HIV expression persists in the cerebrospinal fluid of HIV-associated neurocognitive disorders despite effective ART

ABSTRACT

Despite effective antiretroviral therapy (ART), HIV-associated neurocognitive disorders (HAND) persist in people with HIV (PWH). The central nervous system (CNS) may act as a viral reservoir due to limited ART penetration and virological discordance between plasma and cerebrospinal fluid (CSF). In a cross-sectional study of 24 ART-treated PWH, participants were stratified as cognitively normal (CN, n = 10) or HAND (n = 14), including asymptomatic neurocognitive impairment (ANI, n = 3), mild neurocognitive disorder (MND, n = 9), and HIV-associated dementia (HAD, n = 2). HIV RNA was quantified in paired plasma and CSF by RT-ddPCR. CSF peptidome profiling was performed using mass spectrometry, and ART concentrations were measured by LC-MS/MS. HIV infectivity in CSF was assessed via viral outgrowth assays. HIV RNA was undetectable in plasma but present in CSF from HAND participants, indicating compartmentalized viral persistence. Tenofovir and lamivudine levels were higher in plasma, whereas dolutegravir trended higher in CSF. Nevertheless, all CSF drug concentrations exceeded their IC50 values in effectively suppressing active HIV replication. Peptidomic analysis identified HIV-derived peptides (e.g. Env and Pol) exclusively in HAND samples, accompanied by an early reduction in β-tau. Although HIV RNA and peptides were detectable, no productive infection was established by CSF in permissive immune cells. Together, despite pharmacologically sufficient ART penetration, HIV persists in the CSF of PWH with HAND. These findings suggest that the latent HIV infection with non-replicative viral expression, rather than residual active HIV replication, may contribute to neuroinflammation and cognitive decline in PWH on suppressive ART.

Introduction

The introduction of antiretroviral therapy (ART) has markedly improved the management of human immunodeficiency virus-1 (HIV), yet HIV-associated neurocognitive disorders (HAND) persist as a serious comorbidity [1–3]. Consistent ART adherence successfully suppresses plasma viral load and has led to a marked decline in severe HAND, such as HIV-associated dementia (HAD), within the central nervous system (CNS). However, milder HAND manifestations – specifically asymptomatic neurocognitive impairment (ANI) and mild neurocognitive disorder (MND) – persist at high prevalence even in individuals with undetectable viral RNA in peripheral blood following ART [3]. This is partly due to a virological discordance between the blood and the CNS; viral loads can remain elevated in the cerebrospinal fluid (CSF) despite being undetectable in peripheral blood [4–6]. This challenge is particularly acute in developing countries, such as Brazil, where socioeconomic factors, limited healthcare access, and a high burden of comorbidities likely contribute to both the persistence and underdiagnosis of milder HAND forms [5,7].

The pathophysiology of HAND is closely linked to the ability of HIV to infect and replicate in the CNS [8,9]. The transmigration of activated, circulating monocytes across the blood–brain barrier remains a critical mechanism for HIV neuroinvasion. This process is driven by chemotactic signals originating from brain parenchyma. Once in the CNS, HIV primarily targets the tissue-resident brain microglia, the key and major viral reservoir in the CNS, and possibly astrocytes [10,11], leading to a chronic inflammatory response [12–14]. This neuroinflammation leads to the activation of proinflammatory cytokines, such as TNF-α and IFN-γ, causing CNS cell death and neuronal damage [8,12–15]. Inflammatory cytokines play a crucial role in sustaining immune activation, with their levels directly correlating with the progression of HAND. This is evidenced by significantly elevated CSF levels of IP-10 (CXCL10) and MCP-1 (CCL2) in HAND patients [16–18]. Furthermore, monocyte activation appears to be a key driver of this inflammatory cascade [19]. Elevated levels of soluble CD14 (sCD14), a marker of monocyte activation, are found in the CSF of individuals with HAND compared to cognitively unimpaired counterparts [20–23], suggesting a direct link between activated monocytes and proinflammatory cytokine-driven pathology of neurocognitive decline. Additionally, the persistence of HIV infection could be due to a weaker immune response and limited ART drug penetration, even in those with effective ART suppression in the plasma [8,14,15].

The diagnosis of HAND remains challenging, especially in early stages, which impedes early treatment and facilitates disease progression. Consequently, developing strategies to identify early markers is vital for preventing neurocognitive impairment [24]. Established risk factors for HAND development include a history of severe immune dysfunction, most notably a low nadir CD4⁺ T cell count [25–27], as well as comorbid conditions like diabetes [28], arterial hypertension [29], and dyslipidemia [30]. Furthermore, neuronal damage in people with HIV (PWH) is driven by various metabolites and proteins that are significantly elevated in patients with HAND [31]. As a clear ultrafiltrate of blood, CSF envelops the CNS, where it provides protection and regulates neuronal function by maintaining interstitial fluid homeostasis. This intimate physiological relationship establishes CSF as a reliable source of biomarkers for the diagnosis and therapeutic monitoring of neurological disorders [32].

Here, we examined HIV viral loads in paired plasma and CSF samples collected from PWH on ART who had HAND. We then performed peptidomic analysis and viral outgrowth assays on CSF from these participants. In addition, we measured ART drug concentrations in both plasma and CSF to better understand the mechanisms underlying HIV persistence in the CNS despite suppressive ART in the plasma.

Methods

Participants

This cross-sectional study was conducted at the Institute of Infectious Diseases Emilio Ribas (IIDER) in Brazil from December 2014 to March 2016. A total of 24 PWH were recruited to assess the prevalence of HIV-associated dementia (Table 1). All participants provided written informed consent prior to data collection (CAEE: 44791921.0.1001.0068). For individuals with cognitive impairments affecting their capacity to consent, authorization was obtained from a legal guardian aged 18 or older. Demographic data were collected at the time of consent. Inclusion criteria comprised: a confirmed HIV diagnosis, regular follow-up at the IIER outpatient clinic, age ≥18 years, and a minimum of four years of formal education. Exclusion criteria for all participants included: active opportunistic neurological diseases (e.g. cerebral toxoplasmosis, neurotuberculosis, cryptococcal meningitis, progressive multifocal leukoencephalopathy); pre-existing neurological or metabolic conditions that could confound assessment (e.g. vascular dementia, diabetic neuropathy); use of psychoactive substances; an inability to comprehend the neurocognitive assessment battery; less than four years of schooling; age under 18 years; and pregnancy. This study was approved by the Research Ethics Committees of both the USP Medical School (659.363) and the IIER (1.327.226).

Table 1.

Baseline sociodemographic characteristics of study participants by cognitive status.

	HIV+
Variable	CN (n  =   10)	ANI (n  =   3)	MND (n  =   9)	HAD (n  =   2)	Total (n  =   24)	p Value
Agea	44.4  ±  8.6	45.0  ±  5.0	47.7  ±  11.2	57.0  ±  14.1	46.8  ±  9.8	0.4532
Sex, n (%)						0.4262
Female	9 (90%)	2 (66.7%)	7 (77.8%)	1 (50%)	19 (79.2%)
Male	1 (10%)	1 (33.3%)	2 (22.2%)	1 (50%)	5 (20.8%)
Educationa	11.8  ±  3.3	8.3  ±  3.8	11.6  ±  4.0	7.5  ±  4.9	10.9  ±  3.8	0.1868
Treatment Duration (Years) Mean ± SD	13.4  ±  5.0	13.3  ±  6.6	15.4  ±  4.1	13.2  ±  4.9	14.1  ±  4.6	0.8688
Efavirenz, n (%)
Yes	3 (30%)	2 (66.7%)	1 (77.8%)	0 (0%)	6 (25.0%)	0.2337
No	7 (70%)	1 (33.3%)	8 (22.2%)	2 (100%)	18 (75.0%)
Nadir CD4 count	403.9  ±  257.6	223.0  ±  125.5	273.6  ±  250.3	247.5  ±  48.8	319.5  ±  234.1	0.3737
Mean ± SD
CD4+ T (cells/mm3)	1019.1  ±  468.9	521.0  ±  176.6	755.7  ±  333.5	612.0  ±  55.2	824.1  ±  402.3	0.1691
Mean ± SD
CD8+ T (cells/mm3)	923.8  ±  328.9	1074.3  ±  528.5	836.3  ±  486.4	1064.5  ±  340.1	921.5  ±  402.0	0.6464
Mean ± SD
Plasma VL, n (%)
<20 copies/mL	9 (90%)	3 (100%)	8 (88.9%)	1 (50%)	21 (87.5%)	0.4886
>20 copies/mL	1 (10%)	0 (0%)	1 (11.1%)	1 (50%)	3 (12.5%)

Abbreviations: ANI, asymptomatic neurocognitive impairment; CN, cognitively normal; HAD, HIV-associated dementia; MND, mild neurocognitive disorder; NA, not available. Comparing the two groups uses χ2 for categorical variables or the Kruskal-Wallis test for continuous variables.

Diagnostic classification

Based on the Frascati criteria, HIV positive participants were categorized into the Cognitive Normal control (CN), Asymptomatic Neurocognitive Impairment (ANI), Mild Neurocognitive Disorder (MND), or HIV-Associated Dementia (HAD) [33]. CN individuals demonstrated normal performance on neuropsychological tests and reported no subjective cognitive complaints. Individuals with ANI showed objective cognitive impairment (performance ≥1 SD below demographically adjusted norms) in at least two domains, without any significant interference in everyday functioning. Those with MND met the same threshold for cognitive impairment as ANI but reported at least mild interference in daily activities. The HAD classification required marked cognitive impairment (performance ≥2 SD below norms in at least two domains) that resulted in significant interference with day-to-day functioning [34]. All neuropsychological assessments were conducted by the same specialist neuropsychologist who was blinded to the participants’ HIV status.

CSF and plasma collection

∼ 500 μL CSF samples were obtained by lumbar puncture without anticoagulant, and peripheral blood samples were collected with the addition of ethylenediaminetetraacetic acid (K3 EDTA 7.5%). To preserve biomarker stability, samples were centrifuged at 1,500 rpm for 5 min, aliquoted, and frozen at – 80°C within two hours of collection. Subsequent transport to the UNC HIV Cure Center (University of North Carolina at Chapel Hill, USA) was performed using liquid nitrogen to maintain a continuous cold chain.

Reverse transcription-droplet digital PCR (RT-ddPCR)

Viral RNA was extracted from 200 μL plasma and 200 μL CSF samples using the QIAamp Viral RNA Mini Kit (Qiagen). A total of 20 μL of RNA was treated with DNase I to remove contaminating genomic DNA. Complementary DNA (cDNA) was synthesized from the purified RNA using the SuperScript IV First-Strand Synthesis System (Invitrogen). HIV RNA was quantified via RT-ddPCR with primers/probe sets specific to the HIV gag regions. Primer 1: 5’-TACTGACGCTCTCGCACC3’, Primer 2: 5’-TCTCGACGCAGGACTCG-3’, and Probe 5’-FAM-CTCTCTCCTTCTAGCCTCCGCTAGT-3’. The PCR cycling conditions were as follows: initial denaturation at 95°C for 10 min; 45 cycles of denaturation at 94°C for 30 sec and annealing/extension at 57°C for 60 sec; and a final droplet stabilization step at 98°C for 10 min. Droplets were analysed on a QX200 Droplet Reader using QuantaSoft software (Bio-Rad) in absolute quantification mode. We used plasma and cerebrospinal fluid samples from HIV-negative donors as a negative control. Wells containing fewer than 10,000 accepted droplets were excluded from the analysis to ensure data quality.

Liquid chromatography-tandem mass spectrometry (LC-MS/MS) to analyze antiretroviral (ARV) drugs

Analytes were extracted from plasma and CSF samples by protein precipitation with methanol containing stable, isotopically-labelled internal standards (13C,2H5DTG, 2H9-DRV, 2H6-RTV, 2H5-EFV, 13C5-TFV, and 13C,15N2-3TC). Following extraction, samples were analysed by LC-MS/MS. Analytes were separated from matrix components using reverse-phase chromatography followed by detection on an AB Sciex API-5000 triple quadrupole mass spectrometer. Lower limits of quantitation for the analytes were 50 ng/mL (DRV, darunavir; RTV, ritonavir; EFV, efavirenz), 30 ng/mL (DTG, dolutegravir), and 1 ng/mL (TDF, tenofovir disoproxil fumarate; 3TC, lamivudine) in plasma, with lower limits of quantitation of 5 ng/mL (DTG) and 1 ng/mL (DRV, RTV, EFV, 3TC, TDF) in CSF. Calibration standards and QC samples met the 15% acceptance criteria in these analyses.

In vitro infection of CD8-depleted PHA blasts and Molt-4/CCR5 cells with CSF

CD8-depleted phytohemagglutinin (PHA) blasts (1 × 10⁶) or Molt-4/C–C Chemokine Receptor Type 5 (CCR5) cells (5 × 10⁵) were spinoculated (1,200 × g for 2 h at room temperature) with 250 µL of CSF from PWH. Phosphate-buffered saline (PBS) was used as a mock control. The total volume was adjusted to 500 µL with complete medium. Following spinoculation, the cells were washed once with sterile PBS and resuspended in 2 mL of complete R10 medium. The R10 medium consisted of RPMI 1640 (Gibco) supplemented with 10% FBS (Avantor), 1% penicillin–streptomycin (Gibco), 10 mM HEPES (Gibco), 2 mM L-glutamine (Gibco), and 1 mM sodium pyruvate (Gibco). For CD8-depleted PHA blasts, the medium was supplemented with 40 U/mL of recombinant IL-2 (PeproTech) in addition to the standard medium. Cells were plated in a 24-well plate and cultured at 37 °C. Culture supernatants were collected at the indicated time points, with fresh medium replenished at each collection. Viral RNA was isolated from 400 µL of supernatants using the QIAamp Viral RNA Mini Kit (Qiagen). HIV gag RNA levels were then quantified by RT-ddPCR, as described above.

Peptidome analysis

Three millilitres of CSF were incubated at 80°C for 20 min to inactivate proteases. The samples were then cooled on ice for 30 min, acidified to pH 2 with 0.1 M HCl (Millipore, Burlington, MA, USA), and centrifuged at 3,000 ×g for 30 min at 4°C. The supernatants were filtered using Amicon Ultra-4 centrifugal units (cut-off of 10,000, Millipore) to remove proteins with a molecular mass above 10 kDa. The filtrates were adjusted to pH 2–4 with 50% formic acid and purified using C18 OASIS chromatography columns (Waters, Inc., São Paulo, Brazil). The peptides were resuspended in ultrapure water, and their concentrations were measured at 214 nm using a NanoDrop spectrophotometer (Thermo Fisher, Waltham, MA, USA).

Then, five micrograms of the peptide extracts were diluted in 100 µL of 100 mM TEAB buffer (Sigma-Aldrich, St. Louis, USA). Isotopic forms of formaldehyde (4%) and sodium cyanoborohydride (24 mM) were added, and the samples were incubated for 16 h at room temperature. The dimethyl labelling method used four isotopic forms, introducing mass shifts of 28, 30, 32, or 36 Da per labelling site. The reactions were stopped with ammonium bicarbonate (0.16%) and acidified with formic acid (0.4%). The samples were pooled, purified on C18 OASIS columns (Waters), and 0.2 µg of peptides diluted in 5 µL of 100% acetonitrile with 0.15% formic acid were injected for mass spectrometry analysis. These analyses were performed on an Orbitrap Fusion Lumos spectrometer (Thermo Fisher) coupled to nano-LC. The peptides were separated on a C18 column of 360 µm OD × 150 mm with 3 µm spheres, using a 5–40% acetonitrile gradient in 0.1% formic acid for 120 min at 250 nL/min. Data acquisition parameters included a resolution of 120,000 for full MS, an m/z range of 300–1500, and dynamic exclusion of 60 s. The data were analysed using Xcalibur and Mascot software, with search parameters set to no enzymatic specificity, a precursor mass tolerance of ±0.5 Da, and fragment ion mass tolerance of ±0.5 Da. Peptides showing ≥100% increase or ≥50% decrease compared to controls were considered significant.

Peptidome trend analysis across HAND stages

To assess peptide dynamics across disease progression, peptides were grouped by directional change across Control → ANI → MND → HAD. Trends were defined as increasing (≥1.5× up), decreasing (≤0.67× down), or stable (±0.2× around baseline) using normalized abundance ratios. These trajectories were visualized as (i) line plots showing per-peptide behaviour, and (ii) Sankey diagrams mapping gene → peptide → clinical fate using the networkD3 and ggalluvial R packages. Peptides showing consistent monotonic shifts across independent runs were prioritized.

Measurement of immune and biochemical parameters in the CSF

Cerebrospinal samples were obtained by standard lumbar puncture, performed at the Cerebrospinal Fluid Laboratory, linked to the Central Laboratory Division of the Hospital das Clínicas, Faculty of Medicine, University of São Paulo (HCFMUSP). After collecting, all samples were submitted to routine cytochemical examination at the HCFMUSP Central Laboratory. Total protein quantification (mg/dL) was performed using an automated colorimetric method on a Cobas c501 analyzer (Roche Diagnostics, Basel, Switzerland). The CSF protein profile, which included pre-albumin, albumin, α₁-globulin, α₂-globulin, β-globulins, and γ-globulins, was assessed by high-resolution capillary electrophoresis using the Capillarys 3 TERA system (Sebia, Lisses, France). CSF cellular analysis was performed by automated flow cytometry on a FACSCanto II (BD Biosciences, San Jose, CA, USA), allowing total leukocyte counting and determination of the proportion of lymphocytes, monocytes, and granulocytes. Red blood cell counts (cells/mm³) were also performed to identify potential blood contamination. More specific cell populations, such as plasma cells, basophils, or eosinophils, were only evaluated when samples were previously labelled with fluorochrome-conjugated antibodies.

Measurement of tau proteins

Total tau (b-tau) and phosphorylated tau (p-tau) levels in CSF were quantified using a validated sandwich enzyme-linked immunosorbent assay (ELISA) (INNOTEST assays, Fujirebio, Ghent, Belgium). Absorbance values were obtained using a microplate reader with a 450 nm filter (FilterMax F5, Molecular Devices, San Jose, CA, USA), and concentrations (pg/mL) were calculated from standard curves generated for each assay.

Data visualization and integration

All figures were generated in R v4.3.2 using ggplot2, pheatmap, and ggalluvial. Heatmaps represent scaled Z-scores of peptide intensities, and circle/Sankey plots depict weighted interactions normalized to total peptide flux per condition. Clinical and proteomic datasets were integrated using sample-matched identification codes (IDs), enabling cross-referencing between HIV peptide remodelling and β-tau-linked neurodegeneration.

Statistical analysis

Data on clinical and laboratory parameters were retrieved from electronic medical records in the HCMED system (Hospital das Clínicas, FMUSP). All statistical analyses were carried out with GraphPad Prism 10 (La Jolla, CA, USA). The Shapiro–Wilk test was used to determine the normality of data distribution. Based on this assessment, differences between the cognitive groups (CN, ANI, MND, HAD) were analysed using the Kruskal–Wallis test for non-normally distributed continuous variables and the chi-squared (χ²) test for categorical variables. For the comparison of paired ARV concentrations in CSF and plasma, the Wilcoxon Signed-Rank Test was employed. Statistical significance was defined as a two-tailed p-value < 0.05. For CSF Biochemical and Immune Profiling, data were normalized to control medians and visualized using ggplot2 (v3.4) in R. Statistical comparisons were performed using Kruskal – Wallis ANOVA with Benjamini – Hochberg FDR correction (q < 0.05). Correlations between peptide abundance and CSF biomarkers were assessed using Spearman’s ρ.

Results

Cohort characteristics

The baseline sociodemographic and clinical characteristics of the 24 study participants, stratified by cognitive status, are presented in Table 1. The cohort consisted of 10 CN individuals, 3 with ANI, 9 with MND, and 2 with HAD. The cohort had a mean age of 46.8 years (SD = 9.8), and 79.2% of participants were women. The overall cohort had a mean age of 46.8 ± 9.8 years, was predominantly female (79.2%), and had a mean of 10.9 ± 3.8 years of education. While no statistically significant differences were found across the groups for any variable (all *p* > 0.05), several non-significant trends were observed. The HAD group was, on average, older (57.0 ± 14.1 years) and had fewer years of education (7.5 ± 4.9 years) compared to the other groups. Immunologically, the CN group exhibited the highest mean nadir CD4 count (403.9 ± 257.6 cells/mm³) and current CD4⁺ T count (1019.1 ± 468.9 cells/mm³), whereas ANI and HAD groups trended to have a higher CD4 count, but no statistical difference was reached (Table 1). Metabolic profile was also measured (Supplementary Table 1). In summary, the groups were well-matched at baseline with no significant demographic or clinical differences, though trends suggested that more severe cognitive impairment may be associated with older age, lower education, and a history of greater immunosuppression.

HIV RNA detection in CSF despite plasma ART suppression

Despite undetectable plasma HIV Gag viral RNA RT-ddPCR in all participants, a finding consistent with excellent ART adherence (Figure 1A), viral RNA was identified in the CSF of individuals with HAND although the copied of HIV RNA varied among samples (Figure 1B). This dissociation suggests compartmentalized HIV replication within the CSF despite systemic viral suppression and/or possible suboptimal penetration of ART drugs into the CNS. The persistence of HIV RNA in the CSF underscores the CNS's role as a possible viral reservoir. This sustained viral presence may contribute to ongoing production of viral proteins and inflammatory responses, potentially driving the neurodegeneration and cognitive decline characteristic of HAND.

Figure 1.

HIV RNA is detectable in CSF from people living with HIV despite effective ART in the plasma. (A-B) HIV Gag RNA levels in the plasma and the CSF were measured by RT-ddPCR.

HIV peptidome dynamics across clinical stages

Detection of HIV transcripts indicates possible ongoing viral replication in the cerebrospinal fluid (CSF). If this is the case, HIV particles could be produced. Therefore, we sought to detect HIV proteins in the CSF using peptidomic analysis, which offers higher sensitivity for identifying trace amounts of proteins compared to conventional proteomic approaches [35]. We first mapped the distribution of HIV-derived peptides across viral genes to determine their relative contribution to the CSF peptidome. Pol-derived peptides dominated the landscape (∼42.5%), followed by Env (∼27.5%) and Gag (∼15%), while Nef, Vpr, Vpu, and Protease contributed fewer peptides (Figure 2A). Hierarchical clustering of peptide sequences further highlighted the enrichment of Pol and Env regions, with distinct peptide clusters reflecting viral protein diversity (Figure 2B). A Mascot score was calculated to determine the significance of viral peptides matching to HIV protein (peptide) sequences [36]. Distribution of Mascot scores across viral genes confirmed consistently higher confidence values for Pol, Env, and Gag, whereas accessory proteins (Vpr, Vpu, Nef) and Protease scored lower, consistent with their limited peptide representation (Figure 2C). Together, these data demonstrate that Pol, Env, and Gag represent the most consistently detectable and conserved HIV peptide sources in CSF.

Figure 2.

Distribution and clinical dynamics of HIV-derived peptides in CSF peptidomics. (A) A donut distribution of HIV peptides across viral genes. (B) Heatmap of peptide sequences mapped to viral genes, confirming dominance of Pol and Env regions. (C) Violin plots of Mascot scores per gene, indicating higher confidence identifications for Env, Pol, and Gag peptides, while Vpr, Nef, Vpu, and Protease showed lower scores. (D) Relative abundance heatmap of selected HIV peptides across clinical stages (ANI, MND, and HAD).

Next, we assessed the dynamics of peptide abundance across clinical stages of HIV-associated neurocognitive impairment (ANI, MND, HAD). Heatmap visualization showed heterogeneous detection of HIV peptides, with some peptides stable across stages while others declined or increased relative to controls (Figure 2D). Notably, the Pol peptide DEQLCLDFV exhibited a marked increase in abundance during HAND progression, while multiple Env – and Gag-derived peptides (e.g. RPGSGSK, RAGCNLNGT) decreased, particularly at MND and HAD. Several peptides, including IGGGRTLHTTRAITG (Env) and QYDQVPIEICGHKAIGIVL (Pol), remained stable across clinical stages, suggesting conserved persistence.

Divergent trajectories of HIV-derived peptides across clinical progression

Overall trajectories of peptide abundance across clinical stages revealed that progression from ANI to HAD is characterized by divergent peptide dynamics rather than uniform suppression. Certain peptides, such as DEQLCLDFV (Pol), exhibit a gradual increase in abundance, suggesting ongoing viral activity even under ART. In contrast, peptides including AGAGNSK (Gag), RAGCNLNGT (Env), and RPGSGSK (Gag) display marked declines, particularly at the MND or HAD stages, implying progressive immune clearance or peptide loss. Meanwhile, stable peptides (IGGGRTLHTTRAITG from Env and QYDQVPIEICGHKAIGIVL from Pol) remain unchanged, representing conserved regions likely persisting in the CSF despite clinical worsening (Figure 3A).

Figure 3.

Divergent trajectories of HIV-derived peptides across clinical progression. (A) Overall peptide trajectories across clinical stages (ANI → MND → HAD) showing abundance dynamics. (B) Individual peptide trend classification separates peptides into three major groups: decreasing (red), increasing (green), and stable (blue). (C) Gene – peptide – clinical trajectory mapping peptide origin (gene) to its directional trend across clinical stages.

Dissection of these patterns (Figure 3A) into three functional groups: decreasing (red) peptides showing consistent decline across stages, increasing (green) peptides predominantly Pol-linked, indicating potential low-level replication or reactivation, and stable (blue) peptides that remain invariant, highlighting structurally or immunologically conserved peptides. These distinct trajectories collectively point to selective viral peptide remodelling during neurocognitive progression under ART suppression. Gene – peptide – clinical trend relationships of peptides derived from the Env, Gag, and Pol genes dominate, while those of Protease contribute minimally (Figure 3B). The dominant red flow corresponds to declining peptides across stages, marking regions prone to immune recognition or degradation. A smaller green flow links to Pol-derived peptides, identifying Pol as a potential active reservoir signature, while stable blue flows mark conserved Env/Pol peptides that could serve as biomarker candidates for chronic viral persistence in the CNS.

CSF immune and biochemical parameters in PWH on ART

To evaluate whether peptide-level alterations correspond with changes in CSF composition, we examined total protein fractions, β-tau protein, glucose levels, and immune cell populations across Control, ANI, and HAND stages (Figure 4A–N and Supplementary Table 2). To improve statistical significance and understanding the changes during HAND, we merged MND with HAD (HAND); the two conditions (ANI and HAND) were then compared to controls.

Figure 4.

CSF protein and immune cell alterations across HAND progression. (A – B) CSF α−1 and α−2 globulin fractions across Control, ANI, and HAND stages. (C-E) Levels of β-tau proteins, γ-globulin, and CSF glucose across Control, ANI, and HAND stages. (F-L) Percentage of lymphocytes, monocytes, neutrophils, basophils, plasmacytes, eosinophils, and macrophages across Control, ANI, and HAND stages. (M – N) CSF cell count of total cells and red blood cells across Control, ANI, and HAND stages.

Distinct clinical trends emerged, linking β-tau proteins to neurodegenerative signatures [37] and peptide remodelling. Specifically, β-tau levels (Figure 4C) were significantly reduced in ANI compared with controls but partially stabilized in HAND. α−1 and α−2 globulin fractions (Figure 4A–B) displayed modest variation across stages, while γ-globulins (Figure 4D) had a trend to increase in ANI, consistent with heightened intrathecal immune activity. Glucose levels (Figure 4E) and lymphocyte percentages (Figure 4F) declined in parallel with β-tau, but did not reach statistical significance. By contrast, monocytes (Figure 4G) and neutrophils (Figure 4H) increased in HAND, pointing to monocyte-driven inflammatory infiltration and sustained neuroimmune activation. Minor immune subsets, including basophils (Figure 4I), plasmocytes (Figure 4J), eosinophils (Figure 4K), and macrophages (Figure 4L), remained largely unchanged. Total CSF cell counts (Figure 4M) rose gradually with disease severity, while red blood cells (Figure 4N) remained low, indicating minimal contamination.

ARV penetration into the CSF

Despite achieving ART-mediated suppression of HIV in plasma, both HIV RNA and peptides were detected in the CSF samples. To investigate whether limited drug penetration could account for this virological discordance, we quantified ARV concentrations in paired plasma and CSF samples from 11 PWH across different HAND stages using LC-MS/MS. The participants’ regimens were as follows: seven were on TDF/DTG/3TC, three on TDF/3TC/EFV, and one on DTG/DRV/RTV. The CNS penetration-effectiveness (CPE) score for these regimens, based on a previous report [38], ranged from 6 to 8. Our pharmacological analysis revealed distinct drug partitioning patterns between plasma and CSF. TFV and 3TC concentrations were significantly higher in plasma than in CSF, whereas DTG showed a non-significant trend for higher CSF concentration (p = 0.13). No significant difference was found for EFV between compartments. Insufficient data precluded statistical analysis for DRV and RTV (Figure 5A). These distribution patterns resulted in high CSF-to-plasma penetration ratios for TFV, 3TC, and DTG, while EFV, DRV, and RTV demonstrated substantially lower penetration ratios (Figure 5B). Analysis of absolute CSF concentrations showed the following means and ranges: EFV (13.8, 9–22 ng/mL), TDF (1.6, 0–5.1 ng/mL), 3TC (78, 8–212 ng/mL), and DTG (5.2, 0–19.5 ng/mL), with a single DRV measurement at 20.4 ng/mL. According to previous reports [39–43], the concentrations of DTG, 3TC, DRV, and TDF in CSF, although lower than in plasma, exceed the in vitro IC50 for wild-type HIV. This indicates that the ARV levels in the CSF should be sufficient for viral suppression.

Figure 5.

CSF and plasma concentrations of different ARV drugs. (A) Quantitation of ARV drugs (EFV, DRV, RTV, TFV, 3TC, and DTG) in plasma and CSF using LC-MS/MS. (B) Quantitation of 3TC in CSF. (C) The CSF/plasma 3TC ratio in HIV positive patients with CN, ANI, MND, or HAD.

Given the relatively high absolute levels of 3TC penetration detected in the CSF, we investigated whether its concentration varied with different manifestations of HAND. However, no significant difference in CSF 3TC concentration was observed across HAND groups (Figure 5C). Further, the CSF-to-plasma penetration ratio of 3TC did not differ significantly among the groups (Figure 5D), indicating that the severity of neurocognitive impairment in our cohort might not be influenced by how much 3TC penetrates the CSF.

Failure to produce replication-complete HIV despite detection of HIV RNA and peptides in the CSF

We sought to determine whether HIV present in the CSF of individuals with HAND is capable of infecting phytohemagglutinin (PHA) activated cells, likely enriched for activated CD4⁺ T cells, with the depletion of CD8⁺ T cells to produce the replication-competent HIV in the CSF. Using an equal volume of CSF (250 μL) from eight HAND donors, we spinoculated CSF HIV into CD8-depleted PHA blasts. However, we observed no exponential viral replication in cells infected with CSF and mock-infected controls (Figure 6A). Although low-level outgrowth was detected on day 5 post-infection in cultures infected with CSF from donors AD1355C and AD1373B, viral production did not increase subsequently but declined. As discovered in Figure 1(B), these two donors had the highest CSF viral loads, which may explain the transient detection of the viral outgrowth. To further confirm these results, we used the HIV highly permissive Molt4/CCR5 cell line. A similar result was observed where CSF-derived virus from HAND donors was unable to establish a productive infection in Molt4/CCR5 cells (Figure 6B). Collectively, these findings indicate that HIV in the CSF collected from individuals with HAND on ART is defective and has limited infection capability. These observations further support that ART, although it has less penetration into CSF, remains effective in controlling HIV replication in the CNS. Under ART, HIV in the CSF is in the latent or quiescent stage of infection, which may be associated with neuroinflammation and neuronal damage.

Figure 6.

Failure of HIV from CSF to establish a spreading infection in T cells. (A-B) CSF samples from ANI, MND, or HAD were used to spinoculate CD8-depleted PHA blasts and MOLT-4/CCR5 cells. HIV gag RNA in the culture supernatant was measured by RT-ddPCR.

Discussion

This study provides a multifaceted investigation into the virological and pharmacological factors that may underpin the persistence of HAND in the era of suppressive ART. Our principal findings are fourfold: (1) we confirm the presence of HIV RNA in the CSF of individuals with HAND despite undetectable plasma viral loads; (2) our peptidome analysis detects many HIV peptides in the CSF of HAND patients. (3) we demonstrate heterogeneous penetration of antiretroviral drugs into the CNS, with some agents achieving significantly lower concentrations in the CSF but within the ARV IC50 ranges to effectively suppress HIV replication; and (4) we show that virus found in the CSF is unable to establish a productive infection in highly permissive immune cells in vitro. Collectively, these findings suggest that CSF serves as a sanctuary site for HIV, not primarily through robust, replication-competent virus, but rather through a combination of persistent HIV RNA and peptides, likely driving an inflammatory response that leads to neuronal damage in the brain of HAND.

The detection of HIV RNA in the CSF despite undetectable plasma levels defines the virological discordance observed in our cohort, consistent with prior reports [4–6]. In addition to HIV RNA transcripts, it was somewhat unexpected to detect numerous HIV-derived peptides in the CSF of participants on suppressive ART, enabled by advances in peptidomic analysis. This may be because peptidomics can identify low-abundance, naturally processed, pathogen-derived peptides that are not easily traceable to their parent proteins in traditional proteomic workflows [35]. Pol, Env, and Gag represent the most consistently detectable and conserved HIV peptide sources in CSF. At the same time, Protease contributes minimally to the peptidome, possibly making Pol, Env, and Gag peptides prone to trigger immune activation in the CSF. Unexpectedly, p-β-tau was decreased in ANI but restored in HAND. This is consistent with a previous report [44], in which a depressed p-Tau181 was observed in the CSF, distinguishing PWH with HAND from patients with dementia of Alzheimer's disease, which suggests that reduction of p-Tau may be an early biomarker of neuronal compromise in ART-suppressed individuals. Nevertheless, due to the small cohort of this study, the ANI β-tau reduction should be viewed as a preliminary, early-stage signal that warrants validation in a larger cohort.

While such compartmentalized viral activity has been suggested because of suboptimal ARV penetration into the CNS [45,46], our pharmacological analysis does not support this hypothesis for the regimens studied in this cohort. Quantification of CSF ARV concentrations revealed levels that, for key drugs such as DRV, DTG, 3TC, and TDF, all meet or exceed the concentrations required for in vitro suppression [39–41,43]. This is consistent with studies showing that a higher CPE score is not always predictive of neurocognitive improvement [46,47]. Collectively, these data indicate that the virus in the CSF is exposed to ARV concentrations that should be sufficient to suppress replication, even though ARVs in the CSF are significantly lower than in plasma. This suggests that mechanisms other than insufficient drug exposure – such as local viral resistance or reservoir dynamics – are likely responsible for the observed CSF viral escape [48], arguing against inefficient penetration of current ART drugs in the CSF of PWH.

The detection of HIV RNA and peptidome suggests possible ongoing viral replication in the CSF on ART. Despite this, the virus from the CSF of ART-suppressed individuals was incapable of establishing a productive infection in a standard viral outgrowth assay, which hasn’t been studied before. This suggests that the viral RNA may originate from cellular populations producing defective virions, or from the release of viral transcripts in the absence of functional HIV protein synthesis and/or full viral assembly. Or, even if some viral particles are produced, CNS immune surveillance (microglia, resident macrophages, antibodies, or complement) might neutralize them before they can infect target cells. Lastly, the viral particles might be immature, poorly processed, or non-infectious due to incomplete Gag or Env maturation. Together, despite the presence of HIV RNA and viral peptides, HIV in the CSF is likely defective or latently infected in CNS-resident cells – producing viral products but not infectious virions in ART-suppressed PWH.

With single-cell RNA sequencing (scRNA-seq) analysis, it has been shown that there exist residual microglia/macrophages, or CNS T cells in the CSF on ART, which may contain HIV transcripts [49–52]. These CSF cells could be a potential cellular source of latent HIV. However, neither our data nor scRNA-seq analyses are sufficient to provide direct evidence in defining the full length of replication-competent HIV in the CSF of PWH on ART. Thus, it remains underdetermined whether these extremely low levels of CSF cells serve as the source of viral rebound, in addition to brain microglia and, to a lesser extent, CNS T cells or astrocytes in the brain parenchyma [10]. Future study on CSF cells is needed to uncover this enigma. Despite that, HIV RNA transcripts and peptidome may be associated with neuroinflammation, as we recently reported in the same cohort from the Brazilian population [23] (Ref 21, in press, DOI: 10.1097/QAD.0000000000004368), in which sCD14, IFN-γ, IP10, IL-6, IL-1β, and MCP-1 in the CSF were significantly increased in patients with ANI, MND, and/or HAD, compared with HAND-negative controls.

Our study has several limitations. First, baseline CNS/CSF samples obtained prior to ART initiation are not available for this cohort, as lumbar puncture was performed only after sustained viral suppression to minimize clinical risk and adhere to ethical guidelines. Therefore, a pre-ART positive control could not be included. Second, the modest sample size for the pharmacological (n = 11) and in vitro infectivity (n = 8) analyses limited the statistical power to detect associations between variables such as ARV concentrations and HAND severity. Third, we could not fully assess the impact of efavirenz (EFV), which was administered at a 50% higher dose than the standard dose (400 mg/day instead of 600 mg/day) as first-line ART in Brazil from 1998 to 2014, and is associated with neurological symptoms such as nightmares and panic attacks [53]. With only three patients on an EFV-based regimen, our study was underpowered to evaluate its relationship with HAND. Fourth, while our peptidomic findings are promising, they require validation in a larger, independent cohort to confirm their utility in the discovery of biomarkers. Also, while other HIV proteins, such as Nef and Rev [54,55], have been detected in the CSF and may play important roles in CSF immune activation, none of these proteins were able to be captured by peptidomics, limiting our further analyses. Future studies with expanded sample sizes are needed to substantiate these observations and explore the impact of variables like ART regimens, infection duration, and host genetics.

In conclusion, HIV continues to produce RNA transcripts and proteins in the CSF despite effective ART. However, CSF-derived HIV fails to establish new infections in T cells, suggesting that it is unlikely to represent active residual replication. Instead, these findings point to defective or latent proviruses within CSF-resident immune cells as the source. The chronic production of viral RNA and proteins in the CSF, along with the resulting inflammatory cascade, rather than the generation of new, replication-competent viruses, likely contributes to neuroinflammation and neurocognitive dysfunction in PWH with HAND receiving effective ART.

Funding Statement

G.J. is supported by NIAID (R21 AI167709-01A1 and R01AI186609), NIMH (R21MH128034 and R01MH136852), Collaborative Development Program at B-HIVE (U54AI170855), and NIAID/CARE (1UM1AI164567). G.P. is supported by FAPESP (2019/25511-4 and 2022/06636-3). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Author contributions

G.J. and J.C. conceived and designed the study. G.S. and S.T. recruited participants and collected clinical samples. M.G. and Carolina Gualqui performed neuropsychological screening. V.F., M.M., and C.A. managed clinical data. G.S., R.E., and H.G. performed peptidome assays. G.P. and N.K. performed ddPCR and viral outgrowth assays. X.L. conducted viral outgrowth assays and prepared samples for ARV analysis. Y.T. assisted with the viral outgrowth assay and data analysis. Cassandra Gilbert performed LC-MS/MS analysis. G.P., J.C., and X.L. organized the initial data. C.S.S. analysed the peptome data. G.P., J.C., V.F., X.L., and G.J. prepared the initial manuscript draft. G.J. finalized the data analysis and completed the manuscript for submission. All authors approved the manuscript for publication.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability

All the data have been included in this study.

Standard protocol approvals, registrations, and patient consents

This study was approved by the Research Ethical Standards Committees of both the USP Medical School (659.363) and the IIER (1.327.226). Written informed consent was obtained from all participants in the study.

Supplemental Material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/22221751.2026.2616945.