Central nervous system (CNS) transcriptomic correlates of human immunodeficiency virus (HIV) brain RNA load in HIV-infected individuals

To generate new mechanistic hypotheses on the pathogenesis and disease progression of neuroHIV and identify novel therapeutic targets to improve neuropsychological function in people with HIV, we investigated host genes and pathway dysregulations associated with brain HIV RNA load in gene expression profiles of the frontal cortex, basal ganglia, and white matter of HIV+ patients. Pathway analyses showed that host genes correlated with HIV expression in all three brain regions were predominantly related to inflammation, neurodegeneration, and bioenergetics. HIV RNA load directly correlated particularly with inflammation genesets representative of cytokine signaling, and this was more prominent in white matter and the basal ganglia. Increases in interferon signaling were correlated with high brain HIV RNA load in the basal ganglia and the white matter although not in the frontal cortex. Brain HIV RNA load was inversely correlated with genesets that are indicative of neuronal and synaptic genes, particularly in the cortex, indicative of synaptic injury and neurodegeneration. Brain HIV RNA load was inversely correlated with genesets that are representative of oxidative phosphorylation, electron transfer, and the tricarboxylic acid cycle in all three brain regions. Mitochondrial dysfunction has been implicated in the toxicity of some antiretrovirals, and these results indicate that mitochondrial dysfunction is also associated with productive HIV infection. Genes and pathways correlated with brain HIV RNA load suggest potential therapeutic targets to ameliorate neuropsychological functioning in people living with HIV.


Gene expression and pathway analysis. For pathway analysis, we used Gene Set Enrichment Analysis
(GSEA), a computational method to assess whether a priori defined sets of genes show statistically significant differences between biological states 31 . GSEA was used in conjunction with genesets from the Molecular Signatures Database (MSigDb), including canonical pathways in the C2 collection 32,35 . GSEA uses the Kolmogorov-Smirnov statistical test to assess whether a predefined geneset, here a pathway from the C2 collection, is statistically enriched in differentially expressed genes by testing their distribution in the full list of genes ranked by their differential expression between two biological states 31 . The MSigDB is a comprehensive database of genesets for performing geneset enrichment analysis that represents a wide range of biological processes and diseases 32,35 . A total of 1322 canonical Pearson pathways from the MSigDB C2 collection were scored and ranked using the GSEA algorithm as described also in 12 using R. Multiple testing adjustment was performed and significance was assessed using the False Discovery Rate (FDR) in GSEA algorithm.

Correlation between HIV transcription and host gene and pathway expression in humans.
We first computed Pearson correlations and associated p values between the expression levels of host genes and HIV RNA load. HIV transcription was used as a single value to estimate the overall virus expression level. Adjustments of p values were performed using the Benjamini-Hochberg FDR and significantly correlated host genes with HIV transcripts were selected by applying an appropriate threshold (e.g., 5% threshold by the FDR). Next, we separated the samples into two groups based on the mean expression of HIV transcripts to create groups with graded levels of HIV transcription (e.g., high and low HIV, or higher levels of expression as warranted by the results). Differential expression between groups was computed with Limma and used as an input for GSEA to identify pathways that were differentially regulated between samples with high and low HIV expression (i.e., load). Pathways that were tested were extracted from the MSigDb C2 canonical pathway collection. We selected pathways that were significantly associated with brain HIV load using the GSEA FDR < 0.05.

Correlation between HIV transcription and host gene and pathway expression in a rodent model of HIV.
HIV transgenic (Tg) rats on a mixed Wistar-Fischer background, which ensures variability of host genes and HIV expression levels, were used for the study 36 . Male HIV Tg rats, weighing between 350 and 400 g, were housed two per cage on a reverse 12 h/12 h light/dark cycle (lights off at 8:00 AM) in a temperature (20-22°C) and humidity (45-55%) controlled vivarium with ad libitum access to tap water and food pellets (PJ Noyes, Lancaster, NH, USA). All of the procedures were conducted in strict adherence to the National Institutes of Health Guide for the Care and Use of Laboratory Animals and were approved by the Institutional Animal Care and Use Committee of The Scripps Research Institute and ARRIVE guidelines (http:// www. nc3rs. org. uk/ page. asp? id= 1357). A total of 18 gene expression profiles from prefrontal cortices of HIV Tg rats profiled by RNA-Seq as previously described 36 were separated into two groups based on the mean expression of brain HIV transcripts to create groups with graded levels of HIV transcription (n = 7 for high brain HIV RNA load and n = 10 for low brain HIV RNA load). Differential expression between groups was computed with DESeq2 and used as an input for the GSEA to identify pathways that were differentially regulated between samples with high and low HIV expression. A normalized enrichment score and p-value were calculated for each pathway.

Results
Host transcriptional correlates of brain HIV RNA loads in human brain samples. Gene expression profiles from three brain regions of HIV-infected patients, including the basal ganglia, white matter, and the frontal cortex, were used for the study. The gene expression data were filtered and normalized as described previously 26 . Pearson correlations and associated p values were computed between the expression levels of host genes and HIV transcripts for each brain region as outlined in the Methods section. Genes with a high degree of correlation with brain HIV RNA loads in each of the three brain regions studied are shown in Fig. 1A-C and Supplemental Table S2.
We applied GSEA, a computational method that assesses whether a priori-defined genesets show statistically significant differences between biological states 31 , to compare gene expression profiles associated with high brain HIV RNA loads. GSEA was used in conjunction with genesets from the MSigDb, including canonical pathways in the C2 collection 32 . In total, we identified 254 pathways in the basal ganglia, 104 pathways in white matter, and 393 pathways in the frontal cortex that were significantly enriched/decreased in samples with high brain HIV load (FDR < 0.05).

Figure 1.
Correlation of gene expression and pathway dysregulation with brain HIV RNA load in brain regions of patients with HIV. (A-C) Host genes that correlated with brain HIV RNA load in three brain regions of patients with HIV. Volcano plots depict significant correlation with brain HIV RNA loads in the three brain regions under study (left side; also see Supplemental Table S2)  ). Asterisk indicates pathways significantly concordantly dysregulated in all three brain regions (FDR < 0.05). (E-G) Representative host pathways that were concordantly correlated with HIV expression in all three brain regions that were indicative of (E) increase in cytokine signaling, (F) neurodegeneration, and (G) dysregulation of bioenergetics, between samples from patients with high brain HIV RNA load (HIV-H) and samples from patients with low brain HIV RNA load (HIV-L). The example in panel (E) is from the basal ganglia, the one in panel F is from the frontal cortex, and the one in panel (G) is from white matter. Similar results were obtained for these pathways in the three brain regions as shown in (D). Changes in the expression of the pathway in the GSEA plots, such as the ones in (E-G), are indicated by the asymmetric distribution of genes in the geneset (vertical bars) and of the line plot of the running normalized enrichment score 31  www.nature.com/scientificreports/ The top differentially regulated pathways in the three brain regions (concordantly changed in all regions) ranked by the normalized enrichment score (NES) are shown in Fig. 1D. Overall, the pathway analysis indicated that host genesets correlated with brain HIV RNA load and were concordantly dysregulated in the three brain regions studied, which could mostly be grouped into three broad biological processes: inflammation, neurodegeneration, and bioenergetics. Pathways related to inflammation, neurodegeneration, and bioenergetics were found to be correlated with HIV levels in all three brain regions ( Fig. 1D-G). Pathways that are representative of inflammation and were concordantly dysregulated across brain regions were primarily representative of increased cytokine signaling and directly correlated with brain HIV RNA load. Pathways that are representative of neurodegeneration included pathways that are representative of gene expression changes in Alzheimer's disease (AD), Parkinson's disease (PD), and Huntington's disease (HD) were found to be dysregulated across all brain regions.
Pathways that are representative of bioenergetics, including pathways that are related to oxidative phosphorylation, electron transfer, and the tricarboxylic acid (TCA) cycle, were inversely correlated in all three brain regions studied (Fig. 2). Transcripts of several genes involved in energy metabolism were inversely correlated with HIV RNA load in all three brain regions, including α-enolase (ENO1), a glycolytic enzyme that catalyzes the conversion of 2-phosphoglycerate to phosphoenolpyruvate, glyceraldehyde 3-phosphate dehydrogenase (GAPDH), which catalyzes the sixth step in glycolysis, converting glyceraldehyde 3-phosphate to d-glycerate 1,3-bisphosphate, and the mitochondrial membrane ATP synthase subunit C locus 3 (ATP5G3), Supplemental Table S2.
The differential expression of genes involved in energy metabolism and the TCA cycle was confirmed in an independent set of frontal cortices from individuals with high and low brain HIV RNA loads by quantitative real time PCR (RT-PCR), Fig. 3.
Pathway dysregulations in each of the three brain regions studied are reviewed next.
Basal ganglia. The pathways that showed the highest correlation with brain HIV load in the basal ganglia included pathways indicative of inflammation and immune system activation, including interferon type I and II and cytokine signaling, Toll-like receptor signaling, tumor necrosis factor-α signaling, JAK-STAT signaling, and interferon (Figs. 2 and 4). Inversely correlated with brain HIV RNA load in the basal ganglia were several pathways that are indicative of neural degeneration, including pathways that are representative of gene expression changes in AD, PD, and HD (Figs. 2 and 5), indicative of a reduction of neuronal signaling and trophism. Also inversely correlated with brain HIV RNA load were pathways related to synaptic plasticity, α-amino-3hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptor trafficking, and axonal function (Figs. 2 and 5, Supplemental Table S3). Pathways that are related to energy metabolism, including the TCA cycle, oxidative phosphorylation, and glycolysis, were concomitantly reduced with higher virus brain RNA load (Figs. 2 and 5). Transcriptional evidence of a catabolic milieu and evidence of impairments in proteostasis and mitochondrial protein synthesis were also associated with higher brain HIV RNA load (Figs. 2 and 5). Among the genes that showed higher correlations with virus titers were transforming growth factor, β-induced (TGFBI), signal transducer and activator of transcription 1 (STAT1), meteorin-like (METRNL), interferon-induced with helicase C domain 1 (IFIH1), interferon-inducible protein 16 (IFI16), IFI30, interferon-inducible transmembrane 1 (IFITM1), and γ-interferon-inducible lysosomal thiol reductase (GILT; Fig. 1A, Supplemental Table S4). www.nature.com/scientificreports/   Table S5). Transcriptional evidence of increased neurodegeneration and impaired bioenergetics was also correlated with brain HIV RNA load in white matter, including pathways representative of AD, PD, and HD, reductions of TCA cycle and electron transport, and glycolytic genes (Figs. 2 and 7). Concomitant evidence of a catabolic milieu was also associated with higher brain HIV RNA load (Figs. 2 and 7).
Frontal cortex. Pathways reflecting changes in the expression of neuronal genes, including axonal and synaptic transcripts, trophic signaling, and mitochondrial and other key genes that are associated with neurodegenerative conditions were dysregulated in the frontal cortex ( Figs. 2 and 8). Like the basal ganglia and white matter, these included genesets involved in AD, PD, and HD and evidence of reductions of synaptodendritic injury, synaptic signaling and plasticity, neurotrophin signaling, and energy metabolism (Fig. 7). The latter included reductions of genesets that are representative of TCA cycle, electron transport, and glycolytic genes (Figs. 2, 3 and 8). A catabolic milieu was also indicated by impairments in RNA synthesis and processing and impairments in proteostasis and protein synthesis that are associated with brain HIV RNA load (Fig. 8). Immune activation and inflammation were also correlated with brain HIV RNA load (Fig. 8). Pathways representative of increased inflammation and cytokine signaling were also correlated with brain HIV RNA load in the frontal cortex (Fig. 8).
Transcriptional correlates of brain HIV RNA load in a rodent model. HIV transgenic rats express multiple HIV products in disease-relevant cells in the brain, such as microglia and astrocytes, under control of the viral LTR promoter [37][38][39] . These rats also have similar brain gene expression changes as humans with HIV 26,36,38 . The repertoire of pathway dysregulations in humans with HIV in the present study was generally consistent with observations in HIV transgenic rats 36,38 . As depicted in Fig. 9, using previously reported gene expression profiles of the frontal cortex in HIV transgenic rats on a mixed genetic background 36 , the expression of genesets representative of neurodegeneration and bioenergetics, although not inflammation-related genesets, was significantly correlated with brain HIV RNA loads in the frontal cortex in HIV transgenic rats, despite a lower HIV RNA dynamic range (Fig. 9).

Discussion
The molecular mechanisms of HIV-associated neurotoxicity have been only partially elucidated. cART has itself been implicated in neurotoxicity [40][41][42] . Brain HIV viral load and indirect indicators, such as CSF HIV RNA load, are generally, albeit variably, inversely correlated with neuropsychological scores 10,43-45 . To better highlight genes and pathway dysregulations that have greater potential relevance to neuroHIV severity and progression, www.nature.com/scientificreports/ we correlated transcriptional alterations of functional pathways and genes in three brain regions in human cases of HIV with brain HIV RNA load. We used GSEA, a knowledge-based computational strategy that interrogates a genome-wide expression profile dataset using a priori-defined genesets of functionally related genes 46 . GSEA was used in conjunction with the MSigDb, including canonical pathways in the C2 collection 32 .
Here, we found that transcriptional evidence of inflammation was more pronounced in white matter and the basal ganglia. Pathways that are representative of cytokine signaling were correlated with brain HIV RNA load in all three brain regions, consistent with the contribution of inflammatory processes to cortical, subcortical, and white matter injury 47 . Increases in interferon signaling were associated with high brain HIV RNA load in the basal ganglia and white matter. However, despite considerable activation of interferon-related pathways in the frontal cortex of HIV cases versus uninfected control individuals in the present human neuroHIV dataset 26 , their overall expression was not significantly correlated with brain HIV RNA load in the frontal cortex. In fact, significant activation of interferon signaling is seen in the frontal cortex even in the absence of neurocognitive impairment 26 . Conversely, the present results highlight that in subcortical regions such as the basal ganglia and white matter, the activation of interferon signaling was directly correlated with brain HIV RNA load. www.nature.com/scientificreports/ Effective cART is associated with lower intrathecal inflammation [48][49][50] . A multilevel analysis of the neuropathogenesis of neurocognitive impairment in HIV showed that brain HIV RNA load was significantly correlated with inflammation and markers of neurodegeneration 45 . Significant correlations were also found between inflammatory markers, particularly type I interferon, and brain or CSF HIV RNA load has been observed in both recent cohorts 10,51-53 and earlier ones [54][55][56] .
In the present study, brain HIV RNA load was inversely correlated with genesets that are indicative of neuronal and synaptic genes, particularly in the cortex, indicative of synaptic injury and neurodegeneration. Genesets that are related to neurodegeneration and were found to be dysregulated in cases with high brain HIV RNA load included pathways that are representative of gene expression changes in AD, PD, and HD. Established immunohistochemical markers of neuropathological changes that underlie the progression of neuroHIV include preand postsynaptic markers, such as synaptophysin (SYP) and microtubule-associated protein-2 (MAP2) 45,57 , and abnormal protein aggregates, such as β-amyloid [58][59][60][61][62] . MAP2 and SYP are indicators of synaptic integrity, and their expression in the frontal cortex is inversely correlated with plasma and brain HIV RNA load 45 . β-amyloid plaque burden in the frontal cortex also correlated with brain HIV RNA load, and SYP and β-amyloid were negatively correlated with HIV RNA load 45 . Transcriptional evidence of alterations of proteostasis that were associated with brain HIV RNA load was also found, particularly in the basal ganglia and frontal cortex. Altogether, these results indicate that brain HIV RNA load is correlated with synaptodendritic injury and dysfunctional protein clearance. Evidence of neurodegeneration was also found in the basal ganglia and white matter, in addition to the frontal cortex. This is consistent with mounting evidence of the involvement of the basal ganglia and white matter in aging and neurodegenerative diseases, such as AD and PD [63][64][65][66][67][68][69][70][71] .
Emerging neuropathology concepts in neuroHIV also suggest a degree of pathogenic similarity and possibly overlap between neuroHIV and AD. Several studies reported increases in amyloid deposition in older HIV-infected individuals, suggesting that long-term HIV and antiretroviral therapy might interfere with the clearance of proteins, such as amyloid-β peptide, and worsen neuronal damage and cognitive impairment in this population 58,60,61,72,73 . In one study, amyloid-β plaques in HIV+ brains, although immunohistologically different from those in symptomatic AD brains, were associated with HAND among apolipoprotein E (APOE) ε4 carriers, suggesting the possibility of convergent mechanisms of pathogenesis 61 . In vitro studies indicate that HIV proteins can disrupt different steps of amyloid pathways 59,[74][75][76] . The HIV protein Tat, which is produced by the HIV reservoir in the brain even in the setting of the suppression of viral replication 77 , has been shown to bind amyloid-β peptide, promoting amyloid aggregation and neurotoxicity 78 . When Tat was injected in the brains of amyloid precursor protein and presenilin-1 transgenic mice (APP-PS1), it colocalized with APP 78 . Furthermore, crossing Tat transgenic mice with APP-PS1 transgenic mice resulted in an increase in amyloid-β deposition, neurodegeneration, neuronal apoptotic signaling, and phosphorylated Tau compared with PSAPP mice 79 , indicating that Tat likely contributes to AD-like pathology in HIV. As the HIV-positive population ages, individuals with both HIV and canonical AD pathology are beginning to be diagnosed. A recent case report of AD in an HIV-infected patient who was positive for amyloid on positron emission tomography 80 was followed by approximately 20 more cases 81 . Thus, current issues of debate include the possibility of HAND/AD mixed dementia and the hypothesis that HIV may represent a predisposing factor for AD [80][81][82][83] . Supporting the latter possibility, combined CSF biomarker risk for AD in an Australian HIV-positive cohort was found to be more than 10-times greater than in age-matched controls 82 . www.nature.com/scientificreports/ The present data also show that the gene expression evidence of impairments in bioenergetics are associated with brain HIV RNA load and are correlated with transcriptional evidence of inflammation and neurodegeneration. Mitochondrial dysfunction has been implicated in neurodegenerative disease (e.g., AD [84][85][86] and PD 86,87 ), aging 88 , and neuroHIV, particularly in adverse ageing in older people who live with HIV 89 . Frontal cortices from patients with HAND showed mitochondrial abnormalities, including a reduction of mitochondrial biogenesis 90 . The accumulation of mutations and deletions of mitochondrial DNA (mtDNA) 91,92 altered mitochondrial fission and fusion 93 . Experimental evidence shows that HIV products, such as Tat and gp120, can cause mitochondrial damage [94][95][96][97][98][99][100] . Mitochondrial dysfunction is also implicated in the toxicity of some antiretrovirals 101 . The present data indicate that mitochondrial dysfunction is a key component of the neuropathogenesis of HIV infection itself.
In the present study, we found that a rodent model of HIV, in which several HIV products are expressed in disease-relevant glial cells (e.g., microglia and astrocytes) but not in neurons 37,38 , exhibited dysregulations of bioenergetics-and neurodegeneration-related genesets that were associated with brain HIV RNA paralleled findings in gene expression profiles of patients with HIV. In agreement with the present results, mitochondria from this HIV rodent model were found to have significantly lower oxygen consumption rates, basal respiration, ATP production, maximal respiratory capacity, spare capacity, proton leakage, and non-mitochondrial respiration 96 .
Emphasis has been rightfully placed on impairments in oxidative phosphorylation in neuroHIV and degenerative diseases, such as AD 84,[102][103][104][105] . The present pathway analysis underscores the potential contribution of disruptions of the TCA cycle and β-oxidation in the mitochondrial matrix, in which oxidative phosphorylation depends on the supply of reducing equivalents from the end-oxidation of nutrients 106 . The TCA cycle and oxidative phosphorylation are tightly coupled. The TCA cycle produces the reducing equivalents NADH and FADH2, which are required for electron transfer through the mitochondrial respiratory chain or electron transport chain (ETC). The oxidation of NADH and FADH2 in complexes I and II of the ETC is required to maintain the function of the TCA cycle 106 . As electrons are transported through the complexes, the ETC produces a mitochondrial membrane potential that generates ATP. Mitochondrial complexes I and II in the ETC replenish NAD+ and FAD, respectively, allowing the oxidative TCA cycle to function 106 . Thus, lower expression of TCA cycle-related genes may play a crucial role in impairments in mitochondrial oxidative phosphorylation in neuroHIV. Tricarboxylic acid cycle metabolites are also important for the biosynthesis of nucleotides, lipids, and proteins 106 . Mounting evidence indicates that metabolites in the TCA cycle are also involved in controlling chromatin modifications, DNA methylation, and post-translational modifications of proteins 106 . Acetyl-coenzyme A levels regulate chromatin dynamics by providing acetyl groups for the acetylation of histones by histone acetyltransferases [107][108][109][110] . www.nature.com/scientificreports/ The TCA cycle predominates in neurons, whereas glycolysis predominates in astrocytes 111 . Thus, decreases in TCA cycle-related gene expression in neuroHIV may be a correlate of lower neuronal energy metabolism. Evidence of alterations of TCA cycle capacity and expression that are induced by HIV infection in the scientific literature is limited. Interestingly, CSF metabolomics showed that worsening cognitive status in HIV-infected patients is associated with the accumulation of citrate and acetate, which are suggestive of disruptions of TCA cycle function 112 . A separate CSF metabolomic analysis of young individuals with HIV who were on ART found a dysregulation of TCA cycle intermediates, such as succinate and malate, among other metabolites, differed when compared with HIV-negative controls 113 . Relevant to the present results, we recently found that escalated (dependent) methamphetamine self-administration in HIV transgenic rats was associated with the lower expression of TCA cycle-related genes in conjunction with transcriptional evidence of increases in inflammation and neurodegeneration 36 . Selected genes that encode enzymes that are involved in glycolysis, such as ENO1 and GAPDH, were concordantly dysregulated in all three brain regions that were studied herein, and the expression of genesets that are representative of glycolysis was reduced particularly in the basal ganglia and frontal cortex, thus indicating that alterations of glycolysis and reductions of the TCA cycle and electron transport are key bioenergetic alterations that are correlated with brain HIV RNA load.
In conclusion, we explored functional pathways that are associated with brain HIV load as a correlate of neu-roHIV disease severity and progression by the pathway analyses of transcriptional profiles from brain regions of humans with HIV. We found that brain HIV RNA load correlated with transcriptional evidence of an increase in inflammation. Evidence of increased cytokine signaling was associated with high HIV RNA across the three brain regions studied. Increases in interferon signaling were correlated with high brain HIV RNA load in the basal ganglia and white matter although not in the frontal cortex. We also found transcriptional evidence of neurodegeneration that involved pathways that reflect changes in axonal and synaptic genes and trophic signaling, among others. Lastly, we found transcriptional evidence of impairments in bioenergetics that involved oxidative phosphorylation and the TCA cycle across all three brain regions. Transcriptional evidence of mitochondrial dysfunction was also associated with brain HIV RNA load and correlated with transcriptional evidence of inflammation and neurodegeneration, indicating that mitochondrial dysfunction is a key component of the neuropathogenesis of HIV infection itself. www.nature.com/scientificreports/