by independent groups. One likely reason for this is that the vast majority of these studies utilized behavioral phenotypes (e.g., HAND diagnosis or neurocognitive functioning), with little consistency between studies. Due to the inherent limitations of neurocognitive assessment (e.g., measurement error due to tests’ psychometric properties and engagement by the examinee), coupled with the poor inter-rater agreement of what distinguishes HAND from other causes of neurocognitive impairment (Woods et al., 2004), a more fruitful strategy may be to focus on histopathological phenotypes.

Putative immunohistochemical markers of the neuropathological changes underlying HAND include synaptophysin (SYP) and microtubule-associated protein-2 (MAP 2) (Moore et al., 2006), abnormal protein aggregates such as β-amyloid (Achim et al., 2009; Rempel & Pulliam, 2005; Green et al., 2005; Esiri et al., 1998; Soontornniyomkij et al., 2012), and markers of microglial/macrophage activation, astroglial activation, and dysregulated cytokine expression (Glass et al., 1995; Bell et al., 1998; Persidsky & Gendelman, 2003; Everall et al., 2005; Letendre et al., 2011; Boven et al., 2000; Conant et al., 1998; Eugenin et al., 2006; Kraft-Terry et al., 2009). If previously identified genetic variants modify risk for HAND, it logically follows that those variants also modify the cellular and pathophysiological pathways that underlie HAND. Bridging the informational gap between genotypes and behavioral phenotypes in the context of HAND may provide important insights about pathogenesis. We recently reported results of an ambitious study that bridged genetic, histopathological, virologic, and neurocognitive data within subjects in order to understand which histopathological features and genetic variants were relevant to HAND pathogenesis (Levine et al., 2015). That study identified several genetic susceptibility loci that influenced histopathology and other disease parameters. Most notably, neurocognitive functioning was strongly correlated with levels of MAP 2 and SYP in the frontal cortex, both of which declined as plasma HIV-1 RNA viral load increased. This underscores the widely reported process by which HIV-1 replication-related events lead to synaptodendritic degeneration. Furthermore, an inverse relationship between SYP expression and β-amyloid plaque burden in frontal cortex suggests that HIV-1 replication in the brain may be a driver of the histopathological changes or, alternatively, an initiator of a causal chain of events involving neuroinflammation (reflected by ionized calcium-binding adapter molecule-1 (Iba1)) and dysfunctional protein clearance (reflected by β-amyloid plaque deposition). Downstream to these changes is synaptodendritic degeneration which is the immediate histopathological substrate of HAND, although several factors likely modify this cascade (Guha et al., 2018; Cantres-Rosario et al., 2019; Santerre et al., 2019; Desplats et al., 2013) and the ultimate manifestation of HAND (Geffin & McCarthy, 2018; Saloner et al., 2019; Yu et al., 2019; Tovar-y-Romo et al., 2012; Gannon et al., 2017).

In the current study, we expanded upon the previous findings (Levine et al., 2015). In the previous paper, we were unable to examine how demographic factors, HIV-1 disease variables, and host genotypes predicted MAP 2 and SYP levels due to the low number of cases. In the current study, we first revisited this relationship by examining data from a larger sample of PLWH. Second, we assessed whether a variety of histopathological markers were associated with neurocognitive functioning among the larger sample. Third, we genotyped several additional genetic susceptibility loci and tested for association with the histopathological markers within the frontal cortex. Finally, we examined the causal pathway between these histopathological markers and global neuropsychological functioning. Our goal was to exploit knowledge of functional polymorphisms to identify the relevant genes and histopathological markers involved in HAND pathogenesis.

Methods

Biological specimens were obtained from the National NeuroAIDS Tissue Consortium (NNTC) (Morgello et al., 2001). The NNTC is a longitudinal study of neuroAIDS in existence since 1998. Specimens and data used in the current study generally came from participants recruited because they had one or more diagnoses indicative of advanced HIV-1 disease, were at high risk of death, and agreed to participate in the study evaluations and to donate their organs for research purposes. Participants were typically evaluated every 6–12 months, undergoing comprehensive neurocognitive testing, psychiatric/substance use interview, and neuromedical evaluation. Upon death, brains were harvested for research purposes. Demographic and HIV-1 disease characteristics of individuals included in the current study are shown in Table 1.

Table 1 Sample characteristics
Full size table

For inclusion, all cases were required to be HIV-seropositive, 18 years or older, and diagnosed as either neurocognitively normal or with HAND within 1 year prior to death, per established research criteria (Antinori et al., 2007; No authors, 1996). Those determined to be neurocognitively impaired due to other causes were not included. All cases died well into the cART era (post-1996). Exclusion criteria were (1) pre- or postmortem evidence of non-HIV-related neurological diseases (e.g., stroke, neoplasm, multiple sclerosis, traumatic brain injury, and neurodegenerative illness) and (2) history or evidence of central nervous system (CNS) toxoplasmosis or progressive multifocal leukoencephalopathy. Comorbid medical conditions were in most cases self-reported by participants during the visits just prior to death, and were confirmed via chart review whenever possible. Sample characteristics are displayed in Table 1.

DNA extraction and genotyping

DNA extraction and genotyping methods are provided in the Supplemental Material. Table 2 displays the genes, specific reference SNP cluster ID numbers, and gene function.

Table 2 Candidate genes and their protein’s primary function
Full size table

Clinical variables

Neurocognitive functioning

Neuropsychological clinical ratings were determined for each case based on neurocognitive test scores obtained within 1 year of death. A global ability rating was derived from demographically corrected T-scores from a comprehensive neuropsychological battery, as previously described (Woods et al., 2004). Clinical ratings were assigned on a scale that ranged from 1 (above average) to 9 (severely impaired), with scores greater than or equal to 5 indicative of at least mild impairment. These were summarized as a global clinical rating (GCR). Among PLWH, the GCR was associated with activities of daily living (Heaton et al., 2004), HIV-1 disease variables (Heaton et al., 2011), and synaptodendritic changes on brain histopathology (Moore et al., 2006).

HIV-1 disease measures

Peripheral blood was collected from living participants by venipuncture into EDTA and heparinized tubes prior to death and was assayed using the Roche Amplicor Assay for HIV-1 RNA viral load and by flow cytometry for CD4+ T lymphocyte subsets. Plasma HIV-1 RNA viral load was measured at the last premortem visit within 1 year of death. Plasma measures of viral load and CD4+ T cell count were not available at the time of death because venipuncture cannot be performed after the heart has ceased beating due to intravascular blood coagulation. Duration of HIV-1 infection was based on self-reported date of infection and confirmed by chart review when possible.

Antiretroviral CNS penetration or effectiveness

We employed the antiretroviral CNS penetration or effectiveness (CPE), a score that is based on the pharmacologic characteristics of antiretroviral medications (Letendre, 2011). The CPE of individual antiretroviral drugs is ranked from 1 (poorest) to 4 (best) based on the 2010 ranking system (Letendre et al., 2010). The CPE score for each case was derived by adding ranks of all antiretroviral drugs within the regimen, which was reported at the time of neurocognitive testing. Higher scores indicated a regimen with increased penetration of the blood-brain barrier.

Alcohol and substance use

The Psychiatric Research Interview for Substance and Mental Disorders (PRISM) (Hasin et al., 1996) or Composite International Diagnostic Interview (CIDI) (Robins et al., 1988) was used to ascertain lifetime substance use disorders. Both are structured diagnostic interviews that yield DSM-IV diagnoses. For the purposes of the current study, NNTC participants were classified with none, current, or past substance use disorder for the following drugs: cocaine, opioids, and methamphetamine. Alcohol use was similarly classified.

Immunohistochemistry and histopathological characterization

As described in (Moore et al., 2006; Masliah et al., 1997), brain tissue was harvested from deceased HIV-seropositive NNTC participants as soon as possible after death. Tissue blocks measuring 4 cm3 were taken from the right dorsolateral midfrontal cortex. The blocks were fixed overnight in 4% paraformaldehyde and cut at 40 μm thick with a Leica Vibratome (Vienna, Austria). Histopathological characterization was accomplished using previously described methods (Moore et al., 2006; Masliah et al., 1997) based on immunohistochemistry conducted on formalin-fixed paraffin-embedded sections. The midfrontal gyrus was processed for the following markers: SYP (gray matter), MAP 2 (gray matter), human leukocyte antigen-DR (HLA-DR, gray and white matter separately), Iba1 (gray and white matter separately), glial fibrillary acidic protein (GFAP, gray and white matter separately), and β-amyloid (gray matter). Additional details of immunohistochemistry analysis and quantitative image analysis are in the Supplemental Material.

Statistical analysis

Significance was assessed at a false discovery rate (FDR) of 0.05 (Benjamini & Hochberg, 1995).

All genetic loci except MBL2 and APOE were treated as having a dominant acting risk allele. MBL2 genotypes were treated as three categories (A/A, A/O, and O/O). To test the association of APOE ε4 and ε2 alleles with GCR, we modeled dominant main effects and an interaction between ε4 and ε2 in order to allow the inclusion of participants carrying the APOE ε2/ε4 genotype. Symbolically, this model was:

$$ \mathrm{GC}{\mathrm{R}}_i=\upalpha +{\upbeta}_{\upvarepsilon 4}\ {\left|{}_{\left\{\mathrm{i},\upvarepsilon 4/\upvarepsilon 4,\upvarepsilon 4/\upvarepsilon 3,\upvarepsilon 4/\upvarepsilon 2\right\}}+{\upbeta}_{\upvarepsilon 2}\ \right|}_{\left\{\mathrm{i},\upvarepsilon 2/\upvarepsilon 2,\upvarepsilon 2/\upvarepsilon 3,\upvarepsilon 4/\upvarepsilon 2\right\}}+\upgamma \upvarepsilon 4\upvarepsilon 2\ {\left|{}_{\left\{\mathrm{i},\upvarepsilon 4/\upvarepsilon 4,\upvarepsilon 4/\upvarepsilon 3,\upvarepsilon 4/\upvarepsilon 2\right\}}\ \right|}_{\left\{\mathrm{i},\upvarepsilon 2/\upvarepsilon 2,\upvarepsilon 2/\upvarepsilon 3,\upvarepsilon 4/\upvarepsilon 2\right\}}+{\mathrm{e}}_{\mathrm{i}} $$

where I{ε4/ε4, ε4/ε3, ε4/ε2} = 1 if individual i’s genotype was ε4/ε4, ε4/ε3, or ε4/ε2 and = 0 otherwise, I{ε2/ε2, ε2/ε3, ε4/ε2} = 1 if individual i’s genotype was ε2/ε2, ε2/ε3, or ε4/ε2 and = 0 otherwise, and ei was the residual error. If the evidence for interaction was not significant, we reduced the model to: GCRi = α + βε4 I{i, ε4/ε4, ε4/ε3, ε4/ε2} + βε2 I{i, ε2/ε2, ε2/ε3, ε4/ε2} + ei

To test potential causal pathways involving histopathological markers, we used a multi-stage linear regression-based mediation analysis approach (MacKinnon, 2008; Tingley et al., 2014). The evidence for a particular pathway was provided by the proportion of the effect of a predictor that could be explained by the effect of a mediator (the proportion mediated). The proportion mediated was determined by conducting two linear regressions: (1) regressing the mediator on the predictor and (2) regressing the outcome on the mediator and the predictor. Additional covariates could be included in both regressions if desired.

Results

We first determined if potential confounders were associated with GCR (Table 3), including age at death, sex (female or reference group male), duration of HIV-1 infection, log10 plasma HIV-1 RNA viral load (herein referred to as viral load), NNTC site (reference group site 1), CD4+ T cell count, CPE score, self-reported race/ethnicity (African American, Hispanic, Asian/Native Alaskan, or reference group Caucasian), alcohol use (current, past, or reference group none), or drug use (current, past, or reference group none). Only viral load was associated with GCR using a significance threshold of 0.05. Because the inclusion of race/ethnicity could reduce the effects of population stratification that confounded genetic association analyses, we included race/ethnicity in our subsequent analyses despite lack of statistical significance.

Table 3 Potential confounding covariates of GCR, SYP, and MAP 2
Full size table

We next tested each of the histopathological markers’ association with GCR, controlling for the effects of race/ethnicity and viral load (Table 4). Only MAP 2 and SYP were associated with GCR at an FDR of 0.05. We then determined whether the potential confounders were associated with MAP 2 or SYP. The viral load was significantly associated with SYP. None of the potential confounder variables were associated with MAP 2 (Table 3).

Table 4 Histopathological marker associations with GCR controlling for race/ethnicity and plasma HIV-1 RNA viral load
Full size table

Association of Genotype with GCR, SYP, and MAP 2

We next determined candidate genetic markers’ associations with GCR, SYP, or MAP 2. Because genetic associations could be confounded by population stratification, we included race/ethnicity as a covariate in all subsequent analyses despite its lack of statistical significance. When using a nominal per test p value cutoff of 0.05, HFE, HEPH, and MBL2 were associated with GCR, NFE2L2, and CCR2 were associated with SYP, and NFE2L2, and HCP5 were associated with MAP 2. However, none of the genetic markers were significantly associated with GCR, SYP, or MAP 2 at an FDR of 0.05 (Table 5). Thus, when further testing for the association of MAP 2 and SYP with GCR, we did not include genetic markers.

Table 5 Association of genetic loci with GCR, SYP, and MAP 2
Full size table

Mediation analysis

When both MAP 2 and SYP were simultaneously included as predictors, MAP 2, but not SYP, was associated with GCR (Table 6). This suggests the relationship of SYP and GCR is mediated by MAP 2; however, there was strong collinearity between MAP 2 and SYP. To test this potential pathway, we used mediation analyses to determine the most plausible pathway that could explain the observed association of SYP, MAP 2, and GCR. We assumed that GCR would not influence SYP or MAP 2; therefore, we limited our analyses to four possible pathways: (1) the effect of SYP on GCR was mediated through MAP 2 (SYP ➔ MAP 2 ➔ GCR); (2) the effect of MAP 2 on GCR was mediated through SYP (MAP 2 ➔ SYP ➔ GCR); (3) neither SYP nor MAP 2 was a mediator of the other; and 4) a more complex pathway in which both MAP 2 and SYP mediated the effects of each other, possibly due to unmeasured variables. If pathway 1 (SYP ➔ MAP 2 ➔ GCR) was the predominant pathway, we would expect the mediation effects to be significant for this pathway and not for pathway 2 (MAP 2 ➔ SYP ➔ GCR). If pathway 2 was the predominant pathway, the mediation effects would be significant for this pathway and not for pathway 1. If neither SYP nor MAP 2 mediated, neither analysis would be statistically significant. If both SYP and MAP 2 had direct and mediated effects, both analyses would be expected to be statistically significant. Based on the mediation analyses, the statistical support for pathway 1 (Fig.1) was much stronger than that for pathway 2 (Fig. 2). Therefore, these results were most consistent with the SYP ➔ MAP 2 ➔ GCR causal pathway.

Table 6 Effect sizes and p values of GCR, MAP 2, and SYP as outcomes
Full size table
Fig. 1
figure 1

Mediation analysis results for the model MAP 2 mediating the effect of SYP on GCR. The model includes log10 plasma HIV-1 RNA viral load and race/ethnicity as covariates. Estimates for the average causal mediation effect (ACME), the average direct effect (ADE), and the total effect are denoted as closed circles, and the 95% confidence intervals (CI) are denoted as horizontal lines. Effects significant at a p value less than 0.05 do not cross the hatched vertical line at zero. The ACME p value = 0.032, the ADE p value = 0.282, and the total effect p value = 0.004. The proportion mediated = 0.547 (95% CI = 0.045, 1.832; p value = 0.036), supporting the hypothesis that MAP 2 mediates the effect of SYP on GCR, SYP ➔ MAP 2 ➔ GCR (MAP 2 mediating SYP effect on GCR), a Controlling for log10 plasma HIV-1 RNA viral load and race/ethnicity

Full size image
Fig. 2
figure 2

Mediation analysis results for the model SYP mediating the effect of MAP 2 on GCR. The model includes log10 plasma HIV-1 RNA viral load and race/ethnicity as covariates. Estimates for the average causal mediation effect (ACME), the average direct effect (ADE), and the total effect are denoted as closed circles, and the 95% confidence intervals (CI) are denoted as horizontal lines. Effects significant at a p value less than 0.05 do not cross the hatched vertical line at zero. The ACME p value = 0.296, the ADE p value = 0.030, and the total effect p value = 0.002. The proportion mediated = 0.212 (95% CI = − 0.196, 0.871; p value = 0.294), indicating a lack of support for the hypothesis that SYP mediates the effect of MAP 2 on GCR, MAP 2 ➔ SYP ➔ GCR (SYP mediating MAP 2 effect on GCR), a Controlling for log10 plasma HIV-1 RNA viral load and race/ethnicity

Full size image

Discussion

In the current study, we took several steps to address the question of whether or not host genotype affected the histopathological factors that underlay the neurocognitive dysfunction common to PLWH. The first step involved determining if the quantitative measure of neurocognitive functioning (i.e., GCR) was associated with virologic, demographic, mood, CNS penetration of antiretroviral medication, and/or substance use factors. In this sample, only plasma HIV-1 RNA viral load was significantly associated with GCR. This finding might be somewhat unexpected considering that recent studies did not find viral load to be associated with cognitive functioning, at least in the current era of widespread cART use (Clifford & Ances, 2013). However, the NNTC cohort is composed of individuals with more advanced illness, in which viral load was likely higher and cognitive functioning poorer than in clinical cohorts of living PLWH in general. Furthermore, the higher viral load might reflect poor medication adherence and health in general, which suggests that other medical factors not captured in the current analysis (e.g., metabolic syndrome and comorbid medical illnesses) may be contributing. The lack of association between demographics and GCR may be due to the use of normative data when interpreting neurocognitive test scores. That is, standardized scores were derived using age, education, gender, and/or ethnicity stratified normative data. Neither current nor past substance or alcohol use disorder was associated with GCR, despite our focus on the most neurotoxic drugs (cocaine, methamphetamine, and opioids). This is somewhat surprising considering that the majority of past studies have reported an increased risk of neurocognitive impairment associated with substance use (Martin et al., 2018; Thaler et al., 2015; Persidsky et al., 2011; Meyer et al., 2013). One possible explanation is that the relatively advanced disease stage of our sample, with concomitant higher rates of severe cognitive impairment (i.e., HIV-associated dementia), might simply overshadow the damaging effects of alcohol or substance use. Finally, as a group, our sample was neurocognitively impaired based on the average GCR. The reason for this impairment may not have been captured by our study due to incomplete data collection, as mentioned above. Among those important variables not captured were comorbid medical conditions, which in more recent studies appeared to be among the greatest predictors of neurocognitive impairment (Saloner et al., 2019; Sacktor et al., 2016; Lake et al., 2017; Wu et al., 2018; Vance et al., 2016). As such, the documented adverse effects of the recreational drugs included in our analyses, as well perhaps of depression, may have been overshadowed by factors that were not measured.

The second step involved determining which histopathological markers were associated with GCR. Only MAP 2 and SYP were found to be significant predictors of global cognitive functioning. This finding is consistent with our earlier study (Levine et al., 2015), as well as other studies, many of which included samples from the NNTC (Moore et al., 2006; Bryant et al., 2017; Masliah et al., 1992). There are several mechanisms through which HIV-1 may affect synaptodendritic functioning, including direct actions via envelope protein gp120 and transactivator of transcription (tat) protein, as reviewed in (Wenzel et al., 2019). Furthermore, our inclusion of potential confounding variables found to be significant in the initial step of our analyses revealed that plasma HIV-1 RNA viral load was significantly associated with SYP, but not with MAP 2.

In the third step, we tested for associations of each candidate genetic marker with GCR, SYP, and MAP 2, controlling for plasma HIV-1 RNA viral load and race/ethnicity. We expected to discover some associations, based on our previous finding (Levine et al., 2015). However, there were no significantly associated genetic markers with any of the outcomes after correcting for multiple comparisons using an FDR of 0.05. Therefore, despite the increase in sample size from the previous study, we found no statistically significant association of these outcomes with the genetic markers, suggesting that these genetic pathways are not major determinants of neurocognitive dysfunction or neurodegeneration among PLWH with advanced disease. Such findings further underscore the likely lack of consistently significant genetic influence on neurobehavioral outcomes in PLWH, as reviewed in (Levine et al., 2014a; Kallianpur & Levine, 2014).

The final step of our study was to apply mediation analysis to examine potential causal pathways relating SYP, MAP 2, and GCR. This step was taken to determine if the effect of MAP 2 on GCR was mediated through SYP, or if the effect of SYP on GCR was mediated through MAP 2. As the directed pathway of MAP 2 to SYP to GCR did not show any significant evidence and the directed pathway of SYP to MAP 2 to GCR was significant, the most plausible causal pathway is that MAP 2 mediates the effect of SYP on GCR. Coupled with the findings from earlier steps, an overall model emerges in which elevated plasma HIV-1 RNA viral load results in presynaptic degeneration (as indicated by SYP levels), which in turn leads to somatodendritic degeneration (as indicated by MAP 2 levels) and ultimately neurocognitive dysfunction. However, even if presynaptic degeneration is a downstream effect of more active systemic viral replication, it is almost certain that additional intermediary factors are involved, such as changes in neurogranin and calmodulin (Guha et al., 2018), cathepsin B and serum amyloid p component (Cantres-Rosario et al., 2019), E2F transcription factor-3 (Santerre et al., 2019), and BCL11B-encoded protein (Desplats et al., 2013), to name just a few. Importantly, this putative multicomponent sequence of events leading to neurocognitive dysfunction in PLWH is just one of a list of several other causes, a list that includes medical comorbidities (Saloner et al., 2019; Yu et al., 2019), low-grade systemic chronic inflammation (Geffin & McCarthy, 2018), and antiretroviral medication toxicity (Tovar-y-Romo et al., 2012; Gannon et al., 2017).

The current study had several limitations. Perhaps most important was the absence of key variables in our analysis, including those mentioned in the immediately preceding paragraph (e.g., medical comorbidities), as well as brain HIV-1 viral load, CD163, and germane medical comorbidities. Additionally, the generalizability of the findings to the current cART era is limited, as the NNTC cases typically have more advanced illness and comorbidities than the vast majority of PLWH, at least within the USA (Morgello et al., 2001). This is best demonstrated by the relatively high rate of HIV-associated dementia (29%) and HIV encephalitis (9%) in our sample. Our use of plasma HIV-1 RNA viral load rather than the brain- or CSF-derived viral load also limits the interpretability, as these two measures are not strongly correlated (Gelman et al., 2013; Bavaro et al., 2019). Finally, while our study may be among the largest genetic-histopathological studies of HIV-related neurocognitive impairment, the statistical power was nevertheless limited by the sample size.

In conclusion, while our results affirmed the role of synaptodendritic degeneration in HIV-related neurocognitive impairment, we did not find definitive evidence of a host genetic influence on these histopathological markers among individuals with advanced HIV-1 disease. Our study adds to the existing literature of genetic association studies of HAND, which have focused on behavioral rather than histopathological phenotypes. Our findings do suggest that presynaptic degeneration precedes somatodendritic degeneration in the lead up to neurocognitive impairment.