Single-nucleus detection of rare HIV-infected cells defines the cellular landscape of HIV persistence in the human brain

Single-nucleus detection of rare HIV-infected cells defines the cellular landscape of HIV persistence in the human brain | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Method Article Single-nucleus detection of rare HIV-infected cells defines the cellular landscape of HIV persistence in the human brain Yuan-yuan Cai, Nicholas Jacobs, Dana Gabuzda, Michael Corley, and 9 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9013592/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 8 You are reading this latest preprint version Abstract Background HIV-1 enters the central nervous system early after infection and establishes a long-lived reservoir that persists despite antiretroviral therapy. Single-cell and single-nucleus RNA sequencing provide powerful approaches to study HIV infection in the human brain, yet standardized and sensitive methods for identifying rare HIV-infected cells in these datasets remain limited. Here, we present a scalable multi-reference framework for detecting HIV RNA–positive cells in human CNS single-nucleus RNA-seq data. The pipeline integrates a modified HIV reference genome, subject-specific variant-updated HIV references, and a comprehensive HIV strain collection to improve viral read recovery and specificity. Results We applied this framework to 250 post-mortem brain samples from the SCORCH (Single Cell Opioid Responses in the Context of HIV) consortium spanning 12 brain regions and 102 donors, including people with and without HIV (PWH and PWoH). After screening, 48 samples from 35 donors comprising 559,207 high-quality nuclei were analyzed in depth. We identified 1,939 HIV RNA–positive cells exclusively in samples from PWH. Using conservative thresholds, 908 high-confidence infected cells were retained for downstream analyses. HIV RNA-positive cells were rare overall and strongly enriched in cases with HIV encephalitis. Microglia constituted the predominant infected population (79% of HIV RNA-positive cells), with substantially smaller contributions from oligodendrocytes, astrocytes, and neurons. In non-encephalitic brains, detectable infection was largely restricted to microglia, whereas in encephalitic tissue HIV RNA–positive cells were distributed across multiple CNS (Central Nervous System) lineages. Viral RNA burden followed a long-tailed distribution, with microglia retaining higher HIV transcript counts than other cell types. Recovered HIV reads were concentrated in the U3 region of the 5′ LTR and in the env gene, implicating regulatory and entry-associated regions as focal points of viral diversity in the brain. Conclusions Together, these data establish a harmonized framework for identifying rare HIV-infected cells in CNS single-cell datasets and provide large-scale quantitative evidence that microglia represent the dominant and most persistent HIV-infected population in the human brain. This work offers a reference strategy and resource for future NeuroHIV studies aimed at defining, monitoring, and ultimately targeting CNS viral reservoirs. HIV-1 single-nucleus RNA-seq brain microglia viral read mapping reference genome NeuroHIV HIV encephalitis Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 BACKGROUND Human immunodeficiency virus type 1 (HIV-1) remains a major global health challenge. HIV infection affects both mental and physical health, placing substantial and long-term burdens on people living with HIV, their families, and health systems. In 2024, an estimated 40.8 million people were living with HIV worldwide, and only 31.6 million were receiving antiretroviral therapy (ART), with ART leading to clinical viral suppression in 95% [ 1 ]. Approximately 630,000 people died from AIDS-related illnesses in 2024, compared with 2.1 million in 2004 and 1.4 million in 2010, underscoring both the success and the limitations of current treatment [ 1 ]. As ART has extended survival, more people are living for decades with chronic complications of infection, including HIV-associated neurocognitive disorders (HAND), which range from mild cognitive impairment to severe dementia [ 2 ]. Another term, HIV-associated brain injury (HABI) has been proposed recently to help distinguish legacy damage prior to receiving ART from active damage occurring despite efficacious ART [ 3 ]. Regardless of these concepts, the study of HIV in the brain and its effects are collectively known as NeuroHIV. HIV enters the central nervous system (CNS) early during infection and establishes long-lived reservoirs in brain myeloid cells and other cell types, where it can persist despite systemic viral suppression [ 4 ]. Understanding which CNS cell types harbor virus, how infected cells interact with their local microenvironment, and how these interactions contribute to neurocognitive disorders is essential for developing more effective therapeutic strategies for treating such disorders and for strategies towards a cure for HIV. The emergence of single-cell sequencing technologies has opened new possibilities for dissecting these questions in human brain tissue. Unlike bulk RNA sequencing, which averages gene expression across heterogeneous cell populations, single-cell and single-nucleus RNA sequencing measure mRNA at the level of individual cells, providing the resolution needed to study how specific cell types in the brain are infected by, and/or respond to, HIV infection. However, detecting HIV-infected cells in these datasets is technically challenging. Single-cell and single-nucleus experiments capture only a fraction of the transcriptome, on the order of ~ 30% of mRNA molecules per cell in typical droplet-based experiments, and thus provide only a snapshot of the transcripts present at the time of sampling [ 5 , 6 ]. Single-nucleus sequencing, which is widely used for human post-mortem brain, captures even fewer transcripts than whole-cell approaches and yields a higher proportion of intronic and ribosomal reads. In addition, effective ART suppresses viral replication and reduces viral RNA levels in tissues, so that even truly infected cells may express very low amounts of HIV transcripts [ 7 ], and in the case of true latently infected cells no viral transcripts. These factors, together with the rarity of infected cells and low viral copy numbers, make sensitive and specific detection of HIV-infected cells in CNS single-cell datasets particularly difficult and highlight the need for improved analytical strategies. Single-cell and single-nucleus approaches are increasingly used in NeuroHIV research, but most studies have focused on host gene-expression changes associated with HIV status, neuropathology, substance use, or ART exposure, rather than on systematic detection of HIV-positive cells [ 8 – 10 ]. Existing reports of HIV-positive cells at single-cell resolution are typically limited to small cohorts, single brain regions, or cerebrospinal fluid, and often rely on study-specific reference genomes, alignment settings, and ad hoc thresholds for calling infected cells [ 4 , 8 , 9 ]. As a result, there is currently no standardized, scalable framework for identifying HIV-infected cells across large, heterogeneous CNS single-cell datasets. To begin addressing this need, the Single Cell Opioid Responses in the Context of HIV (SCORCH) consortium was established in 2020 with support from the National Institute on Drug Abuse (NIDA). SCORCH aims to generate resources for comprehensive brain tissue characterization at the single-cell level, with a particular focus on single-nucleus transcriptomic data to identify novel rare cell types and to enrich key cell populations, thereby refining our understanding of CNS pathophysiology associated with substance use disorders and HIV. SCORCH is an interdisciplinary effort that brings together fourteen data generating projects, with investigator teams led by principal investigators from nine institutions, along with a Data Coordination Center [ 11 ]. Most of the human patient samples, and all samples from people with HIV (PWH) provided to SCORCH investigators for single-nucleus sequencing, are provided by the National NeuroHIV Tissue Consortium (NNTC). The NNTC was established in 1998 to collect, store, and distribute central and peripheral nervous system tissue, cerebrospinal fluid, blood, and other organs from PWH who have been characterized in longitudinal neuromedical, neuropsychological, and other assessments. This unique combination of rich clinical metadata and high-quality frozen brain tissue has made the NNTC an essential resource for investigating the neurological and neuropsychiatric complications of HIV [ 12 – 14 ]. To address the gap in standardized viral detection for CNS single-cell datasets, we developed and applied a viral-detection framework for human NeuroHIV single-cell and single-nucleus RNA-seq data that combines modified and variant-updated HIV reference genomes, along with a collection of HIV strains, and explicit criteria for defining HIV RNA–positive cells. Using single-nucleus RNA-seq data generated by SCORCH from a large NNTC post-mortem cohort spanning multiple brain regions and including both PWH and HIV-negative individuals (people without HIV, PWoH), we systematically quantify the frequency and distribution of HIV-infected cells across cell types and anatomical regions in the human brain. This work provides a reference strategy for detecting rare HIV-infected cells in CNS single-cell datasets and offers new insight into the cellular composition of brain reservoirs. METHODS Brain tissue Post-mortem brain tissue was obtained by the SCORCH data generating sites from the National NeuroHIV Tissue Consortium (NNTC) and related cohorts under local institutional review board approval and in accordance with ethical guidelines for the use of human tissue in research. Donors included people with HIV (PWH) and HIV-negative individuals (people without HIV, PWoH). Clinical data such as age, sex, HIV status, plasma viral load, antiretroviral therapy (ART) history, and neuropathological diagnoses were collected by the originating NNTC sites. For this study, we included samples from anatomically defined brain regions deposited in the SCORCH NeMO database, including cortical, subcortical, and cerebellar areas. Each sample consisted of data obtained from frozen tissue from a single region in a single donor. The initial screening dataset comprised 250 samples from 12 brain regions and 102 donors (163 PWH and 95 PWoH). After applying our viral-detection screening and quality-control criteria, the final dataset used for downstream analysis consisted of 48 samples from 35 donors (17 PWH and 18 PWoH). Single-cell and single-nucleus RNA sequencing At each SCORCH data generating site, nuclei were isolated from frozen tissue, quantified, and nuclei concentrations adjusted to manufacturer-recommended concentrations and loaded onto the Chromium Single Cell Multiome ATAC + Gene Expression platform (10x Genomics, Pleasanton, CA, USA). Nuclei were partitioned into Gel Bead-in-Emulsions for simultaneous transposition of accessible chromatin and barcoding of nuclear RNA, followed by construction of separate ATAC and gene expression libraries according to the manufacturer’s instructions. Only the gene expression (GEX) data from the multiome sequencing were used in downstream analyses. Libraries were prepared at the contributing sites, sequenced on Illumina platforms, and raw base call (BCL) files were converted to FASTQ format using cellranger mkfastq or equivalent pipelines. Preprocessing, quality control, and normalization Sequencing reads from single-nucleus gene expression (GEX) data were downloaded from the NEMO database, aligned to a custom reference genome (see below), and gene-level counts were generated using Cell Ranger (version 8.0.1, 10x Genomics) with default parameters unless otherwise specified. The resulting count matrices were first screened based on whether any HIV reads were detected within Cell Ranger–called cells. For the 22 count matrices (22 samples) in which HIV reads were detected in at least one called cell, we then applied per-cell quality control. Barcodes with low library complexity or evidence of poor quality were removed. Specifically, we retained cells with at least 300 detected genes, a total UMI count above 500, and a mitochondrial RNA fraction below 5%. We did not apply a computational doublet-removal step because HIV encephalitis brain tissue contains activated and multinucleated microglia/macrophages that can be misidentified as doublets. After filtering, counts were normalized and variance-stabilized using SCTransform [ 31 ], and highly variable genes were selected for downstream dimension reduction and clustering. Principal component analysis (PCA) was used for initial dimension reduction. Batch correction across all samples was performed using Harmony [ 32 ] run on PCA of top 1000 genes. Cell-type annotation To annotate major CNS cell types, we first performed PCA followed by graph-based clustering. Clustering was carried out using the Louvain algorithm with a resolution of 0.1 and 30 nearest neighbors. We then visualized the resulting clusters using UMAP with settings of 30 neighbors and a minimum distance of 0.3. For each cluster, we summarized cell-type–specific expression by calculating the mean expression of selected marker genes and the percentage of cells expressing each marker. These metrics were used to generate bubble plots to aid visual cell-type identification. Cell-type labels were assigned based on canonical marker genes. Briefly, clusters expressing PLP1 and MBP were annotated as oligodendrocytes; clusters expressing GAD1 and GAD2 as inhibitory neurons; clusters with high SLC17A7 expression as excitatory neurons; clusters expressing PDGFRA and COL9A1 as oligodendrocyte precursor cells; clusters expressing GFAP and AQP4 as astrocytes; clusters expressing CSF1R, P2RY12, and PTPRC as microglia; and clusters expressing CLDN5 and VWF as endothelial cells. Within excitatory and inhibitory neuron clusters, HTR2C expression was used to define HTR2C-positive neuronal subpopulations. To validate our manual cell typing results, we used popV (popular Vote), a cell type annotation method that uses Cell Ontology-aware ensemble voting of eight different algorithms (including random forest, support vector machine, XGBoost, OnClass, CellTypist and k-nearest neighbors on several batch-correction embeddings) [ 18 ]. The method demonstrated ability to reliably transfer labels from an annotated reference dataset to a query dataset. We run 5-fold cross-validation to guarantee stability of predictions and reliability of performance metrics. Along with labels popV also gives prediction scores that we used to create weighted predictions for very rare cases when cross-validation experiment results disagree (Supplementary File 1). For the reference dataset, we used our manual labels based on commonly used marker genes, as well as two external sources of data. The first consisted of 17 dissection datasets from Human Brain Cell Atlas (HBCA) v1.0 [ 33 ]. These HBCA dissection datasets were chosen to span multiple CNS regions, including basal forebrain (septal nuclei), basal nuclei (nucleus accumbens and globus pallidus), hippocampal formation (dentate gyrus/CA4 and subiculum), midbrain (substantia nigra/red nucleus), cerebellum (vermis, deep nuclei, and cerebellar hemisphere), pons, medulla oblongata (including inferior olive and cranial nerve nuclei), perirhinal cortex (A35/A36), and spinal cord. We selected HBCA v1.0 for its large, well-annotated, and widely cited collection of reference cell types, predominantly from three individuals without neurological disorders. Second, we used the Human Microglia Atlas, which separated microglia from a number of datasets from individuals with Alzheimer’s disease, autism spectrum disorders, epilepsy, multiple sclerosis, Lewy body disease, and severe COVID-19 [ 34 ]. Because HIV primarily affects microglia, the latter atlas was essential for providing diverse disease-relevant microglial states that are not represented in healthy-only datasets. Manual labels dataset in the reference excluded one sample that had overwhelming number of HIV+ cells and was considered an outlier (Supplementary File 1). Construction of modified and variant-updated HIV reference genomes Standard reference genomes can under-detect HIV reads because of viral sequence diversity and patient-specific variation. To increase sensitivity for HIV RNA detection, we extended the human reference genome with modified HIV sequences. We combined the human reference genome (GRCh38 from Ensembl [ 35 ]) with the HIV-1 reference genome HXB2 (GenBank: K03455.1). Because HIV-1 contains two long terminal repeat (LTR) regions with nearly identical sequences, reads mapping to both regions are discarded by the Cell Ranger count pipeline as it excludes multi-mapped reads. To retain reads originating from the LTR, we removed the 3′ LTR and kept only the 5′ LTR when building the combined reference. This customized Human + HIV reference genome was generated using cellranger mkref and used as the initial screening reference to identify infected cells across all samples. HIV has one of the highest mutation rates among viruses, and after many rounds of reverse transcription, viral genomes in individual patients can differ substantially from the canonical HXB2 sequence. When reads are aligned only to an unmodified reference, accumulated mismatches can reduce mapping quality and lead to missed HIV reads. To mitigate this, we generated subject-specific, variant-updated HIV references for donors with substantial numbers of infected cells. To ensure sufficient coverage for reliable variant calling, we selected subjects with more than 10 infected cells detected using the modified HXB2 reference, pooled all HIV-aligned reads using samtools (version 1.22), performed variant calling to identify single-nucleotide variants using bcftools (version 1.18), and built a consensus sequence based on the proportion of alternate alleles [ 36 ]. This consensus HIV genome was then combined with GRCh38 and used with cellranger mkref to create a subject-specific reference genome for re-alignment. Construction of HIV strain collection reference genomes A total of 9,444 HIV-1 clade B sequences in FASTA format were downloaded from Los Alamos HIV Sequence Database ( https://www.hiv.lanl.gov/ ) and used to construct a strain collection reference. STAR(V) genomeGenerate was used to index the sequences and create the a composite HIV strain collection reference for alignment [ 37 ]. Alignment and quantification of HIV transcripts For both the 5’ LTR-modified reference and the reference update methods, reads were aligned to the corresponding human + HIV reference genomes, and gene-level UMI counts were generated using the standard 10x Genomics cellranger count pipeline. HIV transcripts were quantified by summing UMIs assigned to HIV genomic regions for each cell barcode. For the HIV strain collection method, sequencing reads were aligned to HIV-1 clade B collection reference using STAR with multimapping parameters (--outFilterMultimapNmax 1000) to capture candidate viral reads. To reduce false-positive alignments driven by homopolymeric artifacts, reads containing poly-G tails were removed prior to downstream analysis. Viral reads were then collapsed by UMIs to quantify unique viral UMIs Finally, we mapped the cell barcodes associated with these UMIs back to the set of called cells. In all samples from PWoH donors, we did not detect any cells with HIV UMIs. For PWH samples, HIV UMI counts per cell were carried forward and integrated with the host gene expression matrix for downstream analyses. Definition of HIV RNA–positive cells. A cell was considered HIV RNA–positive if HIV UMIs were detected using any of the reference strategies described above. For cells processed with more than one method (e.g., generic 5′ LTR-modified HXB2, subject-specific variant-updated reference, or the HIV strain collection), we took the maximum HIV UMI count across methods as the value for that cell. Unless otherwise noted, our primary analyses defined HIV RNA–positive cells as those with HIV UMI > 1. Analyses using UMI > 0 and > 2 were used to help determine the HIV UMI cutoff (Fig. 9 and Supplementary Table 1). Immunohistochemistry Immunohistochemistry for HIV-1 p24 was performed on 5 µM sections of formalin-fixed paraffin-embedded (FFPE) brain sections. Sections were deparaffinized using xylene and rehydrated prior to Tris-EDTA antigen retrieval buffer (10 mM Tris base, 1 mM EDTA solution, 0.05% Tween 20, pH 9.0) (40min, 100°C). Slides were blocked in 5% normal horse serum in PBS, followed by primary antibody (HIV-1 p24, SantaCruz Biotechnology sc-65918 Lot#2325) incubation at a dilution of 1:50 in blocking buffer overnight at 4°C. HRP linked anti-mouse secondary was developed for 6 minutes using ImmPACT NovaRed (Vector SK-4805), followed by a 30 second Gill 2 Hematoxylin nuclear stain (StatLab SL94-16). Results Infected Cell Screening The screening dataset comprised raw sequencing data files from 250 samples from 12 anatomically defined brain regions across 102 participants, including 156 samples from 73 people with HIV (PWH) and 94 samples from 32 people without HIV (PWoH). The only criteria used for choosing data was availability in the SCORCH consortium as of May 13th 2025, and that the samples were from an NNTC participant (SCORCH also utilizes samples from other studies for PWoH). Using the 10xGenomics cellranger pipeline with the modified 5’ LTR reference (Human + HIV removing 3’ LTR), similar to our studies in SIV-infected rhesus monkeys [ 15 ] we first examined the identified total of 1,885,561 cells from 156 PWH samples, and 782,202 cells from 94 PWoH samples. We next restricted the study to 22 PWH samples in which HIV reads were detected and 26 selected PWoH samples, yielding 48 samples from 35 donors (17 PWH, 18 PWoH) and 559,207 high-quality cells (314,409 PWH, 244,798 PWoH) from PWH for downstream analyses. The 26 PWoH samples were selected based on attempting to best match the number of donors and brain regions to confirm major cell types across PWoH and PWH samples. Infected cell detection Figure 1 summarizes the infected HIV RNA-positive cell detection workflow, which combines three complementary approaches: (i) a modified 5′ LTR HIV reference method, here using the HXB2 reference viral sequence (with deletion of the 3’ LTR) combined with the GRCh38 human genome, (ii) subject-specific variant-updated HIV references (viral reference update method), and (iii) an HIV strain collection reference method. Using the modified 5′ LTR HIV reference genome, we identified 1,739 cells with detectable HIV RNA (at least one HIV UMI). These HIV RNA-positive cells were confined to samples from PWH while no HIV-positive cells were detected in PWoH samples at the thresholds used. Six of the positive PWH samples were from the three donors with HIV encephalitis (HIVE) and contained 1,636 infected cells in total, whereas the remaining 16 non-HIVE samples (from 14 donors) together contributed only 103 infected cells. One sample from subject 6800127569 (with HIVE, from the ventral striatum) was particularly enriched with 1,466 infected cells. When this sample was excluded, 255 infected cells remained, corresponding to ~ 0.05% of all nuclei from 22 PWH samples. While this study was not designed to compare infection in specific brains regions, since multiple regions were assayed from many of the donors we examined how often the virus was found in one region of the brain while it could be found in a different distinct region in the same donor, and how this related to HIVE and viral suppression. Of the 17 donors in which HIV RNA-positive cells were found, 15 had more than one brain region sampled. Three of the donors had HIVE, and two had more than one region with positive cells, while in total six of the eight brain regions examined had positive cells (67%). Nine donors with more than one region examined did not have viral suppression at their last clinical visit (plasma viral loads ranging from 123 to 441,668 RNA copies/ml, with a median of 1,523), and three had more than one region with HIV RNA-positive cells, resulting in 12 of the 20 brain regions examined having positive cells (60%). The remaining three donors were virally suppressed at the last measurement of HIV load (plasma viral loads < 50 RNA copies/ml), and each only had one region with positive cells, resulting in three of the eight regions examined with HIV RNA-positive cells (38%). This relative rarity is also reflected in our finding that multiple regions were assessed from an additional 12 virally suppressed donors (34 samples) with no HIV-infected cells detected. Including another five suppressed cases in which only a single region was examined (all without positive cells), only four HIV-infected cells were found among the 436,304 nuclei examined in virally suppressed PWH (0.001%). From the results of modified 5’ LTR reference method, we identified six subjects from 10 samples with more than ten infected cells each and applied the viral reference update method to these samples. We reasoned that by improved mapping of the viral reference to the viral sequence in the brain we would identify more reads mapping to HIV, and likely more infected cells. This is similar to the single-cell ECCITE-seq approach described by Collora et al. [ 16 ], where incorporating subject-specific (autologous) HIV-1 sequences in addition to HXB2 for read mapping increased the HIV mapping rate and improved HIV RNA capture/detection of rare HIV RNA-positive cells. We therefore extracted all HIV-aligned reads from the BAM files generated with the modified HXB2 reference, merged reads across regions from the same subject, and performed variant calling using bcftools. As summarized in Table 1, subjects with more infected cells yielded more called variants. Variants with an alternate-allele frequency > 50% were incorporated into a subject-specific HXB2 consensus, which was then combined with GRCh38 to build an updated reference genome for each subject. Re-alignment against these subject-specific references recovered an additional 162 infected cells in 6 of the 10 samples (the remaining 4 samples did not yield additional infected cells), resulting in 1,895 cell barcodes with associated HIV UMI counts over 1 when combined with the original 5′ LTR-based calls. Visualization of representative samples from different brain regions of subject 4084 in the genome browser confirmed improved coverage and alignment in regions with subject-specific variation (Fig. 2 ). Using the HIV strain collection reference method in which reads were mapped to a panel of 9444 HIV strain reference genomes [ 17 ] we independently detected 1,739 cells with at least one HIV UMI. As shown in Fig. 3 , 1 ,696 HIV RNA-positive cells were identified by both approaches (here, the 5′ LTR-modified reference plus variant-updated reference are grouped as “5’ LTR+RefUpdated”), while 43 cells were uniquely identified by the strain-collection method and 199 cells were uniquely identified by the 5’ LTR_RefUpdated approach. Thus, the modified 5′ LTR plus variant-updated references captured most HIV RNA-positive cells, while the strain-collection reference contributed to a small but non-negligible set of additional cells. Taking the union of HIV RNA-positive cells identified by all three methods, yielded a final set of 1,939 HIV RNA-positive cells with HIV UMI > 0. For each cell, we assigned the maximum HIV UMI count observed across the three methods as its HIV UMI value for downstream analyses. As shown in Fig. 4 , HIV RNA was detected at low copy number in most HIV RNA-positive cells (1,014 cells with 1 HIV UMI, 180 cells with 2 HIV UMIs), with a sharp decline in cell counts beyond HIV UMI ≥ 2, consistent with a long-tailed distribution of viral RNA burden. When stratified by annotated cell types (Fig. 5 ), cumulative infected-cell counts as a function of the HIV UMI threshold displayed clear saturation behavior for most populations, indicating that increasing stringency rapidly prunes low-read HIV-positive cells. Notably, microglia — the most abundant HIV RNA-positive cell population — showed a slower decline and remained detectable at substantially higher HIV UMI levels (with the increase in retained infected cells beginning to plateau at higher thresholds), whereas other cell types saturated much earlier (e.g., endothelial cells, oligodendrocytes, and neurons). Despite the known limitation that snRNA-seq captures only a fraction of cellular RNA and is particularly insensitive to low-level signal, these patterns are consistent with cell-type-specific differences in HIV RNA burden, with microglia harboring the highest levels, followed by oligodendrocytes and astrocytes, and lower levels in endothelial cells and neurons (with most HIV RNA-positive cells originating from HIVE samples). To reduce the impact of potential false positives due to free-floating/ambient viral RNA and background alignment, we used HIV UMI > 1 as the primary threshold in further analyses to define infection in the remaining cells. After additionally applying per-cell quality filters (total UMI counts and mitochondrial read percentage), 908 HIV RNA-positive cells remained for downstream analyses. Distribution of infected cells across cell types Using the cell-by-gene expression matrix from the 48 selected samples (combining those from PWH and PWoH), we confirmed expected expression of major cell-type and subtype marker genes and assigned cluster identities. As shown in Fig. 6 A–B, among 559,207 cells, 171,604 (30.69%) were identified as oligodendrocytes based on MBP and PLP1 expression, and 77,356 (13.83%) were annotated as astrocytes based on AQP4 and GFAP. SNAP25 was used as a general neuronal marker, and neurons were further divided into excitatory and inhibitory populations, with or without expression of the serotonin receptor gene HTR2C. This resulted in five excitatory neuron subtypes, two inhibitory neuron subtypes, two HTR2C⁺ inhibitory neuron subtypes, and one HTR2C⁺ excitatory neuron subtype, with a total of 201,823 (36.09%) neurons. In addition, 48,125 cells (8.61%) were classified as microglia based on the expressions of CSF1R, P2RY12, and PTPRC. As summarized in Fig. 6 C and Supplementary Table 1, in the infected cell pool, 78.6% of HIV RNA-positive cells were microglia, representing the largest infected cell population, followed by oligodendrocytes (6.06%) and astrocytes (4.85%); 5.73% of HIV RNA-positive cells (52 of 908) were classified as neurons (combining all neuron subtypes). In the total PWH cell pool of the 22 samples with HIV RNA-positive cells detected, 2.23% of microglia were HIV RNA-positive, followed by 0.24% of endothelial cells, 0.1% of astrocytes, and 0.05% of the combined neuronal cell types. The total detectable HIV RNA-positive rate of all cells was 0.29%. While excluding sample 6800127569_ventral_striatum with the exceptionally large number of HIV RNA-positive cells, 0.38% of microglia, 0.02% of oligodendrocytes, 0.01% of astrocytes, and 0.02% of the combined neuronal cell types were HIV RNA-positive. The total infection rate represented by HIV RNA-positive cells was reduced to 0.05% (Supplementary Fig. 1). If we consider all PWH samples (including those without HIV RNA-positive cells detected, a total of 156 samples with 1,885,561 cells), the infection rates became 0.05% of all cells, dropping to 0.01% when excluding sample 6800127569_ventral_striatum. Validation of Cell type annotation To validate our manual cell typing results, we used popV (popular Vote), a cell type annotation method that uses Cell Ontology-aware ensemble voting of eight different algorithms with label transfer from multiple reference datasets [ 18 ]. Table 2 shows agreement between manual annotation and popV results represented on cells with HIV UMI detected, represented in classification metrics, giving us confidence in our cell type annotation. In Fig. 7 , the bubble heatmap shows how the weighted predicted labels correspond to our manually annotated HIV-infected cells. The color of the bubbles shows the average popV prediction score, indicates how many annotation methods support a cell’s final consensus type, including votes propagated to ancestor types in the ontology. In cross-validation, by combining these scores across experiments we computed weighted final cell type predictions. We also applied the popV to all the data sets that we used for screening, as shown in Supplementary File 1, cross validation of popV showed great performance and stability of the cell annotation across experiments, data sources, cell types. N/A means that a cell type is not represented in manual labels but is present in reference datasets. In tissue from cases with HIVE, as expected, finding HIV-infected cells using immunohistochemistry easily identified HIV p24-positive cells resembling microglia and macrophages, as well as the pathognomonic multinucleated giant cells (Fig. 8 A). However, in non-HIVE cases detection of positive cells was difficult and were only rarely seen, including in a donor in which infected oligodendrocytes were identified, in which a positive cell with the morphological characteristics of an oligodendrocyte was found (Fig. 8 B). Due to the rarity of such cells, double labeling for cell-type markers could not be performed. HIV encephalitis versus non-encephalitis across cell types Beyond the total number of infected cells per UMI count identified in Fig. 4 , we examined how the distribution of HIV RNA-positive cells across cell types changed under different HIV UMI thresholds. As shown in Fig. 9 and Supplementary Table 1, when HIV UMI > 0, 19.5% of infected cells (371 cells) were oligodendrocytes and 13.83% (213 cells) were astrocytes. When we applied a more stringent cutoff of HIV UMI > 1, these proportions dropped to 6.06% (55 cells) for oligodendrocytes and 4.85% (44 cells) for astrocytes. A similar trend was observed for most other non-microglial cell types. In contrast, microglia showed a different pattern. Among microglia, 870 cells had HIV UMI > 0, 714 had HIV UMI > 1, and 638 had HIV UMI > 2, indicating a relatively modest decrease as thresholds became more stringent. This slower decline in microglial infection frequency is consistent with microglia being much more susceptible to HIV infection and expression than other CNS cell types. When we stratified cells by the presence or absence of HIV encephalitis (HIVE) (Fig. 9 ) the proportions of infected cells in most cell types decreased or disappeared as the HIV UMI threshold increased in the non-HIVE group. As expected, most infected cells originated from HIVE cases, reflecting the abundant expression of HIV transcripts in encephalitic tissue. Notably, however, even within the smaller pool of infected cells in non-HIVE cases, microglia consistently constituted a large proportion of infected cells, reinforcing their role as a principal CNS HIV reservoir [ 19 , 20 ]. Together, these patterns indicate that in HIVE, HIV RNA-positive cells are distributed across multiple CNS lineages, whereas in the absence of HIVE, detectable infection appears largely restricted to microglia, with only rare oligodendrocyte or astrocyte involvement. Location of HIV sequence variants Using the reference-updated mapping strategy, we found that most HIV-derived reads localized to the U3 region of the 5′ LTR and to the env gene ingp120 and gp41 (Fig. 2 ). The U3 region encodes key cis-regulatory elements, including the core promoter, enhancer, and multiple transcription factor binding sites (e.g., NF-κB, Sp1) that control viral transcriptional activation and integration site–dependent expression [ 21 ]. These functional constraints, together with host- and tissue-specific selective pressures on transcriptional regulation, may contribute to the accumulation of sequence diversity in U3, especially in enhancer and promoter modules that modulate replication fitness in different cellular environments, including CNS myeloid cells. Likewise, env, particularly gp120 and the extracellular portion of gp41, is under strong immune and receptor-mediated selection, as these regions mediate CD4/coreceptor binding and membrane fusion and are primary targets of neutralizing antibodies [ 22 ]. Thus, the high density of variants we observed in U3 and env likely reflects a combination of their central regulatory and entry functions and the strong, ongoing selection exerted by the host environment in the CNS. DISCUSSION In this study, we developed a scalable, multi-reference framework for detecting rare HIV RNA–positive cells in human CNS single-nucleus RNA-seq data. By combining the modified 5′ LTR reference method, viral reference update method, and the HIV strain collection method, we identified a high-confidence set of HIV RNA–positive cells in post-mortem brains from people with HIV. Within this framework, we show that microglia constitute the predominant infected population, with oligodendrocytes, astrocytes, and neurons contributing smaller fractions largely restricted to cases with HIV encephalitis (HIVE). These results provide an innovative approach and a quantitative baseline for studying HIV-infected CNS cell types in NeuroHIV single-cell datasets. Infected cells were also detected in several non-microglial lineages, including neurons, oligodendrocytes, and astrocytes, but this pattern was strongly dependent on neuropathological state. In HIVE cases, HIV RNA-positive nuclei were distributed across multiple CNS cell types, with clear representation in oligodendrocytes and astrocytes and occasional infected neurons, consistent with postmortem and genomic studies showing that, although microglia are the dominant target, astrocytes and neurons can occasionally harbor HIV DNA or RNA in encephalitic brain tissue [ 23 , 24 ]. In contrast, in non-HIVE tissue, infected cells were almost exclusively confined to microglia, with only a small fraction observed in oligodendrocytes and virtually no signal in neurons or astrocytes. This is aligned with recent work demonstrating that microglia constitute the principal and most persistent HIV reservoir in the brain across disease stages and even during suppressive ART, whereas evidence of infection of other CNS resident cells appears less frequent [ 25 ]. The restriction of detectable infection to microglia (and to a lesser extent oligodendrocytes) in non-HIVE cases versus other cell types found in HIVE likely reflects the high level of virus, inflammation, and cellular damage in encephalitis. present in the brain. In the absence of encephalitis and especially in the case of systemic viral suppression, microglia maintain their role as the dominant CNS reservoir while infection of other lineages is rare at best. The complete absence of HIV UMIs in PWoH donors, together with the strong enrichment of HIV UMIs in microglial clusters rather than diffusely across cell types, supports the specificity of our cell-calling and viral transcript identification strategies. In this study, we analyzed single-nucleus derived gene expression (snRNA-seq) to detect HIV-associated transcripts at single-cell resolution. This modality primarily captures the transcriptional footprint of infection including viral RNA abundance, cellular activation states, and host-response programs. However, we acknowledge some limitations. In particular, snRNA-seq does not directly interrogate stages of the viral life cycle. Nuclear profiling can be biased towards different splicing stages of mRNA, and in the case of HIV, once the Rev protein is made it functions to transport unspliced and singly spliced RNA from the nucleus to the cytoplasm [ 26 ], increasing the problem of technical dropout. Combined, these factors limit sensitivity for detecting low-level or latent infection. Recent conceptual frameworks emphasize that understanding CNS HIV persistence requires integrating viral transcription, chromatin accessibility, and cell-state–specific immune programs, rather than relying on viral detection alone [ 27 ]. In future work, integrating the paired snATAC-seq modality will enable direct assessment of LTR/proviral accessibility and surrounding regulatory landscapes that govern latency versus reactivation, and will provide additional context for integration-site preferences in relation to open chromatin and active gene neighborhoods [ 28 , 29 ]. Finally, infected-cell detection was highly imbalanced across pathological groups (e.g., substantially more HIV positive cells in HIVE than in non-HIVE), limiting our ability to characterize virus-positive states in non-HIVE cases, particularly in ART-suppressed individuals where viral transcription is expected to be low [ 30 ]. The number of cells sampled in single cell analyses is a large factor, limited largely by cost, as given the low infection rate in the absence of HIVE detection is largely stochastic. Addressing these limitations will require increased sampling of brains from brains without HIV, ART-treated donors and/or complementary enrichment and validation strategies (e.g., targeted HIV RNA capture, proviral DNA/integration assays, and orthogonal in situ detection) to improve sensitivity and confirm cell-type attribution. Beyond these biological and sampling constraints, the practical details of reference construction, alignment, and thresholding can strongly influence what is called “HIV RNA–positive” in snRNA-seq. HIV-1 mapping is particularly sensitive to reference design because duplicated regions (most notably the two LTRs) create multi-mapping; reads that align equally well to both ends can be down-weighted or discarded by standard quantification pipelines. To reduce this ambiguity while preserving sensitivity, we retained a single LTR in the viral reference (keeping the 5′ LTR and removing the 3′ LTR), which makes LTR-derived signal easier to interpret—an important consideration when many infected nuclei are supported by only a few viral UMIs. A second practical issue is viral diversity: even modest divergence from a canonical reference can lower alignment confidence for short reads and low-abundance transcripts. When we had sufficient HIV-aligned reads to support variant calling, subject-specific variant-updated references improved matching to within-host sequence, and mapping to a curated strain collection provided a complementary way to accommodate diversity without requiring de novo assembly for every donor. Because viral signal is sparse, thresholding is also not a minor detail; we recommend reporting infected-cell counts and cell-type proportions across simple UMI cutoffs (e.g., HIV UMI > 0, >1, > 2) and then choosing a primary cutoff based on the analytic goal, using higher stringency when specificity and cross-sample comparability are priorities while still showing the full sensitivity series for transparency. Finally, rare-event detection is vulnerable to technical artifacts (ambient RNA, sporadic misalignment, barcode-related effects), so negative controls processed through the identical pipeline provide a direct estimate of background; in our dataset, we did not detect HIV UMIs in samples from individuals without HIV at the thresholds used. To make the approach easy to reproduce, we report software versions, reference-build steps, and key alignment/counting parameters. CONCLUSION In summary, by combining modified 5′ LTR-based references, subject-specific variant-updated HIV genomes, and a comprehensive HIV strain collection method, we developed a scalable framework for detecting rare HIV RNA–positive cells in human CNS single-nucleus datasets. We recommend starting with a custom genome combining the human reference genome (currently GRCh38 from Ensembl) with the modified HIV-1 reference genome HXB2 (NCBI GenBank: K03455.1), and analyzing the sequencing reads with Cell Ranger. If additional sensitivity is desired, use of the HIV strain collection, and if sufficient HIV UMIs are detected, the subject-specific variant updated HIV genome. The UMI threshold can be adjusted depending on the neuropathology (e.g., HIVE) and objective of the study. When we applied this pipeline to 314,409 nuclei from 22 post-mortem brain samples from PWH, and required > 1 UMI, this approach identified 908 HIV RNA-positive cells and revealed that microglia constitute the predominant infected population, with oligodendrocytes, astrocytes, and neurons contributing smaller fractions, particularly in HIVE cases. The strong enrichment of infected microglia in both HIVE and non-HIVE tissue, together with the more restricted involvement of other lineages outside of encephalitic brains, reinforces the concept of microglia as a principal and persistent brain reservoir of HIV. Our variant analyses further show that HIV reads in CNS tissue are concentrated in the U3 region of the LTR and in env (gp120/gp41), highlighting regulatory and entry-associated regions as focal points of viral diversity in the brain. Although our study is constrained by the inherent sparsity of single-nucleus RNA-seq, the cross-sectional nature of post-mortem tissue, and the focus on viral RNA rather than proviral DNA, the framework and reference resources presented here provide a roadmap for harmonized viral detection in NeuroHIV single-cell studies. Future work integrating viral detection and genotypes with longitudinal clinical data, proviral sequencing, and functional perturbation of infected microglia will be essential to translate these insights into strategies for monitoring and targeting CNS HIV reservoirs. Declarations Ethics approval and consent to participate Human post-mortem tissues and associated de-identified metadata were obtained by SCORCH investigators from the National NeuroHIV Tissue Consortium (NNTC). Work on decedents is not considered human subject research. Consents for participation in the NNTC studies were through multiple IRB protocols since 1999 to the present time, approved by the IRBs at the clinical sites ( UTMB, UCSD, UCLA, ISMMS). Availability of data and materials Data presented in this study were produced as part of the Single Cell Opioid Response in the Context of HIV consortium (SCORCH: RRID:SCR_022600). Publicly accessible data is available at NeMO Archive (RRID:SCR_002001) under identifier nemo:col-12t9m0h (https://assets.nemoarchive.org/col-12t9m0h). Access to all protected data associated with this study is managed by dbGaP and can be requested with the identifier phs003991.v1.p1 . Competing interests The authors declare that they have no competing interests. Funding This work was supported by NIDA 1U01DA053624. Authors’ contributions HSF, MN, YC, NJ was responsible for the conceptualization of the study. Single-cell analyses were performed by MN, YC, NJ and DM. Visualization of the data was performed by MN, CY, DM. Supervision of the study was the responsibility of HSF, DG and MN. The original draft of the manuscript was prepared by HSF, MN, and all authors (MN, HSF, YC, NJ, DG, MC, BH, DV, JR, DM, OW, YK, and CC) reviewed, edited, and approved the final version. Acknowledgements The single-nucleus data examined were generated by members of the SCORCH consortium: Y-SCORCH (S. Spudich, M. Gerstein and Y. Kluger, PIs, NIH grant UM1DA051410), M-SCORCH (Y-C Ho, PI, NIH grant U01DA053628), BROAD-SCORCH (M, Kellis and M. Heiman, PIs, NIH grant U01DA053631), UCSD SCORCH (T. Rana, PI, NIH grant U01DA053630), Weill Cornell SCORCH (H. Tilgner, T. Milner, and L. Ndhlovu, PIs, NIH grant U01DA053625). Data sets were curated and provided by the SCORCH Data Coordinating Center (O. White, S. Ament, and A. Mahurkar, PIs, NIH grant UM1DA052244). The brain tissue specimens and associated metadata were provided by the National NeuroHIV Tissue Consortium (NNTC), in which the member institutions have NIH contracts supported by multiple institutes (NIMH, NIA, NIDA, and NINDS) to support the their role in the NNTC, with the following contract numbers: Texas NeuroAIDS Research Center (TNRC), PI B. Gelman, University of Texas Medical Branch, Galveston: 75N95023C00016; California NeuroAIDS Tissue Network (CNTN), PI D. Moore, University of California, San Diego: 75N95023C00014; National Neurological AIDS Bank (NNAB), PI E. Singer, University of California, Los Angeles: 75N95023C00017; Manhattan HIV Brain Bank (MHBB), PI S. Morgello, Icahn School of Medicine at Mt. Sinai, New York, NY: 75N95023C00015; Data Coordinating Center (DCC), PI S. Sherman, Emmes Company, Rockville, MD: 75N95023C00013. References WHO. HIV data and statistics. 2025 [cited 2025 Dec. 16]; Available from: https://www.who.int/teams/global-hiv-hepatitis-and-stis-programmes/hiv/strategic-information/hiv-data-and-statistics#:~:text=People%20living%20with%20HIV,HIV%2Drelated%20causes%20in%202024 Heaton RK, et al. HIV-associated neurocognitive disorders persist in the era of potent antiretroviral therapy: CHARTER Study. Neurology. 2010;75(23):2087–96. Nightingale S, et al. Cognitive impairment in people living with HIV: consensus recommendations for a new approach. Nat Rev Neurol. 2023;19(7):424–33. Plaza-Jennings A, Akbarian S. Genomic Exploration of the Brain in People Infected with HIV-Recent Progress and the Road Ahead. Curr HIV/AIDS Rep. 2023;20(6):357–67. Zheng GX, et al. Massively parallel digital transcriptional profiling of single cells. Nat Commun. 2017;8:14049. 10XGenomics. What fraction of mRNA transcripts are captured per cell?. 2025 [cited 2025 Dec. 16]; Available from: https://kb.10xgenomics.com/s/article/360001539051-What-fraction-of-mRNA-transcripts-are-captured-per-cell Viswanathan S, et al. Level of adherence and HIV RNA suppression in the current era of highly active antiretroviral therapy (HAART). AIDS Behav. 2015;19(4):601–11. Plaza-Jennings AL, et al. HIV integration in the human brain is linked to microglial activation and 3D genome remodeling. Mol Cell. 2022;82(24):4647–e46638. Farhadian SF et al. Single-cell RNA sequencing reveals microglia-like cells in cerebrospinal fluid during virologically suppressed HIV. JCI Insight, 2018. 3(18). Wilson AM, et al. Transcriptional impacts of substance use disorder and HIV on human ventral midbrain neurons and microglia. Nat Commun. 2025;16(1):9123. Ament SA, et al. The single-cell opioid responses in the context of HIV (SCORCH) consortium. Mol Psychiatry. 2024;29(12):3950–61. Morgello S, et al. The National NeuroAIDS Tissue Consortium: a new paradigm in brain banking with an emphasis on infectious disease. Neuropathol Appl Neurobiol. 2001;27(4):326–35. Heithoff AJ et al. The integrated National NeuroAIDS Tissue Consortium database: a rich platform for neuroHIV research. Database (Oxford), 2019. 2019. Cserhati MF et al. The National NeuroAIDS Tissue Consortium (NNTC) Database: an integrated database for HIV-related studies. Database (Oxford), 2015. 2015: p. bav074. Xu X, et al. Transformation of brain myeloid cell populations by SIV in rhesus macaques revealed by multiomics. Commun Biol. 2025;8(1):100. Collora JA, et al. Single-cell multiomics reveals persistence of HIV-1 in expanded cytotoxic T cell clones. Immunity. 2022;55(6):1013–e10317. Wei Y, et al. Single-cell epigenetic, transcriptional, and protein profiling of latent and active HIV-1 reservoir revealed that IKZF3 promotes HIV-1 persistence. Immunity. 2023;56(11):2584–e26017. Ergen C, et al. Consensus prediction of cell type labels in single-cell data with popV. Nat Genet. 2024;56(12):2731–8. Wallet C, et al. Microglial Cells: The Main HIV-1 Reservoir in the Brain. Front Cell Infect Microbiol. 2019;9:362. Tang Y et al. Brain microglia serve as a persistent HIV reservoir despite durable antiretroviral therapy. J Clin Invest, 2023. 133(12). Cullen BR. Regulation of HIV-1 gene expression. FASEB J. 1991;5(10):2361–8. Wyatt R, Sodroski J. The HIV-1 envelope glycoproteins: fusogens, antigens, and immunogens. Science. 1998;280(5371):1884–8. Trillo-Pazos G, et al. Detection of HIV-1 DNA in microglia/macrophages, astrocytes and neurons isolated from brain tissue with HIV-1 encephalitis by laser capture microdissection. Brain Pathol. 2003;13(2):144–54. Donoso M et al. Identification, Quantification, and Characterization of HIV-1 Reservoirs in the Human Brain. Cells, 2022. 11(15). Joseph SB, et al. HIV-1 target cells in the CNS. J Neurovirol. 2015;21(3):276–89. Cullen BR. Nuclear mRNA export: insights from virology. Trends Biochem Sci. 2003;28(8):419–24. Filippidis P, Corley MJ. Single cell analyses of the HIV reservoir in the CNS and CSF: recent insights and implications. Curr Opin HIV AIDS. 2025;20(5):493–501. Mbonye U, Karn J. The Molecular Basis for Human Immunodeficiency Virus Latency. Annu Rev Virol. 2017;4(1):261–85. Schroder AR, et al. HIV-1 integration in the human genome favors active genes and local hotspots. Cell. 2002;110(4):521–9. Bruner KM, et al. A quantitative approach for measuring the reservoir of latent HIV-1 proviruses. Nature. 2019;566(7742):120–5. Hafemeister C, Satija R. Normalization and variance stabilization of single-cell RNA-seq data using regularized negative binomial regression. Genome Biol. 2019;20(1):296. Korsunsky I, et al. Fast, sensitive and accurate integration of single-cell data with Harmony. Nat Methods. 2019;16(12):1289–96. Siletti K, et al. Transcriptomic diversity of cell types across the adult human brain. Science. 2023;382(6667):eadd7046. Martins-Ferreira R, et al. The Human Microglia Atlas (HuMicA) unravels changes in disease-associated microglia subsets across neurodegenerative conditions. Nat Commun. 2025;16(1):739. Dyer SC, et al. Ensembl 2025. Nucleic Acids Res. 2025;53(D1):D948–57. Danecek P et al. Twelve years of SAMtools and BCFtools. Gigascience, 2021. 10(2). Dobin A, et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics. 2013;29(1):15–21. Tables Tables are available in the Supplementary Files section. Additional Declarations No competing interests reported. Supplementary Files table1.xlsx table2.xlsx supplementarytable1.xlsx Supplementary Table 1. Cell counts for each of the cell types across PWH samples, PWoH samples, and different thresholds of HIV UMI counts. supplementarytable2.xlsx Supplementary Table 2. Master sheet of samples in this study including subject IDs, HIV status, cognitive status, last plasma viral load, brain pathology, brain region, sequencing metrics, and HIV RNA-positive cell counts. The file contains four worksheets: 1) PWH – all files from PWH analyzed, those with HIV RNA-positive cells in red font, light green shading indicates duplicate samples; 2) HIV detected – subset of files from PWH containing HIV RNA-positive cells; 3) PWoH – all files from PWoH analyzed; and 4) PWoH selected – subset of files from PWoH chosen for comparison to PWH HIV detected. supplementaryfile1.docx Supplementary File 1. PopV configuration, quality control procedure, annotation reference datasets, and cross validation. Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 19 May, 2026 Reviews received at journal 13 May, 2026 Reviewers agreed at journal 20 Apr, 2026 Reviewers agreed at journal 20 Apr, 2026 Reviewers invited by journal 26 Mar, 2026 Editor assigned by journal 03 Mar, 2026 Submission checks completed at journal 03 Mar, 2026 First submitted to journal 02 Mar, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9013592","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Method Article","associatedPublications":[],"authors":[{"id":600066952,"identity":"d19626a7-5912-482e-b5d3-4392c16e315a","order_by":0,"name":"Yuan-yuan Cai","email":"","orcid":"","institution":"University of California, San Diego","correspondingAuthor":false,"prefix":"","firstName":"Yuan-yuan","middleName":"","lastName":"Cai","suffix":""},{"id":600066953,"identity":"7d211db8-d665-47a1-aa30-ec4939a21b2f","order_by":1,"name":"Nicholas Jacobs","email":"","orcid":"","institution":"Yale University","correspondingAuthor":false,"prefix":"","firstName":"Nicholas","middleName":"","lastName":"Jacobs","suffix":""},{"id":600066954,"identity":"dcc5d2f6-ed7a-4930-869d-c20fc38341c0","order_by":2,"name":"Dana Gabuzda","email":"","orcid":"","institution":"Harvard University","correspondingAuthor":false,"prefix":"","firstName":"Dana","middleName":"","lastName":"Gabuzda","suffix":""},{"id":600066955,"identity":"b7555cb3-e341-4e53-9dd9-0ef7c1729e53","order_by":3,"name":"Michael Corley","email":"","orcid":"","institution":"University of California, San Diego","correspondingAuthor":false,"prefix":"","firstName":"Michael","middleName":"","lastName":"Corley","suffix":""},{"id":600066956,"identity":"e7ad9942-184f-4818-8ae8-0e9b69fc4a7e","order_by":4,"name":"Brian Herb","email":"","orcid":"","institution":"University of Maryland, Baltimore","correspondingAuthor":false,"prefix":"","firstName":"Brian","middleName":"","lastName":"Herb","suffix":""},{"id":600066957,"identity":"1d60acd6-abd2-455e-8238-12216250ddb5","order_by":5,"name":"David Volsky","email":"","orcid":"","institution":"Icahn School of Medicine at Mount Sinai","correspondingAuthor":false,"prefix":"","firstName":"David","middleName":"","lastName":"Volsky","suffix":""},{"id":600066958,"identity":"bf62fec1-23de-49a0-b13c-c764d7bea633","order_by":6,"name":"Joseph Receveur","email":"","orcid":"","institution":"University of Maryland, Baltimore","correspondingAuthor":false,"prefix":"","firstName":"Joseph","middleName":"","lastName":"Receveur","suffix":""},{"id":600066959,"identity":"e6733a16-4e1e-489e-8f4d-ac5b026968c6","order_by":7,"name":"Dorzhey Matkhanov","email":"","orcid":"","institution":"University of Nebraska Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Dorzhey","middleName":"","lastName":"Matkhanov","suffix":""},{"id":600066960,"identity":"bd71206e-6226-45c1-ba45-54f03cb537e8","order_by":8,"name":"Owen White","email":"","orcid":"","institution":"University of Maryland, Baltimore","correspondingAuthor":false,"prefix":"","firstName":"Owen","middleName":"","lastName":"White","suffix":""},{"id":600066961,"identity":"1830a007-6407-4aee-8b71-50e6a515cb08","order_by":9,"name":"Yuval Kluger","email":"","orcid":"","institution":"Yale University","correspondingAuthor":false,"prefix":"","firstName":"Yuval","middleName":"","lastName":"Kluger","suffix":""},{"id":600066962,"identity":"9a97d23a-ec4d-450b-bae4-d9d72bb7001d","order_by":10,"name":"Christine Chen","email":"","orcid":"","institution":"University of California, San Diego","correspondingAuthor":false,"prefix":"","firstName":"Christine","middleName":"","lastName":"Chen","suffix":""},{"id":600066963,"identity":"b7657064-a74d-4944-8e72-1741e67f2e45","order_by":11,"name":"Howard Fox","email":"","orcid":"","institution":"University of Nebraska Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Howard","middleName":"","lastName":"Fox","suffix":""},{"id":600066964,"identity":"6bf372cd-87c9-4a21-8954-130a524f93a7","order_by":12,"name":"Meng Niu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAvUlEQVRIiWNgGAWjYPACCQZ+KIuxgWgtkg0kamFgMDhArBbd9t7DLxh3WCRuvpF7dMMPBhvZDQcIaDE7cy7NgvGMROK2G3lpN3sY0owJa7mRY2bA2AbSkmN2m4HhcCJhLfffQLRsngHW8p8ILTd4jB+AtGyQAGs5QISWMzlmDIltEsYzzrwxu9ljkGw8k6CW42eMP3xsq5Ptb88xu/Gjwk62j5AWIGCTSICzDQgrBwHmD8SpGwWjYBSMghELAFotReXf4QCkAAAAAElFTkSuQmCC","orcid":"","institution":"University of Nebraska Medical Center","correspondingAuthor":true,"prefix":"","firstName":"Meng","middleName":"","lastName":"Niu","suffix":""}],"badges":[],"createdAt":"2026-03-02 20:38:02","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9013592/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9013592/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":104342310,"identity":"50940557-b407-43a8-8bef-714c51149fed","added_by":"auto","created_at":"2026-03-10 17:00:34","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":2160473,"visible":true,"origin":"","legend":"\u003cp\u003eWorkflow for detection of HIV RNA–positive cells in human brain single-nucleus RNA-seq data.\u003c/p\u003e","description":"","filename":"figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-9013592/v1/0610230f100d5fa6fad23dcb.png"},{"id":104342322,"identity":"14cb584b-c8b4-4a20-a963-0d15d0a2144e","added_by":"auto","created_at":"2026-03-10 17:00:34","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":2150308,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eA.\u003c/strong\u003e Browser plots of HIV reads aligned to the HIV-1 HXB2 genome detected in subject 4084, with tracks showing from top to bottom: reads from 5’ LTR reference genome method, additional reads retrieved from reference variant update method, all reads from reference variant update method. \u003cstrong\u003eB.\u003c/strong\u003e Variants count and density (normalized by the sequence length of each region) per HIV genomic regions. As we removed the 3' LTR region, we removed part of the nef gene as well, which is identical to 1..332 bp of the 5' LTR, so we added 27 variants from this region to nef.\u003c/p\u003e","description":"","filename":"figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-9013592/v1/c3894e0937cb4be9c69ac98b.png"},{"id":104342312,"identity":"e7c5b4be-efb7-4e93-9e6a-e82b61fb1cd7","added_by":"auto","created_at":"2026-03-10 17:00:34","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":614291,"visible":true,"origin":"","legend":"\u003cp\u003eVenn diagram of HIV RNA-positive cells with HIV UMI \u0026gt; 0 showing overlap between the HIV strain collection method and the combined 5’ LTR method and reference update method.\u003c/p\u003e","description":"","filename":"figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-9013592/v1/336f1ab1d03d4bd23597fefb.png"},{"id":104342314,"identity":"01d477c9-5c1e-4db1-8dcf-2d2388334b7c","added_by":"auto","created_at":"2026-03-10 17:00:34","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":273743,"visible":true,"origin":"","legend":"\u003cp\u003eHistogram of cell count per UMI count for 1,939 HIV RNA-positive cells with at least one HIV UMI.\u003c/p\u003e","description":"","filename":"figure4.png","url":"https://assets-eu.researchsquare.com/files/rs-9013592/v1/8e1f1b0aa0b095cc3dcc5a8f.png"},{"id":104406143,"identity":"3f12bc7d-196d-4fa6-9416-894e443632c6","added_by":"auto","created_at":"2026-03-11 12:24:54","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":843905,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eA.\u003c/strong\u003eAccumulative count of HIV RNA-positive cell number per cell type. \u003cstrong\u003eB.\u003c/strong\u003eBlow up of accumulative count of HIV RNA-positive cell number per cell type after removing microglia.\u003c/p\u003e","description":"","filename":"figure5.png","url":"https://assets-eu.researchsquare.com/files/rs-9013592/v1/ae451497d76e0854bfb379f2.png"},{"id":104342317,"identity":"7bf7899c-d7e9-4ea1-b447-a44c7c04bc2e","added_by":"auto","created_at":"2026-03-10 17:00:34","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":3574317,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eA.\u003c/strong\u003e UMAP projection of 559,207 cells from 22 PWH samples and 27 PWoH samples. The annotation for each cluster is labeled. \u003cstrong\u003eB.\u003c/strong\u003e Bubble plot showing the expression of representative marker genes for each cell type. \u003cstrong\u003eC.\u003c/strong\u003e HIV RNA-positive cell count per cell type.\u003c/p\u003e","description":"","filename":"figure6.png","url":"https://assets-eu.researchsquare.com/files/rs-9013592/v1/679917a64a04428c5a189d8e.png"},{"id":104342321,"identity":"875f1bf8-4a84-4014-853b-8bddbfbfdb4d","added_by":"auto","created_at":"2026-03-10 17:00:34","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":1150788,"visible":true,"origin":"","legend":"\u003cp\u003eBubble plot showing agreement between our manual cell typing and popV cell type predictions, for HIV RNA-positive cells after excluding cells from the outlier sample.\u003c/p\u003e","description":"","filename":"figure7.png","url":"https://assets-eu.researchsquare.com/files/rs-9013592/v1/92d13148755e974b9d4c2648.png"},{"id":104342324,"identity":"6d7b065d-2cc4-4953-b1d4-ee03a07720ea","added_by":"auto","created_at":"2026-03-10 17:00:34","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":34177867,"visible":true,"origin":"","legend":"\u003cp\u003eImmunohistochemical staining for HIV p24 Gag protein in brain tissue from HIVE cases. A) Case 6800127569 with HIV p24-positive cells with the morphological characteristics of macrophages, as well as multinucleated giant cells (red arrows). B) Case 7200597771 with a positive cell with the morphological characteristics of an oligodendrocyte (red arrow).\u003c/p\u003e","description":"","filename":"figure8.png","url":"https://assets-eu.researchsquare.com/files/rs-9013592/v1/abc3b8bd624b005329f6e2d5.png"},{"id":104406161,"identity":"c912123f-1130-49ed-9d60-f11f9f33a966","added_by":"auto","created_at":"2026-03-11 12:24:56","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":1832696,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eA.\u003c/strong\u003e Number of HIVE and non-HIVE samples in this study. \u003cstrong\u003eB.\u003c/strong\u003e Number of infected cells detected with HIV UMI \u0026gt;0, \u0026gt;1 and \u0026gt;2, separated by HIVE and non-HIVE. \u003cstrong\u003eC.\u003c/strong\u003eProportions of HIV RNA-positive cells detected with HIV UMI \u0026gt;0, \u0026gt;1 and \u0026gt;2, separated by HIVE and non-HIVE.\u003c/p\u003e","description":"","filename":"figure9.png","url":"https://assets-eu.researchsquare.com/files/rs-9013592/v1/14cae51e6a609188ed67a7f1.png"},{"id":104784145,"identity":"639f4cb9-cd24-4b20-bc47-5ac5d08ef6a2","added_by":"auto","created_at":"2026-03-17 08:05:27","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":43403889,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9013592/v1/b1ffd8f6-0b50-4187-9d3f-e84bdb68245a.pdf"},{"id":104779965,"identity":"e3aeb8a0-b09f-4689-856a-25e489cb9c2a","added_by":"auto","created_at":"2026-03-17 07:48:36","extension":"xlsx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":10337,"visible":true,"origin":"","legend":"","description":"","filename":"table1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-9013592/v1/2e779160e27a81922972a9d1.xlsx"},{"id":104342319,"identity":"7518aac2-e0de-4476-9f93-e4c571114549","added_by":"auto","created_at":"2026-03-10 17:00:34","extension":"xlsx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":10247,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cbr\u003e\u003c/p\u003e","description":"","filename":"table2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-9013592/v1/8b4ec91831e60be126eddbf8.xlsx"},{"id":104779791,"identity":"2b15dfa1-15bc-40c1-95de-d30407012346","added_by":"auto","created_at":"2026-03-17 07:46:24","extension":"xlsx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":14036,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSupplementary Table 1.\u003c/strong\u003e Cell counts for each of the cell types across PWH samples, PWoH samples, and different thresholds of HIV UMI counts.\u003c/p\u003e","description":"","filename":"supplementarytable1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-9013592/v1/58a6d84fa603ba430f074971.xlsx"},{"id":104406159,"identity":"f86d7799-603f-4086-9163-9d33161f9ed1","added_by":"auto","created_at":"2026-03-11 12:24:56","extension":"xlsx","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":77978,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSupplementary Table 2.\u003c/strong\u003e Master sheet of samples in this study including subject IDs, HIV status, cognitive status, last plasma viral load, brain pathology, brain region, sequencing metrics, and HIV RNA-positive cell counts. The file contains four worksheets: 1) PWH – all files from PWH analyzed, those with HIV RNA-positive cells in red font, light green shading indicates duplicate samples; 2) HIV detected – subset of files from PWH containing HIV RNA-positive cells; 3) PWoH – all files from PWoH analyzed; and 4) PWoH selected – subset of files from PWoH chosen for comparison to PWH HIV detected.\u003c/p\u003e","description":"","filename":"supplementarytable2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-9013592/v1/252d4864ff1bf8ef0113f9d2.xlsx"},{"id":104342323,"identity":"c4f39b47-c039-46be-9755-a7a92d12d34f","added_by":"auto","created_at":"2026-03-10 17:00:34","extension":"docx","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":792263,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSupplementary File 1.\u003c/strong\u003e PopV configuration, quality control procedure, annotation reference datasets, and cross validation.\u003c/p\u003e","description":"","filename":"supplementaryfile1.docx","url":"https://assets-eu.researchsquare.com/files/rs-9013592/v1/17e42601f38c4e70b693192f.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Single-nucleus detection of rare HIV-infected cells defines the cellular landscape of HIV persistence in the human brain","fulltext":[{"header":"BACKGROUND","content":"\u003cp\u003eHuman immunodeficiency virus type 1 (HIV-1) remains a major global health challenge. HIV infection affects both mental and physical health, placing substantial and long-term burdens on people living with HIV, their families, and health systems. In 2024, an estimated 40.8\u0026nbsp;million people were living with HIV worldwide, and only 31.6\u0026nbsp;million were receiving antiretroviral therapy (ART), with ART leading to clinical viral suppression in 95% [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Approximately 630,000 people died from AIDS-related illnesses in 2024, compared with 2.1\u0026nbsp;million in 2004 and 1.4\u0026nbsp;million in 2010, underscoring both the success and the limitations of current treatment [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. As ART has extended survival, more people are living for decades with chronic complications of infection, including HIV-associated neurocognitive disorders (HAND), which range from mild cognitive impairment to severe dementia [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Another term, HIV-associated brain injury (HABI) has been proposed recently to help distinguish legacy damage prior to receiving ART from active damage occurring despite efficacious ART [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Regardless of these concepts, the study of HIV in the brain and its effects are collectively known as NeuroHIV. HIV enters the central nervous system (CNS) early during infection and establishes long-lived reservoirs in brain myeloid cells and other cell types, where it can persist despite systemic viral suppression [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Understanding which CNS cell types harbor virus, how infected cells interact with their local microenvironment, and how these interactions contribute to neurocognitive disorders is essential for developing more effective therapeutic strategies for treating such disorders and for strategies towards a cure for HIV.\u003c/p\u003e \u003cp\u003eThe emergence of single-cell sequencing technologies has opened new possibilities for dissecting these questions in human brain tissue. Unlike bulk RNA sequencing, which averages gene expression across heterogeneous cell populations, single-cell and single-nucleus RNA sequencing measure mRNA at the level of individual cells, providing the resolution needed to study how specific cell types in the brain are infected by, and/or respond to, HIV infection. However, detecting HIV-infected cells in these datasets is technically challenging. Single-cell and single-nucleus experiments capture only a fraction of the transcriptome, on the order of ~\u0026thinsp;30% of mRNA molecules per cell in typical droplet-based experiments, and thus provide only a snapshot of the transcripts present at the time of sampling [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Single-nucleus sequencing, which is widely used for human post-mortem brain, captures even fewer transcripts than whole-cell approaches and yields a higher proportion of intronic and ribosomal reads. In addition, effective ART suppresses viral replication and reduces viral RNA levels in tissues, so that even truly infected cells may express very low amounts of HIV transcripts [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e], and in the case of true latently infected cells no viral transcripts. These factors, together with the rarity of infected cells and low viral copy numbers, make sensitive and specific detection of HIV-infected cells in CNS single-cell datasets particularly difficult and highlight the need for improved analytical strategies.\u003c/p\u003e \u003cp\u003eSingle-cell and single-nucleus approaches are increasingly used in NeuroHIV research, but most studies have focused on host gene-expression changes associated with HIV status, neuropathology, substance use, or ART exposure, rather than on systematic detection of HIV-positive cells [\u003cspan additionalcitationids=\"CR9\" citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Existing reports of HIV-positive cells at single-cell resolution are typically limited to small cohorts, single brain regions, or cerebrospinal fluid, and often rely on study-specific reference genomes, alignment settings, and ad hoc thresholds for calling infected cells [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. As a result, there is currently no standardized, scalable framework for identifying HIV-infected cells across large, heterogeneous CNS single-cell datasets.\u003c/p\u003e \u003cp\u003eTo begin addressing this need, the Single Cell Opioid Responses in the Context of HIV (SCORCH) consortium was established in 2020 with support from the National Institute on Drug Abuse (NIDA). SCORCH aims to generate resources for comprehensive brain tissue characterization at the single-cell level, with a particular focus on single-nucleus transcriptomic data to identify novel rare cell types and to enrich key cell populations, thereby refining our understanding of CNS pathophysiology associated with substance use disorders and HIV. SCORCH is an interdisciplinary effort that brings together fourteen data generating projects, with investigator teams led by principal investigators from nine institutions, along with a Data Coordination Center [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Most of the human patient samples, and all samples from people with HIV (PWH) provided to SCORCH investigators for single-nucleus sequencing, are provided by the National NeuroHIV Tissue Consortium (NNTC). The NNTC was established in 1998 to collect, store, and distribute central and peripheral nervous system tissue, cerebrospinal fluid, blood, and other organs from PWH who have been characterized in longitudinal neuromedical, neuropsychological, and other assessments. This unique combination of rich clinical metadata and high-quality frozen brain tissue has made the NNTC an essential resource for investigating the neurological and neuropsychiatric complications of HIV [\u003cspan additionalcitationids=\"CR13\" citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTo address the gap in standardized viral detection for CNS single-cell datasets, we developed and applied a viral-detection framework for human NeuroHIV single-cell and single-nucleus RNA-seq data that combines modified and variant-updated HIV reference genomes, along with a collection of HIV strains, and explicit criteria for defining HIV RNA\u0026ndash;positive cells. Using single-nucleus RNA-seq data generated by SCORCH from a large NNTC post-mortem cohort spanning multiple brain regions and including both PWH and HIV-negative individuals (people without HIV, PWoH), we systematically quantify the frequency and distribution of HIV-infected cells across cell types and anatomical regions in the human brain. This work provides a reference strategy for detecting rare HIV-infected cells in CNS single-cell datasets and offers new insight into the cellular composition of brain reservoirs.\u003c/p\u003e"},{"header":"METHODS","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eBrain tissue\u003c/h2\u003e \u003cp\u003e Post-mortem brain tissue was obtained by the SCORCH data generating sites from the National NeuroHIV Tissue Consortium (NNTC) and related cohorts under local institutional review board approval and in accordance with ethical guidelines for the use of human tissue in research. Donors included people with HIV (PWH) and HIV-negative individuals (people without HIV, PWoH). Clinical data such as age, sex, HIV status, plasma viral load, antiretroviral therapy (ART) history, and neuropathological diagnoses were collected by the originating NNTC sites.\u003c/p\u003e \u003cp\u003eFor this study, we included samples from anatomically defined brain regions deposited in the SCORCH NeMO database, including cortical, subcortical, and cerebellar areas. Each sample consisted of data obtained from frozen tissue from a single region in a single donor. The initial screening dataset comprised 250 samples from 12 brain regions and 102 donors (163 PWH and 95 PWoH). After applying our viral-detection screening and quality-control criteria, the final dataset used for downstream analysis consisted of 48 samples from 35 donors (17 PWH and 18 PWoH).\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eSingle-cell and single-nucleus RNA sequencing\u003c/h3\u003e\n\u003cp\u003eAt each SCORCH data generating site, nuclei were isolated from frozen tissue, quantified, and nuclei concentrations adjusted to manufacturer-recommended concentrations and loaded onto the Chromium Single Cell Multiome ATAC\u0026thinsp;+\u0026thinsp;Gene Expression platform (10x Genomics, Pleasanton, CA, USA). Nuclei were partitioned into Gel Bead-in-Emulsions for simultaneous transposition of accessible chromatin and barcoding of nuclear RNA, followed by construction of separate ATAC and gene expression libraries according to the manufacturer\u0026rsquo;s instructions. Only the gene expression (GEX) data from the multiome sequencing were used in downstream analyses. Libraries were prepared at the contributing sites, sequenced on Illumina platforms, and raw base call (BCL) files were converted to FASTQ format using cellranger mkfastq or equivalent pipelines.\u003c/p\u003e\n\u003ch3\u003ePreprocessing, quality control, and normalization\u003c/h3\u003e\n\u003cp\u003eSequencing reads from single-nucleus gene expression (GEX) data were downloaded from the NEMO database, aligned to a custom reference genome (see below), and gene-level counts were generated using Cell Ranger (version 8.0.1, 10x Genomics) with default parameters unless otherwise specified. The resulting count matrices were first screened based on whether any HIV reads were detected within Cell Ranger\u0026ndash;called cells.\u003c/p\u003e \u003cp\u003eFor the 22 count matrices (22 samples) in which HIV reads were detected in at least one called cell, we then applied per-cell quality control. Barcodes with low library complexity or evidence of poor quality were removed. Specifically, we retained cells with at least 300 detected genes, a total UMI count above 500, and a mitochondrial RNA fraction below 5%. We did not apply a computational doublet-removal step because HIV encephalitis brain tissue contains activated and multinucleated microglia/macrophages that can be misidentified as doublets.\u003c/p\u003e \u003cp\u003eAfter filtering, counts were normalized and variance-stabilized using SCTransform [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e], and highly variable genes were selected for downstream dimension reduction and clustering. Principal component analysis (PCA) was used for initial dimension reduction. Batch correction across all samples was performed using Harmony [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e] run on PCA of top 1000 genes.\u003c/p\u003e\n\u003ch3\u003eCell-type annotation\u003c/h3\u003e\n\u003cp\u003eTo annotate major CNS cell types, we first performed PCA followed by graph-based clustering. Clustering was carried out using the Louvain algorithm with a resolution of 0.1 and 30 nearest neighbors. We then visualized the resulting clusters using UMAP with settings of 30 neighbors and a minimum distance of 0.3.\u003c/p\u003e \u003cp\u003eFor each cluster, we summarized cell-type\u0026ndash;specific expression by calculating the mean expression of selected marker genes and the percentage of cells expressing each marker. These metrics were used to generate bubble plots to aid visual cell-type identification.\u003c/p\u003e \u003cp\u003eCell-type labels were assigned based on canonical marker genes. Briefly, clusters expressing PLP1 and MBP were annotated as oligodendrocytes; clusters expressing GAD1 and GAD2 as inhibitory neurons; clusters with high SLC17A7 expression as excitatory neurons; clusters expressing PDGFRA and COL9A1 as oligodendrocyte precursor cells; clusters expressing GFAP and AQP4 as astrocytes; clusters expressing CSF1R, P2RY12, and PTPRC as microglia; and clusters expressing CLDN5 and VWF as endothelial cells. Within excitatory and inhibitory neuron clusters, HTR2C expression was used to define HTR2C-positive neuronal subpopulations.\u003c/p\u003e \u003cp\u003eTo validate our manual cell typing results, we used popV (popular Vote), a cell type annotation method that uses Cell Ontology-aware ensemble voting of eight different algorithms (including random forest, support vector machine, XGBoost, OnClass, CellTypist and k-nearest neighbors on several batch-correction embeddings) [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. The method demonstrated ability to reliably transfer labels from an annotated reference dataset to a query dataset. We run 5-fold cross-validation to guarantee stability of predictions and reliability of performance metrics. Along with labels popV also gives prediction scores that we used to create weighted predictions for very rare cases when cross-validation experiment results disagree (Supplementary File 1).\u003c/p\u003e \u003cp\u003eFor the reference dataset, we used our manual labels based on commonly used marker genes, as well as two external sources of data. The first consisted of 17 dissection datasets from Human Brain Cell Atlas (HBCA) v1.0 [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. These HBCA dissection datasets were chosen to span multiple CNS regions, including basal forebrain (septal nuclei), basal nuclei (nucleus accumbens and globus pallidus), hippocampal formation (dentate gyrus/CA4 and subiculum), midbrain (substantia nigra/red nucleus), cerebellum (vermis, deep nuclei, and cerebellar hemisphere), pons, medulla oblongata (including inferior olive and cranial nerve nuclei), perirhinal cortex (A35/A36), and spinal cord. We selected HBCA v1.0 for its large, well-annotated, and widely cited collection of reference cell types, predominantly from three individuals without neurological disorders. Second, we used the Human Microglia Atlas, which separated microglia from a number of datasets from individuals with Alzheimer\u0026rsquo;s disease, autism spectrum disorders, epilepsy, multiple sclerosis, Lewy body disease, and severe COVID-19 [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. Because HIV primarily affects microglia, the latter atlas was essential for providing diverse disease-relevant microglial states that are not represented in healthy-only datasets. Manual labels dataset in the reference excluded one sample that had overwhelming number of HIV+ cells and was considered an outlier (Supplementary File 1).\u003c/p\u003e\n\u003ch3\u003eConstruction of modified and variant-updated HIV reference genomes\u003c/h3\u003e\n\u003cp\u003eStandard reference genomes can under-detect HIV reads because of viral sequence diversity and patient-specific variation. To increase sensitivity for HIV RNA detection, we extended the human reference genome with modified HIV sequences.\u003c/p\u003e \u003cp\u003eWe combined the human reference genome (GRCh38 from Ensembl [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]) with the HIV-1 reference genome HXB2 (GenBank: K03455.1). Because HIV-1 contains two long terminal repeat (LTR) regions with nearly identical sequences, reads mapping to both regions are discarded by the Cell Ranger count pipeline as it excludes multi-mapped reads. To retain reads originating from the LTR, we removed the 3\u0026prime; LTR and kept only the 5\u0026prime; LTR when building the combined reference. This customized Human\u0026thinsp;+\u0026thinsp;HIV reference genome was generated using cellranger mkref and used as the initial screening reference to identify infected cells across all samples.\u003c/p\u003e \u003cp\u003eHIV has one of the highest mutation rates among viruses, and after many rounds of reverse transcription, viral genomes in individual patients can differ substantially from the canonical HXB2 sequence. When reads are aligned only to an unmodified reference, accumulated mismatches can reduce mapping quality and lead to missed HIV reads. To mitigate this, we generated subject-specific, variant-updated HIV references for donors with substantial numbers of infected cells. To ensure sufficient coverage for reliable variant calling, we selected subjects with more than 10 infected cells detected using the modified HXB2 reference, pooled all HIV-aligned reads using samtools (version 1.22), performed variant calling to identify single-nucleotide variants using bcftools (version 1.18), and built a consensus sequence based on the proportion of alternate alleles [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. This consensus HIV genome was then combined with GRCh38 and used with cellranger mkref to create a subject-specific reference genome for re-alignment.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eConstruction of HIV strain collection reference genomes\u003c/h2\u003e \u003cp\u003eA total of 9,444 HIV-1 clade B sequences in FASTA format were downloaded from Los Alamos HIV Sequence Database (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.hiv.lanl.gov/\u003c/span\u003e\u003cspan address=\"https://www.hiv.lanl.gov/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) and used to construct a strain collection reference. STAR(V) genomeGenerate was used to index the sequences and create the a composite HIV strain collection reference for alignment [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eAlignment and quantification of HIV transcripts\u003c/h3\u003e\n\u003cp\u003eFor both the 5\u0026rsquo; LTR-modified reference and the reference update methods, reads were aligned to the corresponding human\u0026thinsp;+\u0026thinsp;HIV reference genomes, and gene-level UMI counts were generated using the standard 10x Genomics cellranger count pipeline. HIV transcripts were quantified by summing UMIs assigned to HIV genomic regions for each cell barcode.\u003c/p\u003e \u003cp\u003eFor the HIV strain collection method, sequencing reads were aligned to HIV-1 clade B collection reference using STAR with multimapping parameters (--outFilterMultimapNmax 1000) to capture candidate viral reads. To reduce false-positive alignments driven by homopolymeric artifacts, reads containing poly-G tails were removed prior to downstream analysis. Viral reads were then collapsed by UMIs to quantify unique viral UMIs Finally, we mapped the cell barcodes associated with these UMIs back to the set of called cells.\u003c/p\u003e \u003cp\u003eIn all samples from PWoH donors, we did not detect any cells with HIV UMIs. For PWH samples, HIV UMI counts per cell were carried forward and integrated with the host gene expression matrix for downstream analyses.\u003c/p\u003e \u003cp\u003e \u003cb\u003eDefinition of HIV RNA\u0026ndash;positive cells.\u003c/b\u003e \u003c/p\u003e \u003cp\u003eA cell was considered HIV RNA\u0026ndash;positive if HIV UMIs were detected using any of the reference strategies described above. For cells processed with more than one method (e.g., generic 5\u0026prime; LTR-modified HXB2, subject-specific variant-updated reference, or the HIV strain collection), we took the maximum HIV UMI count across methods as the value for that cell.\u003c/p\u003e \u003cp\u003eUnless otherwise noted, our primary analyses defined HIV RNA\u0026ndash;positive cells as those with HIV UMI\u0026thinsp;\u0026gt;\u0026thinsp;1. Analyses using UMI\u0026thinsp;\u0026gt;\u0026thinsp;0 and \u0026gt;\u0026thinsp;2 were used to help determine the HIV UMI cutoff (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e9\u003c/span\u003e and Supplementary Table\u0026nbsp;1).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eImmunohistochemistry\u003c/h3\u003e\n\u003cp\u003eImmunohistochemistry for HIV-1 p24 was performed on 5 \u0026micro;M sections of formalin-fixed paraffin-embedded (FFPE) brain sections. Sections were deparaffinized using xylene and rehydrated prior to Tris-EDTA antigen retrieval buffer (10 mM Tris base, 1 mM EDTA solution, 0.05% Tween 20, pH 9.0) (40min, 100\u0026deg;C). Slides were blocked in 5% normal horse serum in PBS, followed by primary antibody (HIV-1 p24, SantaCruz Biotechnology sc-65918 Lot#2325) incubation at a dilution of 1:50 in blocking buffer overnight at 4\u0026deg;C. HRP linked anti-mouse secondary was developed for 6 minutes using ImmPACT NovaRed (Vector SK-4805), followed by a 30 second Gill 2 Hematoxylin nuclear stain (StatLab SL94-16).\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eInfected Cell Screening\u003c/h2\u003e \u003cp\u003eThe screening dataset comprised raw sequencing data files from 250 samples from 12 anatomically defined brain regions across 102 participants, including 156 samples from 73 people with HIV (PWH) and 94 samples from 32 people without HIV (PWoH). The only criteria used for choosing data was availability in the SCORCH consortium as of May 13th 2025, and that the samples were from an NNTC participant (SCORCH also utilizes samples from other studies for PWoH). Using the 10xGenomics cellranger pipeline with the modified 5\u0026rsquo; LTR reference (Human\u0026thinsp;+\u0026thinsp;HIV removing 3\u0026rsquo; LTR), similar to our studies in SIV-infected rhesus monkeys [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e] we first examined the identified total of 1,885,561 cells from 156 PWH samples, and 782,202 cells from 94 PWoH samples. We next restricted the study to 22 PWH samples in which HIV reads were detected and 26 selected PWoH samples, yielding 48 samples from 35 donors (17 PWH, 18 PWoH) and 559,207 high-quality cells (314,409 PWH, 244,798 PWoH) from PWH for downstream analyses. The 26 PWoH samples were selected based on attempting to best match the number of donors and brain regions to confirm major cell types across PWoH and PWH samples.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eInfected cell detection\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e1\u003c/span\u003e summarizes the infected HIV RNA-positive cell detection workflow, which combines three complementary approaches: (i) a modified 5\u0026prime; LTR HIV reference method, here using the HXB2 reference viral sequence (with deletion of the 3\u0026rsquo; LTR) combined with the GRCh38 human genome, (ii) subject-specific variant-updated HIV references (viral reference update method), and (iii) an HIV strain collection reference method. Using the modified 5\u0026prime; LTR HIV reference genome, we identified 1,739 cells with detectable HIV RNA (at least one HIV UMI). These HIV RNA-positive cells were confined to samples from PWH while no HIV-positive cells were detected in PWoH samples at the thresholds used. Six of the positive PWH samples were from the three donors with HIV encephalitis (HIVE) and contained 1,636 infected cells in total, whereas the remaining 16 non-HIVE samples (from 14 donors) together contributed only 103 infected cells. One sample from subject 6800127569 (with HIVE, from the ventral striatum) was particularly enriched with 1,466 infected cells. When this sample was excluded, 255 infected cells remained, corresponding to ~\u0026thinsp;0.05% of all nuclei from 22 PWH samples.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eWhile this study was not designed to compare infection in specific brains regions, since multiple regions were assayed from many of the donors we examined how often the virus was found in one region of the brain while it could be found in a different distinct region in the same donor, and how this related to HIVE and viral suppression. Of the 17 donors in which HIV RNA-positive cells were found, 15 had more than one brain region sampled. Three of the donors had HIVE, and two had more than one region with positive cells, while in total six of the eight brain regions examined had positive cells (67%). Nine donors with more than one region examined did not have viral suppression at their last clinical visit (plasma viral loads ranging from 123 to 441,668 RNA copies/ml, with a median of 1,523), and three had more than one region with HIV RNA-positive cells, resulting in 12 of the 20 brain regions examined having positive cells (60%). The remaining three donors were virally suppressed at the last measurement of HIV load (plasma viral loads\u0026thinsp;\u0026lt;\u0026thinsp;50 RNA copies/ml), and each only had one region with positive cells, resulting in three of the eight regions examined with HIV RNA-positive cells (38%). This relative rarity is also reflected in our finding that multiple regions were assessed from an additional 12 virally suppressed donors (34 samples) with no HIV-infected cells detected. Including another five suppressed cases in which only a single region was examined (all without positive cells), only four HIV-infected cells were found among the 436,304 nuclei examined in virally suppressed PWH (0.001%).\u003c/p\u003e \u003cp\u003eFrom the results of modified 5\u0026rsquo; LTR reference method, we identified six subjects from 10 samples with more than ten infected cells each and applied the viral reference update method to these samples. We reasoned that by improved mapping of the viral reference to the viral sequence in the brain we would identify more reads mapping to HIV, and likely more infected cells. This is similar to the single-cell ECCITE-seq approach described by Collora et al. [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e], where incorporating subject-specific (autologous) HIV-1 sequences in addition to HXB2 for read mapping increased the HIV mapping rate and improved HIV RNA capture/detection of rare HIV RNA-positive cells. We therefore extracted all HIV-aligned reads from the BAM files generated with the modified HXB2 reference, merged reads across regions from the same subject, and performed variant calling using bcftools. As summarized in Table\u0026nbsp;1, subjects with more infected cells yielded more called variants. Variants with an alternate-allele frequency\u0026thinsp;\u0026gt;\u0026thinsp;50% were incorporated into a subject-specific HXB2 consensus, which was then combined with GRCh38 to build an updated reference genome for each subject. Re-alignment against these subject-specific references recovered an additional 162 infected cells in 6 of the 10 samples (the remaining 4 samples did not yield additional infected cells), resulting in 1,895 cell barcodes with associated HIV UMI counts over 1 when combined with the original 5\u0026prime; LTR-based calls. Visualization of representative samples from different brain regions of subject 4084 in the genome browser confirmed improved coverage and alignment in regions with subject-specific variation (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eUsing the HIV strain collection reference method in which reads were mapped to a panel of 9444 HIV strain reference genomes [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e] we independently detected 1,739 cells with at least one HIV UMI. As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e3\u003c/span\u003e, \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e1\u003c/span\u003e,696 HIV RNA-positive cells were identified by both approaches (here, the 5\u0026prime; LTR-modified reference plus variant-updated reference are grouped as \u0026ldquo;5\u0026rsquo; LTR+RefUpdated\u0026rdquo;), while 43 cells were uniquely identified by the strain-collection method and 199 cells were uniquely identified by the 5\u0026rsquo; LTR_RefUpdated approach. Thus, the modified 5\u0026prime; LTR plus variant-updated references captured most HIV RNA-positive cells, while the strain-collection reference contributed to a small but non-negligible set of additional cells. Taking the union of HIV RNA-positive cells identified by all three methods, yielded a final set of 1,939 HIV RNA-positive cells with HIV UMI\u0026thinsp;\u0026gt;\u0026thinsp;0. For each cell, we assigned the maximum HIV UMI count observed across the three methods as its HIV UMI value for downstream analyses.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAs shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e4\u003c/span\u003e, HIV RNA was detected at low copy number in most HIV RNA-positive cells (1,014 cells with 1 HIV UMI, 180 cells with 2 HIV UMIs), with a sharp decline in cell counts beyond HIV UMI\u0026thinsp;\u0026ge;\u0026thinsp;2, consistent with a long-tailed distribution of viral RNA burden. When stratified by annotated cell types (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e5\u003c/span\u003e), cumulative infected-cell counts as a function of the HIV UMI threshold displayed clear saturation behavior for most populations, indicating that increasing stringency rapidly prunes low-read HIV-positive cells. Notably, microglia \u0026mdash; the most abundant HIV RNA-positive cell population \u0026mdash; showed a slower decline and remained detectable at substantially higher HIV UMI levels (with the increase in retained infected cells beginning to plateau at higher thresholds), whereas other cell types saturated much earlier (e.g., endothelial cells, oligodendrocytes, and neurons). Despite the known limitation that snRNA-seq captures only a fraction of cellular RNA and is particularly insensitive to low-level signal, these patterns are consistent with cell-type-specific differences in HIV RNA burden, with microglia harboring the highest levels, followed by oligodendrocytes and astrocytes, and lower levels in endothelial cells and neurons (with most HIV RNA-positive cells originating from HIVE samples).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTo reduce the impact of potential false positives due to free-floating/ambient viral RNA and background alignment, we used HIV UMI\u0026thinsp;\u0026gt;\u0026thinsp;1 as the primary threshold in further analyses to define infection in the remaining cells. After additionally applying per-cell quality filters (total UMI counts and mitochondrial read percentage), 908 HIV RNA-positive cells remained for downstream analyses.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eDistribution of infected cells across cell types\u003c/h2\u003e \u003cp\u003eUsing the cell-by-gene expression matrix from the 48 selected samples (combining those from PWH and PWoH), we confirmed expected expression of major cell-type and subtype marker genes and assigned cluster identities. As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e6\u003c/span\u003eA\u0026ndash;B, among 559,207 cells, 171,604 (30.69%) were identified as oligodendrocytes based on MBP and PLP1 expression, and 77,356 (13.83%) were annotated as astrocytes based on AQP4 and GFAP. SNAP25 was used as a general neuronal marker, and neurons were further divided into excitatory and inhibitory populations, with or without expression of the serotonin receptor gene HTR2C. This resulted in five excitatory neuron subtypes, two inhibitory neuron subtypes, two HTR2C⁺ inhibitory neuron subtypes, and one HTR2C⁺ excitatory neuron subtype, with a total of 201,823 (36.09%) neurons. In addition, 48,125 cells (8.61%) were classified as microglia based on the expressions of CSF1R, P2RY12, and PTPRC.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAs summarized in Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e6\u003c/span\u003eC and Supplementary Table\u0026nbsp;1, in the infected cell pool, 78.6% of HIV RNA-positive cells were microglia, representing the largest infected cell population, followed by oligodendrocytes (6.06%) and astrocytes (4.85%); 5.73% of HIV RNA-positive cells (52 of 908) were classified as neurons (combining all neuron subtypes). In the total PWH cell pool of the 22 samples with HIV RNA-positive cells detected, 2.23% of microglia were HIV RNA-positive, followed by 0.24% of endothelial cells, 0.1% of astrocytes, and 0.05% of the combined neuronal cell types. The total detectable HIV RNA-positive rate of all cells was 0.29%. While excluding sample 6800127569_ventral_striatum with the exceptionally large number of HIV RNA-positive cells, 0.38% of microglia, 0.02% of oligodendrocytes, 0.01% of astrocytes, and 0.02% of the combined neuronal cell types were HIV RNA-positive. The total infection rate represented by HIV RNA-positive cells was reduced to 0.05% (Supplementary Fig.\u0026nbsp;1). If we consider all PWH samples (including those without HIV RNA-positive cells detected, a total of 156 samples with 1,885,561 cells), the infection rates became 0.05% of all cells, dropping to 0.01% when excluding sample 6800127569_ventral_striatum.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eValidation of Cell type annotation\u003c/h2\u003e \u003cp\u003eTo validate our manual cell typing results, we used popV (popular Vote), a cell type annotation method that uses Cell Ontology-aware ensemble voting of eight different algorithms with label transfer from multiple reference datasets [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Table\u0026nbsp;2 shows agreement between manual annotation and popV results represented on cells with HIV UMI detected, represented in classification metrics, giving us confidence in our cell type annotation. In Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e7\u003c/span\u003e, the bubble heatmap shows how the weighted predicted labels correspond to our manually annotated HIV-infected cells. The color of the bubbles shows the average popV prediction score, indicates how many annotation methods support a cell\u0026rsquo;s final consensus type, including votes propagated to ancestor types in the ontology. In cross-validation, by combining these scores across experiments we computed weighted final cell type predictions. We also applied the popV to all the data sets that we used for screening, as shown in Supplementary File 1, cross validation of popV showed great performance and stability of the cell annotation across experiments, data sources, cell types. N/A means that a cell type is not represented in manual labels but is present in reference datasets.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn tissue from cases with HIVE, as expected, finding HIV-infected cells using immunohistochemistry easily identified HIV p24-positive cells resembling microglia and macrophages, as well as the pathognomonic multinucleated giant cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e8\u003c/span\u003eA). However, in non-HIVE cases detection of positive cells was difficult and were only rarely seen, including in a donor in which infected oligodendrocytes were identified, in which a positive cell with the morphological characteristics of an oligodendrocyte was found (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e8\u003c/span\u003eB). Due to the rarity of such cells, double labeling for cell-type markers could not be performed.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eHIV encephalitis versus non-encephalitis across cell types\u003c/h2\u003e \u003cp\u003eBeyond the total number of infected cells per UMI count identified in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e4\u003c/span\u003e, we examined how the distribution of HIV RNA-positive cells across cell types changed under different HIV UMI thresholds. As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e9\u003c/span\u003e and Supplementary Table\u0026nbsp;1, when HIV UMI\u0026thinsp;\u0026gt;\u0026thinsp;0, 19.5% of infected cells (371 cells) were oligodendrocytes and 13.83% (213 cells) were astrocytes. When we applied a more stringent cutoff of HIV UMI\u0026thinsp;\u0026gt;\u0026thinsp;1, these proportions dropped to 6.06% (55 cells) for oligodendrocytes and 4.85% (44 cells) for astrocytes. A similar trend was observed for most other non-microglial cell types. In contrast, microglia showed a different pattern. Among microglia, 870 cells had HIV UMI\u0026thinsp;\u0026gt;\u0026thinsp;0, 714 had HIV UMI\u0026thinsp;\u0026gt;\u0026thinsp;1, and 638 had HIV UMI\u0026thinsp;\u0026gt;\u0026thinsp;2, indicating a relatively modest decrease as thresholds became more stringent. This slower decline in microglial infection frequency is consistent with microglia being much more susceptible to HIV infection and expression than other CNS cell types.\u003c/p\u003e \u003cp\u003eWhen we stratified cells by the presence or absence of HIV encephalitis (HIVE) (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e9\u003c/span\u003e) the proportions of infected cells in most cell types decreased or disappeared as the HIV UMI threshold increased in the non-HIVE group. As expected, most infected cells originated from HIVE cases, reflecting the abundant expression of HIV transcripts in encephalitic tissue. Notably, however, even within the smaller pool of infected cells in non-HIVE cases, microglia consistently constituted a large proportion of infected cells, reinforcing their role as a principal CNS HIV reservoir [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Together, these patterns indicate that in HIVE, HIV RNA-positive cells are distributed across multiple CNS lineages, whereas in the absence of HIVE, detectable infection appears largely restricted to microglia, with only rare oligodendrocyte or astrocyte involvement.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eLocation of HIV sequence variants\u003c/h2\u003e \u003cp\u003eUsing the reference-updated mapping strategy, we found that most HIV-derived reads localized to the U3 region of the 5\u0026prime; LTR and to the env gene ingp120 and gp41 (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e2\u003c/span\u003e). The U3 region encodes key cis-regulatory elements, including the core promoter, enhancer, and multiple transcription factor binding sites (e.g., NF-κB, Sp1) that control viral transcriptional activation and integration site\u0026ndash;dependent expression [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. These functional constraints, together with host- and tissue-specific selective pressures on transcriptional regulation, may contribute to the accumulation of sequence diversity in U3, especially in enhancer and promoter modules that modulate replication fitness in different cellular environments, including CNS myeloid cells. Likewise, env, particularly gp120 and the extracellular portion of gp41, is under strong immune and receptor-mediated selection, as these regions mediate CD4/coreceptor binding and membrane fusion and are primary targets of neutralizing antibodies [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Thus, the high density of variants we observed in U3 and env likely reflects a combination of their central regulatory and entry functions and the strong, ongoing selection exerted by the host environment in the CNS.\u003c/p\u003e \u003c/div\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eIn this study, we developed a scalable, multi-reference framework for detecting rare HIV RNA\u0026ndash;positive cells in human CNS single-nucleus RNA-seq data. By combining the modified 5\u0026prime; LTR reference method, viral reference update method, and the HIV strain collection method, we identified a high-confidence set of HIV RNA\u0026ndash;positive cells in post-mortem brains from people with HIV. Within this framework, we show that microglia constitute the predominant infected population, with oligodendrocytes, astrocytes, and neurons contributing smaller fractions largely restricted to cases with HIV encephalitis (HIVE). These results provide an innovative approach and a quantitative baseline for studying HIV-infected CNS cell types in NeuroHIV single-cell datasets.\u003c/p\u003e \u003cp\u003eInfected cells were also detected in several non-microglial lineages, including neurons, oligodendrocytes, and astrocytes, but this pattern was strongly dependent on neuropathological state. In HIVE cases, HIV RNA-positive nuclei were distributed across multiple CNS cell types, with clear representation in oligodendrocytes and astrocytes and occasional infected neurons, consistent with postmortem and genomic studies showing that, although microglia are the dominant target, astrocytes and neurons can occasionally harbor HIV DNA or RNA in encephalitic brain tissue [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. In contrast, in non-HIVE tissue, infected cells were almost exclusively confined to microglia, with only a small fraction observed in oligodendrocytes and virtually no signal in neurons or astrocytes. This is aligned with recent work demonstrating that microglia constitute the principal and most persistent HIV reservoir in the brain across disease stages and even during suppressive ART, whereas evidence of infection of other CNS resident cells appears less frequent [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. The restriction of detectable infection to microglia (and to a lesser extent oligodendrocytes) in non-HIVE cases versus other cell types found in HIVE likely reflects the high level of virus, inflammation, and cellular damage in encephalitis. present in the brain. In the absence of encephalitis and especially in the case of systemic viral suppression, microglia maintain their role as the dominant CNS reservoir while infection of other lineages is rare at best. The complete absence of HIV UMIs in PWoH donors, together with the strong enrichment of HIV UMIs in microglial clusters rather than diffusely across cell types, supports the specificity of our cell-calling and viral transcript identification strategies.\u003c/p\u003e \u003cp\u003eIn this study, we analyzed single-nucleus derived gene expression (snRNA-seq) to detect HIV-associated transcripts at single-cell resolution. This modality primarily captures the transcriptional footprint of infection including viral RNA abundance, cellular activation states, and host-response programs. However, we acknowledge some limitations. In particular, snRNA-seq does not directly interrogate stages of the viral life cycle. Nuclear profiling can be biased towards different splicing stages of mRNA, and in the case of HIV, once the Rev protein is made it functions to transport unspliced and singly spliced RNA from the nucleus to the cytoplasm [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], increasing the problem of technical dropout. Combined, these factors limit sensitivity for detecting low-level or latent infection. Recent conceptual frameworks emphasize that understanding CNS HIV persistence requires integrating viral transcription, chromatin accessibility, and cell-state\u0026ndash;specific immune programs, rather than relying on viral detection alone [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. In future work, integrating the paired snATAC-seq modality will enable direct assessment of LTR/proviral accessibility and surrounding regulatory landscapes that govern latency versus reactivation, and will provide additional context for integration-site preferences in relation to open chromatin and active gene neighborhoods [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. Finally, infected-cell detection was highly imbalanced across pathological groups (e.g., substantially more HIV positive cells in HIVE than in non-HIVE), limiting our ability to characterize virus-positive states in non-HIVE cases, particularly in ART-suppressed individuals where viral transcription is expected to be low [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. The number of cells sampled in single cell analyses is a large factor, limited largely by cost, as given the low infection rate in the absence of HIVE detection is largely stochastic. Addressing these limitations will require increased sampling of brains from brains without HIV, ART-treated donors and/or complementary enrichment and validation strategies (e.g., targeted HIV RNA capture, proviral DNA/integration assays, and orthogonal in situ detection) to improve sensitivity and confirm cell-type attribution.\u003c/p\u003e \u003cp\u003eBeyond these biological and sampling constraints, the practical details of reference construction, alignment, and thresholding can strongly influence what is called \u0026ldquo;HIV RNA\u0026ndash;positive\u0026rdquo; in snRNA-seq.\u0026nbsp;HIV-1 mapping is particularly sensitive to reference design because duplicated regions (most notably the two LTRs) create multi-mapping; reads that align equally well to both ends can be down-weighted or discarded by standard quantification pipelines. To reduce this ambiguity while preserving sensitivity, we retained a single LTR in the viral reference (keeping the 5\u0026prime; LTR and removing the 3\u0026prime; LTR), which makes LTR-derived signal easier to interpret\u0026mdash;an important consideration when many infected nuclei are supported by only a few viral UMIs. A second practical issue is viral diversity: even modest divergence from a canonical reference can lower alignment confidence for short reads and low-abundance transcripts. When we had sufficient HIV-aligned reads to support variant calling, subject-specific variant-updated references improved matching to within-host sequence, and mapping to a curated strain collection provided a complementary way to accommodate diversity without requiring de novo assembly for every donor. Because viral signal is sparse, thresholding is also not a minor detail; we recommend reporting infected-cell counts and cell-type proportions across simple UMI cutoffs (e.g., HIV UMI\u0026thinsp;\u0026gt;\u0026thinsp;0, \u0026gt;1, \u0026gt;\u0026thinsp;2) and then choosing a primary cutoff based on the analytic goal, using higher stringency when specificity and cross-sample comparability are priorities while still showing the full sensitivity series for transparency. Finally, rare-event detection is vulnerable to technical artifacts (ambient RNA, sporadic misalignment, barcode-related effects), so negative controls processed through the identical pipeline provide a direct estimate of background; in our dataset, we did not detect HIV UMIs in samples from individuals without HIV at the thresholds used. To make the approach easy to reproduce, we report software versions, reference-build steps, and key alignment/counting parameters.\u003c/p\u003e"},{"header":"CONCLUSION","content":"\u003cp\u003eIn summary, by combining modified 5\u0026prime; LTR-based references, subject-specific variant-updated HIV genomes, and a comprehensive HIV strain collection method, we developed a scalable framework for detecting rare HIV RNA\u0026ndash;positive cells in human CNS single-nucleus datasets. We recommend starting with a custom genome combining the human reference genome (currently GRCh38 from Ensembl) with the modified HIV-1 reference genome HXB2 (NCBI GenBank: K03455.1), and analyzing the sequencing reads with Cell Ranger. If additional sensitivity is desired, use of the HIV strain collection, and if sufficient HIV UMIs are detected, the subject-specific variant updated HIV genome. The UMI threshold can be adjusted depending on the neuropathology (e.g., HIVE) and objective of the study. When we applied this pipeline to 314,409 nuclei from 22 post-mortem brain samples from PWH, and required\u0026thinsp;\u0026gt;\u0026thinsp;1 UMI, this approach identified 908 HIV RNA-positive cells and revealed that microglia constitute the predominant infected population, with oligodendrocytes, astrocytes, and neurons contributing smaller fractions, particularly in HIVE cases. The strong enrichment of infected microglia in both HIVE and non-HIVE tissue, together with the more restricted involvement of other lineages outside of encephalitic brains, reinforces the concept of microglia as a principal and persistent brain reservoir of HIV. Our variant analyses further show that HIV reads in CNS tissue are concentrated in the U3 region of the LTR and in env (gp120/gp41), highlighting regulatory and entry-associated regions as focal points of viral diversity in the brain. Although our study is constrained by the inherent sparsity of single-nucleus RNA-seq, the cross-sectional nature of post-mortem tissue, and the focus on viral RNA rather than proviral DNA, the framework and reference resources presented here provide a roadmap for harmonized viral detection in NeuroHIV single-cell studies. Future work integrating viral detection and genotypes with longitudinal clinical data, proviral sequencing, and functional perturbation of infected microglia will be essential to translate these insights into strategies for monitoring and targeting CNS HIV reservoirs.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eHuman post-mortem tissues and associated de-identified metadata were obtained by SCORCH investigators from the National NeuroHIV Tissue Consortium (NNTC). Work on decedents is not considered human subject research. Consents for participation in the NNTC studies were through multiple IRB protocols since 1999 to the present time, approved by the IRBs at the clinical sites (\u003cu\u003eUTMB, UCSD, UCLA, ISMMS).\u003c/u\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData presented in this study were produced as part of the Single Cell Opioid Response in the Context of HIV consortium (SCORCH: RRID:SCR_022600). Publicly accessible data is available at NeMO Archive (RRID:SCR_002001) under identifier nemo:col-12t9m0h (https://assets.nemoarchive.org/col-12t9m0h). Access to all protected data associated with this study is managed by dbGaP and can be requested with the identifier phs003991.v1.p1 .\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by NIDA 1U01DA053624.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026rsquo; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eHSF, MN, YC, NJ was responsible for the conceptualization of the study. Single-cell analyses were performed by MN, YC, NJ and DM. Visualization of the data was performed by MN, CY, DM. Supervision of the study was the responsibility of HSF, DG and MN. The original draft of the manuscript was prepared by HSF, MN, and all authors (MN, HSF, YC, NJ, DG, MC, BH, DV, JR, DM, OW, YK, and CC) reviewed, edited, and approved the final version.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe single-nucleus data examined were generated by members of the SCORCH consortium: Y-SCORCH (S. Spudich, M. Gerstein and Y. Kluger, PIs, NIH grant UM1DA051410), M-SCORCH (Y-C Ho, PI, NIH grant U01DA053628), BROAD-SCORCH (M, Kellis and M. Heiman, PIs, NIH grant U01DA053631), UCSD SCORCH (T. Rana, PI, NIH grant U01DA053630), Weill Cornell SCORCH (H. Tilgner, T. Milner, and L. Ndhlovu, PIs, NIH grant U01DA053625). Data sets were curated and provided by the SCORCH Data Coordinating Center (O. White, S. Ament, and A. Mahurkar, PIs, NIH grant UM1DA052244). The brain tissue specimens and associated metadata were provided by the National NeuroHIV Tissue Consortium (NNTC), in which the member institutions have NIH contracts supported by multiple institutes (NIMH, NIA, NIDA, and NINDS) to support the their role in the NNTC, with the following contract numbers: Texas NeuroAIDS Research Center (TNRC), PI B. Gelman, University of Texas Medical Branch, Galveston: 75N95023C00016; California NeuroAIDS Tissue Network (CNTN), PI D. Moore, University of California, San Diego: 75N95023C00014; National Neurological AIDS Bank (NNAB), PI E. Singer, University of California, Los Angeles: 75N95023C00017; Manhattan HIV Brain Bank (MHBB), PI S. Morgello, Icahn School of Medicine at Mt. Sinai, New York, NY: 75N95023C00015; Data Coordinating Center (DCC), PI S. Sherman, Emmes Company, Rockville, MD: 75N95023C00013.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eWHO. HIV data and statistics. 2025 [cited 2025 Dec. 16]; Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.who.int/teams/global-hiv-hepatitis-and-stis-programmes/hiv/strategic-information/hiv-data-and-statistics#:~:text=People%20living%20with%20HIV,HIV%2Drelated%20causes%20in%202024\u003c/span\u003e\u003cspan address=\"https://www.who.int/teams/global-hiv-hepatitis-and-stis-programmes/hiv/strategic-information/hiv-data-and-statistics#:~:text=People%20living%20with%20HIV,HIV%2Drelated%20causes%20in%202024\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHeaton RK, et al. HIV-associated neurocognitive disorders persist in the era of potent antiretroviral therapy: CHARTER Study. Neurology. 2010;75(23):2087\u0026ndash;96.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNightingale S, et al. Cognitive impairment in people living with HIV: consensus recommendations for a new approach. Nat Rev Neurol. 2023;19(7):424\u0026ndash;33.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePlaza-Jennings A, Akbarian S. Genomic Exploration of the Brain in People Infected with HIV-Recent Progress and the Road Ahead. Curr HIV/AIDS Rep. 2023;20(6):357\u0026ndash;67.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZheng GX, et al. Massively parallel digital transcriptional profiling of single cells. Nat Commun. 2017;8:14049.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e10XGenomics. What fraction of mRNA transcripts are captured per cell?. 2025 [cited 2025 Dec. 16]; Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://kb.10xgenomics.com/s/article/360001539051-What-fraction-of-mRNA-transcripts-are-captured-per-cell\u003c/span\u003e\u003cspan address=\"https://kb.10xgenomics.com/s/article/360001539051-What-fraction-of-mRNA-transcripts-are-captured-per-cell\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eViswanathan S, et al. Level of adherence and HIV RNA suppression in the current era of highly active antiretroviral therapy (HAART). AIDS Behav. 2015;19(4):601\u0026ndash;11.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePlaza-Jennings AL, et al. HIV integration in the human brain is linked to microglial activation and 3D genome remodeling. Mol Cell. 2022;82(24):4647\u0026ndash;e46638.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFarhadian SF et al. Single-cell RNA sequencing reveals microglia-like cells in cerebrospinal fluid during virologically suppressed HIV. JCI Insight, 2018. 3(18).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWilson AM, et al. Transcriptional impacts of substance use disorder and HIV on human ventral midbrain neurons and microglia. Nat Commun. 2025;16(1):9123.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAment SA, et al. The single-cell opioid responses in the context of HIV (SCORCH) consortium. Mol Psychiatry. 2024;29(12):3950\u0026ndash;61.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMorgello S, et al. The National NeuroAIDS Tissue Consortium: a new paradigm in brain banking with an emphasis on infectious disease. Neuropathol Appl Neurobiol. 2001;27(4):326\u0026ndash;35.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHeithoff AJ et al. \u003cem\u003eThe integrated National NeuroAIDS Tissue Consortium database: a rich platform for neuroHIV research.\u003c/em\u003e Database (Oxford), 2019. 2019.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCserhati MF et al. \u003cem\u003eThe National NeuroAIDS Tissue Consortium (NNTC) Database: an integrated database for HIV-related studies.\u003c/em\u003e Database (Oxford), 2015. 2015: p. bav074.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu X, et al. Transformation of brain myeloid cell populations by SIV in rhesus macaques revealed by multiomics. Commun Biol. 2025;8(1):100.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCollora JA, et al. Single-cell multiomics reveals persistence of HIV-1 in expanded cytotoxic T cell clones. Immunity. 2022;55(6):1013\u0026ndash;e10317.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWei Y, et al. Single-cell epigenetic, transcriptional, and protein profiling of latent and active HIV-1 reservoir revealed that IKZF3 promotes HIV-1 persistence. Immunity. 2023;56(11):2584\u0026ndash;e26017.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eErgen C, et al. Consensus prediction of cell type labels in single-cell data with popV. Nat Genet. 2024;56(12):2731\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWallet C, et al. Microglial Cells: The Main HIV-1 Reservoir in the Brain. Front Cell Infect Microbiol. 2019;9:362.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTang Y et al. Brain microglia serve as a persistent HIV reservoir despite durable antiretroviral therapy. J Clin Invest, 2023. 133(12).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCullen BR. Regulation of HIV-1 gene expression. FASEB J. 1991;5(10):2361\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWyatt R, Sodroski J. The HIV-1 envelope glycoproteins: fusogens, antigens, and immunogens. Science. 1998;280(5371):1884\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTrillo-Pazos G, et al. Detection of HIV-1 DNA in microglia/macrophages, astrocytes and neurons isolated from brain tissue with HIV-1 encephalitis by laser capture microdissection. Brain Pathol. 2003;13(2):144\u0026ndash;54.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDonoso M et al. Identification, Quantification, and Characterization of HIV-1 Reservoirs in the Human Brain. Cells, 2022. 11(15).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJoseph SB, et al. HIV-1 target cells in the CNS. J Neurovirol. 2015;21(3):276\u0026ndash;89.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCullen BR. Nuclear mRNA export: insights from virology. Trends Biochem Sci. 2003;28(8):419\u0026ndash;24.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFilippidis P, Corley MJ. Single cell analyses of the HIV reservoir in the CNS and CSF: recent insights and implications. Curr Opin HIV AIDS. 2025;20(5):493\u0026ndash;501.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMbonye U, Karn J. The Molecular Basis for Human Immunodeficiency Virus Latency. Annu Rev Virol. 2017;4(1):261\u0026ndash;85.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchroder AR, et al. HIV-1 integration in the human genome favors active genes and local hotspots. Cell. 2002;110(4):521\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBruner KM, et al. A quantitative approach for measuring the reservoir of latent HIV-1 proviruses. Nature. 2019;566(7742):120\u0026ndash;5.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHafemeister C, Satija R. Normalization and variance stabilization of single-cell RNA-seq data using regularized negative binomial regression. Genome Biol. 2019;20(1):296.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKorsunsky I, et al. Fast, sensitive and accurate integration of single-cell data with Harmony. Nat Methods. 2019;16(12):1289\u0026ndash;96.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSiletti K, et al. Transcriptomic diversity of cell types across the adult human brain. Science. 2023;382(6667):eadd7046.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMartins-Ferreira R, et al. The Human Microglia Atlas (HuMicA) unravels changes in disease-associated microglia subsets across neurodegenerative conditions. Nat Commun. 2025;16(1):739.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDyer SC, et al. Ensembl 2025. Nucleic Acids Res. 2025;53(D1):D948\u0026ndash;57.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDanecek P et al. Twelve years of SAMtools and BCFtools. Gigascience, 2021. 10(2).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDobin A, et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics. 2013;29(1):15\u0026ndash;21.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTables are available in the Supplementary Files section.\u003c/p\u003e\n"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"virology-journal","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"virj","sideBox":"Learn more about [Virology Journal](http://virologyj.biomedcentral.com/)","snPcode":"12985","submissionUrl":"https://submission.nature.com/new-submission/12985/3","title":"Virology Journal","twitterHandle":"@VirologyJ","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"HIV-1, single-nucleus RNA-seq, brain, microglia, viral read mapping, reference genome, NeuroHIV, HIV encephalitis","lastPublishedDoi":"10.21203/rs.3.rs-9013592/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9013592/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eHIV-1 enters the central nervous system early after infection and establishes a long-lived reservoir that persists despite antiretroviral therapy. Single-cell and single-nucleus RNA sequencing provide powerful approaches to study HIV infection in the human brain, yet standardized and sensitive methods for identifying rare HIV-infected cells in these datasets remain limited. Here, we present a scalable multi-reference framework for detecting HIV RNA\u0026ndash;positive cells in human CNS single-nucleus RNA-seq data. The pipeline integrates a modified HIV reference genome, subject-specific variant-updated HIV references, and a comprehensive HIV strain collection to improve viral read recovery and specificity.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eWe applied this framework to 250 post-mortem brain samples from the SCORCH (Single Cell Opioid Responses in the Context of HIV) consortium spanning 12 brain regions and 102 donors, including people with and without HIV (PWH and PWoH). After screening, 48 samples from 35 donors comprising 559,207 high-quality nuclei were analyzed in depth. We identified 1,939 HIV RNA\u0026ndash;positive cells exclusively in samples from PWH. Using conservative thresholds, 908 high-confidence infected cells were retained for downstream analyses. HIV RNA-positive cells were rare overall and strongly enriched in cases with HIV encephalitis. Microglia constituted the predominant infected population (79% of HIV RNA-positive cells), with substantially smaller contributions from oligodendrocytes, astrocytes, and neurons. In non-encephalitic brains, detectable infection was largely restricted to microglia, whereas in encephalitic tissue HIV RNA\u0026ndash;positive cells were distributed across multiple CNS (Central Nervous System) lineages. Viral RNA burden followed a long-tailed distribution, with microglia retaining higher HIV transcript counts than other cell types. Recovered HIV reads were concentrated in the U3 region of the 5\u0026prime; LTR and in the env gene, implicating regulatory and entry-associated regions as focal points of viral diversity in the brain.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eTogether, these data establish a harmonized framework for identifying rare HIV-infected cells in CNS single-cell datasets and provide large-scale quantitative evidence that microglia represent the dominant and most persistent HIV-infected population in the human brain. This work offers a reference strategy and resource for future NeuroHIV studies aimed at defining, monitoring, and ultimately targeting CNS viral reservoirs.\u003c/p\u003e","manuscriptTitle":"Single-nucleus detection of rare HIV-infected cells defines the cellular landscape of HIV persistence in the human brain","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-10 17:00:29","doi":"10.21203/rs.3.rs-9013592/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-05-19T07:11:01+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-13T13:24:32+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"141876302812645423477349438980916792782","date":"2026-04-20T12:53:38+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"210599646873734723424099315522858665443","date":"2026-04-20T08:00:44+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-03-26T10:41:42+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-03T14:46:45+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-03T07:21:14+00:00","index":"","fulltext":""},{"type":"submitted","content":"Virology Journal","date":"2026-03-02T20:19:45+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"virology-journal","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"virj","sideBox":"Learn more about [Virology Journal](http://virologyj.biomedcentral.com/)","snPcode":"12985","submissionUrl":"https://submission.nature.com/new-submission/12985/3","title":"Virology Journal","twitterHandle":"@VirologyJ","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"1efb6c99-ef88-44f0-9fbe-79d5765b4520","owner":[],"postedDate":"March 10th, 2026","published":true,"recentEditorialEvents":[{"type":"editorInvitedReview","content":"","date":"2026-05-19T07:11:01+00:00","index":61,"fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-13T13:24:32+00:00","index":60,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-03-26T10:53:56+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-10 17:00:29","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9013592","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9013592","identity":"rs-9013592","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}