Abstract Background CNS infection by HIV-1 contributes to neuroinflammation, cognitive impairments, and the establishment of viral reservoirs. Although HIV-1 is known to enter the brain early in infection via “Trojan horse” leukocytes, including infected monocytes and CD4⁺ T cells, the specific cellular phenotypes facilitating this process during acute infection remain incompletely characterized. Objective This study aims to characterize the roles of brain lymphocytes during acute infection and primary CD4 + T cell phenotypes seeding the SIV to the CNS. Methods scRNA-seq was performed on brain and blood cells of three acutely SIV-infected rhesus macaques. The transcriptomic data were analyzed using bioinformatics approaches and validated through in vitro co-culture assays and re-analysis of a publicly available scRNA-seq dataset. Results scRNA-seq of brain and blood immune cells from acutely SIV-infected rhesus macaques revealed an expansion of proliferating CD4⁺ cytotoxic T lymphocytes (CTLs) in the blood, characterized by high CD4, CCR5, and adhesion molecule expression, indicating strong potential for CNS infiltration. In the brain, CD4⁺ CTLs, tissue-resident memory cells, and a unique Myeloid–T cell cluster were enriched for SIV⁺ cells. Integration of brain and blood data revealed transcriptomic maturation of CD4⁺ CTLs upon brain entry. To validate the biological relevance of the Myeloid–T cluster, we used a macrophage–T cell co-culture system, which reproduced similar dual-marker expression and identified chemokines (e.g., CCL3, CCL4) as potential markers of T cell–myeloid cell interaction. Conclusion Our findings suggest that CD4⁺ cytotoxic-like T cells represent a key lymphocyte subset responsible for initiating SIV entry into the brain and triggering neuroinflammatory processes. Furthermore, interactions between infiltrating lymphocytes and brain-resident myeloid cells, potentially through chemokine signaling, may facilitate viral propagation within the CNS. Supplementary Information The online version contains supplementary material available at 10.1186/s12974-025-03543-y. Keywords: CD4 + T cells, Acute SIV infection, ScRNA-seq, Viral entry Introduction Human Immunodeficiency Virus (HIV) is an enveloped retrovirus that harbors two copies of a single-stranded RNA genome. It leads to acquired immunodeficiency syndrome (AIDS) by severely compromising the immune system. HIV continues to pose a significant global health challenge, affecting individuals, communities, and societies at large. As of 2022, an estimated 39 million people worldwide are living with HIV (PWH) [[35]1]. As of the latest epidemiological study in 2021, there are approximately 1.2 million PWH in the United States [[36]2]. HIV invades the brain during the acute phase of the systemic infection and can establish a latent reservoir pool as well as lead to neurocognitive dysfunction, even in the presence of antiretroviral therapy (ART). Unveiling the cellular and molecular mechanisms for infection of the brain, establishment of a reservoir, and damaging the brain that might lead to preventative or therapeutic methods is essential. In this effort, we employed a Simian Immunodeficiency Virus (SIV) infected rhesus macaque model, which can mimic HIV infection of people to understand the lymphocyte involvement in this process [[37]3–[38]5]. Using the SIV/rhesus macaque model, the progression of viral infection after mucosal (vaginal and rectal) exposure has been studied in detail [[39]6, [40]7]. CD4 + T cells are recruited to the initial site of infection, which spreads locally within the tissue. Dissemination of infection then occurs via the lymphatic system [[41]8]. Similar effects, without the initial localization of infection, occur after intravenous viral inoculation [[42]9]. Pronounced infection can then be found in the gut lymphoid tissue, which then suffers a rapid depletion of CD4 + T cells, found in both this model and in PWH [[43]9, [44]10]. In the absence of treatment, systemic depletion of CD4 + T cells occurs leading to immune deficiency and the development of AIDS. Once this occurs, despite the administration of ART, full restoration of CD4 + T-cells is not achieved in PWH [[45]11]. Multiple mechanisms contribute to the depletion of CD4 + T-cells [[46]12], including killing infected cells by HIV/SIV-specific CD8 + T-cells [[47]13], cytolysis triggered by viral release [[48]9], and programmed cell death [[49]14]. The cytotoxic virus proteins and the inflammatory environment could also trigger uninfected cells’ death [[50]15, [51]16]. Infection of the brain is less clear-cut. Brain infection is thought to be initiated by the “Trojan horse” mechanism, by which infected immune cells from the blood seed the CNS [[52]17, [53]18]. The primary targets for HIV- and SIV-infection in the CNS are characterized as cells of the myeloid lineage. In acute brain infection in SIV-inoculated monkeys, the infected brain cells are predominately perivascular macrophages [[54]19]. Both macrophages and microglia can be found to harbor HIV and SIV as well as, if encephalitis (SIVE or HIVE) develops, myeloid-derived multinucleated giant cells [[55]19–[56]24]. Given the recognition that myeloid cells are the predominant infected cells in the brain in HIV and SIV infection, studies on HIV infection of the brain have focused on these cells. However, in the blood, the dominant SIV/HIV infected cell type is the CD4 + T cell. Thus, the true nature of the infected cell type responsible for brain invasion is not known. Studies have found that blocking a monocyte and lymphocyte adhesion molecule could reduce the invasion of HIV to the gut and central nervous system (CNS) [[57]25]. It has been found that lymphocyte-mediated CNS inflammation was concomitant with HIV/SIV infection, thus serving as an early source of neuroinflammation [[58]26]. The two broadly classified T-cell lymphocyte phenotypes, CD4 + and CD8 + T-cells, are present in the uninfected and HIV/SIV-infected brains at varied amounts. CD4 + T cells, as the main target for infection in the blood and the lymphoid system, may contribute to neuroinvasion of HIV/SIV, whereas CD8 + T cells, known to mediate antiviral functions, are more likely to promote neuroinflammation and neuropathogenesis [[59]27, [60]28]. However, this broad classification might oversimplify the brain lymphocyte phenotypes and their roles in HIV/SIV infection. Recent studies found that a unique CD4^dimCD8^bright lymphocyte phenotype could mediate neuroinvasion and anti-HIV responses [[61]29, [62]30]. Those findings highlight the potential contributions of the various brain lymphocyte phenotypes in the HIV/SIV-infected brains. Therefore, this study used high-throughput single-cell RNA sequencing (scRNA-seq) technique to identify and characterize varied lymphocyte phenotypes early after viral inoculation in SIV-infected as well as uninfected rhesus macaque brains. In the characterization of brain lymphocytes, we found an interesting cluster with co-expression of myeloid and lymphoid marker genes on scRNA-seq. Given that the phagocytosis/efferocytosis/cell fusion between macrophages and infected T cells might serve as one of the SIV/HIV cellular transmission mechanisms, we further validated and characterized this observation in a macrophage-T cell (Mac-T) cocultivation system. In addition, we also performed scRNA-seq of CD4 + lymphocytes in the blood at the same point to examine the relationship of the dominant viral targets in blood to what is found in the brain. We found that the CD4 + cytotoxic-like T cells identified in both blood and brain might be the primary T-cell phenotype seeding the SIV to the brain, which was further confirmed by integrating the dataset from another brain lymphocyte study with a longer time of acute SIV infection (21 days) [[63]31]. Materials and methods Experimental design To investigate the phenotypes of primary T cells that may facilitate SIV entry into the brain, as well as the responses of brain-resident lymphocytes during acute infection, we utilized a nonhuman primate model. Three independent biological replicates of adult rhesus macaques (Macaca mulatta) were intravenously inoculated with SIVmac251 and sacrificed at 12 days p.i. to capture early viral dynamics. Three uninfected macaques served as negative controls. Animals were randomly assigned to the infected or control groups. Immune cells were isolated from the six brain tissue samples (the three SIV-infected and three uninfected animals) and CD4 + T cells were isolated at the same time point from peripheral blood samples (from the three infected and one of the uninfected macaques). Single-cell RNA sequencing (scRNA-seq) was performed using the 10x Genomics Chromium platform for both compartments. In order to examine a later timepoint p.i., a published dataset [[64]31] comprising four SIV-infected rhesus macaques and two uninfected animals was downloaded and used for comparative analyses. To assess the biological validity of a distinct Myeloid-T cell population observed in scRNA-seq data, we conducted an in vitro co-culture experiment using human Jurkat T cells and human monocyte-derived macrophages (MDMs). Duplicate samples were prepared for each of the following three conditions: [[65]1] co-culture with Jurkat cells pretreated with Camptothecin (to induce apoptosis) [[66]2], co-culture with Jurkat cells pretreated with anti-CD95 antibody (Fas-mediated apoptosis), and [[67]3] co-culture with untreated Jurkat cells. As a control for potential doublet artifacts during single-cell encapsulation, an additional fourth condition involved the physical mixing of Jurkat cells and MDMs immediately prior to loading on the 10x Genomics platform without prior co-culture. All eight samples from these four experimental conditions were processed for scRNA-seq using the 10x Genomics Chromium platform. Animals The six male adult rhesus macaques used in this study were purchased from PrimGen (Hines, IL) and New Iberia (LA) and tested negative for the indicated viral pathogens: SIV, SRV, STLV1, Herpes B-virus, and measles; and bacterial pathogens: salmonella, shigella, campylobacter, yersinia, and vibrio. Macaques were housed in compliance with the Animal Welfare Act and the Guide for the Care and Use of Laboratory Animals in the NHP facilities of the Department of Comparative Medicine, University of Nebraska Medical Center (UNMC). The primate facility at UNMC has been accredited by the American Association for Accreditation of Laboratory Animal Care International. The UNMC Institutional Animal Care and Use Committee (IACUC) reviewed and approved of this study under protocols 19-145-12-FC and 16-073-07-FC. Animals were maintained in a temperature-controlled (23 ± 2° C) indoor climate with a 12-h light/dark cycle. They were fed Teklad Global 25% protein primate diet (Envigo, Madison, WI) supplemented with fresh fruit, vegetables, and water and libitum. The animal care and veterinary personnel observed the monkeys twice daily to check their health status. Three of the six animals were intravenously inoculated with a stock of SIVmac251 to establish acute SIV infection (93T, 94 T, and 95 T). The other three macaques were uninfected (92T, 104 T, and 111 T) and used as control. Virus stocks were provided by the Virus Characterization, Isolation, and Production Core at Tulane National Primate Research Center. Our previously published paper has reported the viral loads of plasma and different brain regions in the three infected animals (i.e. 93 T, 94 T, and 95 T) that were also used in the study [[68]32]. Isolation of immune cells from the brain Twelve days after viral inoculation, a necropsy was performed on deeply anesthetized (ketamine plus xylazine) animals, following intracardial perfusion with sterile PBS containing 1 U/ml heparin. Brains were harvested, and approximately half of the brain was taken for cell isolation. Immune cell-enriched cellular isolation from the brain was performed using our previously described procedure [[69]33]. Briefly, the brain was minced and homogenized in cold Hank’s Balanced Salt Solution (HBSS, Invitrogen, Carlsbad, CA, USA). After being centrifuged, the brain tissue was digested at 37° C in HBSS containing 28 U/ml DNase I and 8 U/ml papain for 30 min. After digestion, the enzymes were inactivated by adding 2.5% FBS, and the cells were centrifuged and resuspended in cold HBSS. The cell suspension was brought to a final concentration of 20% Percoll by the addition of 90% Percoll (GE HealthCare, Pittsburg, PA, USA) and centrifuged at 4° C for 15 min at 550 x g. The cell pallet at the bottom was resuspended in HBSS and passed through a 40 μm strainer to remove cell clumps and/or aggregates. Cells were again pelleted by centrifugation. A final wash was performed before the resulting cells were quantified on a hemocytometer and Coulter Counter Z1. The brain cells were cryopreserved before the cell sorting and scRNA-seq. Specifically, after isolation as above, the cells were centrifuged at 4° C at 550 x g for 5 min, and the supernatant was removed. The pellet was dissociated by tapping and then resuspended by the dropwise addition of a solution of 4° C 10% DMSO in FBS at a concentration of 10^6 cells per milliliter. Cells were transferred to cryopreservation tubes and controlled slow freezing at −80° C. After 24 h, cryotubes were transferred to liquid nitrogen for long-term storage. scRNA-seq sample preparations for studying brain immune cells in acute SIV infection In our prior analysis of brain immune cells, we isolated cells using positive selection for CD11b and/or CD45 selection [[70]32]. In order to capture more lymphocytes for analysis, we repeated the isolation of cells for scRNA-seq analysis using just CD45 selection. Samples of cryopreserved brain cell isolates, stored in Liquid nitrogen described above, were rapidly thawed in a 37° C water bath. The cell recovery procedures were well described in our previous publications [[71]33]. After the recovery, cells were washed and counted by Coulter Counter Z1. Once cell concentration was known, cells were transferred to ice-cold PBS, staining with Live/Dead and PE-labeled anti-CD45 antibody (Biolegend, San Diego, CA). The live cells that were CD45 + were collected for scRNA-seq capture and library preparation. Isolation of CD4 + T cells from the blood Freshly isolated EDTA-anticoagulated blood was centrifuged, and the cells washed with PBS and then stained with UV-blue live/dead for 30 min at 4° C. After being blocked with FACS buffer for 3 min, the cells were centrifuged and resuspended in the antibody cocktail mixed in Brilliant Stain buffer (Catalog # 563794, BD Biosciences, San Jose, CA, USA) for 45 min. After incubation, red blood cell (RBC) lysing buffer was added to the sample and the cells were pelleted at 300 x g for 8 min. After centrifugation, the cells were resuspended in RBC lysing buffer, which was repeated to completely remove RBCs. The cells were then pelleted and resuspended in MACS buffer for FACS sorting. The antibody cocktail included V500-labeled anti-CD3 antibody (Catalog # 561417, BD Biosciences, San Jose, CA, USA), BV786-labeled anti-CD4 antibody (Catalog # 563881, BD Biosciences, San Jose, CA, USA), PE-CF594-labeled anti-CD20 antibody (Catalog # 562295, BD Biosciences, San Jose, CA, USA) The selection of cells was based on the size, singlet, live, and expression of CD20, CD3, and CD4. The CD20-positive cells were excluded, and the CD20-negative cells that were positive for CD3 and CD4 were collected for scRNA-seq library preparations. Jurkat cell culture and apoptosis A human T-cell cell line (BCL2 Jurkat, ATCC CRL-2899) was obtained from ATCC and cultured in RPMI 1640 medium (Catalog # SH30027.02, Hyclone, Logan, UT, USA) supplemented with 10% fetal bovine serum (Catalog # F8067, Sigma-Aldrich, St. Louis, MO, USA) and 1% antibiotic-antimycotic (Catalog # SV30079.01, Hyclone, Logan, UT, USA). The Jurkat cells were seeded in 24-well plates at a density of 300,000 cells per well one day before the experiment. On the day of the experiment, apoptosis was induced in designated groups using either 1 µg CD95 (Catalog # 16-0958-81, Invitrogen, Carlsbad, CA, USA) or 10 µM Camptothecin (Catalog # J62523.MD, Thermo Scientific, Ward Hill, MA) for 3–6 h. Following incubation, apoptosis was assessed using the Alexa Fluor 488 Annexin V/Dead Cell Kit (Catalog # V13241, Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instructions. Prepared samples were then analyzed using an LSR II flow cytometer (BD Biosciences, San Jose, CA, USA). Monocyte differentiation Monocytes were obtained from the University of Nebraska Medical Center Elutriation Core Facility under IRB approval and isolated from the peripheral blood mononuclear cells (PBMCs) of healthy donors. They were cultured in 6-well plates (3 million cells per well) or 12-well plates (1 million cells per well) using MDM culture medium for one week before the experiment. The culture medium was half changed once a week. The MDM culture medium consisted of 50 mL macrophage serum-free medium (Catalog # 12065-074, Gibco, Grand Island, NY, USA), 500 µL HEPES (Catalog # 15630-080, Gibco, Grand Island, NY, USA), 50 µL Gentamicin (Catalog # G1397, Sigma-Aldrich, St. Louis, MO, USA), 5 µL Ciprofloxacin (20 mg/mL dissolved in 1% HCL, Catalog # 16626133, Honeywell Fluka, Charlotte, NC, USA), 500 µL Nutridoma (Catalog # 21952600, Roche, Indianapolis, IN), and 5 µL M-CSF (Catalog # 300-25-250UG, Gibco, Grand Island, NY, USA), which was filtered using Millipore Steriflip filter (Catalog # SCGP00525, Sigma-Aldrich, St. Louis, MO, USA) before use. Flow cytometry and scRNA-seq for MDMs-Jurkats cocultivation Jurkat cells were plated one day prior to the experiment in 6-well plates. For flow cytometry, Jurkat cells and MDMs were stained with 1µM CellTracker Deep Red (Catalog # [72]C34565, Invitrogen, Carlsbad, CA, USA) and 1µM CellTracker Green (Catalog # C7025, Invitrogen, Carlsbad, CA, USA), respectively, for 45 min. After staining, the MDMs were washed with prewarmed PBS three times. Apoptosis was induced in designated Jurkat cell groups using either 1 µg anti-CD95 or 10 µM camptothecin for 3 h. After centrifugation and removal of the supernatant containing CD95 or camptothecin, apoptotic or non-apoptotic Jurkat cells were cocultured with MDMs at a 10:1 ratio in MDM culture medium for 1 h. After cocultivation, the supernatant was removed, and the cells were washed three times with pre-warmed PBS to remove floating Jurkats. Cells were dissociated from the plate by incubating with TrypLE (Catalog # 12605-010, Gibco, Grand Island, NY, USA) at 37 °C for 10 min, followed by scraping. Cells designated for scRNA-seq were resuspended in PBS (Catalog # 14190-136, Gibco, Grand Island, NY, USA) containing 0.04% BSA (Catalog # 130-091-376, Miltenyi Biotec, Bergisch Gladbach, Germany), while those for flow cytometry were resuspended in FACS buffer (Catalog # 00-4222-26, Invitrogen, Carlsbad, CA, USA). Flow cytometry was performed using an LSR II flow cytometer (BD Biosciences, San Jose, CA, USA). Staining of MDMs and Jurkats MDMs were cultured on 12-well plates with glass coverslips (Catalog # 12541006, Fisher Scientific, Pittsburg, PA) for a week and Jurkat cells were plated one day prior to the experiment also in 12-well plates. The staining procedures, apoptosis induction, and cocultivation were the same as flow cytometry. After cocultivation, the supernatant was removed, and the cells were washed three times with pre-warmed PBS to remove floating Jurkats. Cells were fixed with 4% paraformaldehyde (PFA) at room temperature for 15 min, stained with DAPI (Catalog # D1306, Invitrogen, CarIsbad, CA, USA) for 5 min, and mounted by ProLong Gold antifade reagent (Catalog # [73]P36930, Invitrogen, Eugene, OR, USA). Staining of MDMs’ phagolysosome and Jurkats MDMs were cultured on 12-well plates with glass coverslips for a week and Jurkat cells were plated one day prior to the experiment also in 12-well plates. On the day of experiment, the Jurkat cells were centrifuged and resuspended in 1µM CellTracker Deep Red for 45 min. Then, the Jurkat cells in designated groups were induced apoptosis and cultivated with MDMs as mentioned above. After cocultivation, the supernatant was removed, and the cells were washed three times with pre-warmed PBS to remove floating Jurkats. Cells were fixed with 4% paraformaldehyde (PFA) at room temperature for 15 min, incubated with 1% Triton at room temperature for 15 min and blocked with 1% BSA for 1 h. Then, MDMs were stained with CoraLite Plus 488-conjugated CD68 antibody (Catalog # CL488-25747, Proteintech, Rosemont, IL, USA) overnight at 4 °C. The CD68 antibody was diluted in 1% BSA at concentration of 1 µg/mL for staining. After staining, cells were rinsed with 1% BSA for three times and stained with DAPI for 5 min before being mounted by ProLong Gold antifade reagent. Microscopy The colocalization of MDMs and Jurkat T cells was captured at 25x water-immersed Olympus FVMPE-RS multiphoton laser scanning microscope (Evident Corporation, Waltham, MA, USA). Z-stacked images were obtained at 1024 × 1024-pixel size with excitation/emission wavelength of 647/665 nm (CellTracker Deep Red), 498/517 nm (CellTracker Green), 499/520 nm (Coralite488), and 359/461 nm (DAPI). The number of steps for z-acquisition was determined for each cell independently but each step was set to 0.5 μm. The z-stacked images were further processed with Fiji ImageJ for orthogonal view with scale bar. scRNA-seq library preparation Post-sorting, isolates were concentrated to approximately 1000 cells per µL and assessed by trypan blue for viability and concentration. Brain immune cells and blood CD4 + T cells are isolated from the acutely infected macaques that were processed using the 10× Genomics (Pleasanton, CA, USA) Chromium 3’ v3 GEM kit targeting 8000 cells. And cocultivation study used 10× Genomics GEM-X Universal 3’ Gene Expression v4 4-plex kit targeting 5000 cells. Based on the targeting cell number, the ideal volume of cells was loaded onto the 10× Genomics Chromium GEM Chip and placed into the Chromium Controller for cell capturing and Library preparation. In short, this occurs through microfluidics and combining with Single Cell 3’ Gel Beads containing unique barcoded primers with a unique molecular identifier (UMI), followed by lysis of cells and barcoded reverse transcription of RNA, amplification of barcoded cDNA, fragmentation of cDNA to 200 bp, 5’ adapter attachment, and sample indexing as the manufacturer instructed with version 3 reagent kits. The prepared libraries were sequenced using Illumina (San Diego, CA, USA) Nextseq550 and Novaseq6000 sequencers. The sequences have been deposited in NCBI GEO (accession number: GSE293543). Preprocessing with 10× genomics pipelines The sequenced brain and blood samples were processed using the 10× Genomics Cell Ranger pipelines (7.1.0). Specifically, the scRNA data were demultiplexed and aligned to the customized Mmul10 rhesus macaque reference genome (NCBI RefSeq assembly) and a chromosome representing the SIV genome, as described [[74]31]. The scRNA-seq data from the MDM-T cell cocultivation experiments were processed using the 10× Genomics Cell Ranger pipelines (9.0.1) with GRCh38 human reference genome for alignment. Given the on-chip multiplexing was applied in Library preparation, we used cellranger multi for processing the fastq files and generating count matrix. The counting summary statistics generated by 10x Genomics for aforementioned scRNA-seq data are shown in Suppl. Table 1. The downstream analyses were implemented with R (version 4.3) and Python (version 3.12.7). The code used for analyzing the data in this study was deposited in this GitHub repository: [75]https://github.com/Howard-Fox-Lab/Brain_lymphocytes_acute_SIV_infec tion. Identification and characterization of brain immune cells and blood CD4 + T cells We first combined our previous published data [[76]32] from brains ([77]GSE253835) with those newly sequenced brain data using Seurat R package (version 4.4.0) [[78]34] to generate a single Seurat object. The four samples from the blood were merged to generate a separate Seurat object. Then, quality controls were performed for UMI counts, gene counts, and mitochondria percentage. We kept the cells from the blood and brain with UMI counts from 400 to 20,000, gene counts from 400 to 10,000, and mitochondria read percentage less than 15%. This filtering step removed ~ 9.7% of cells from the brain and ~ 3.1% of cells from the blood. After removing the low-quality cells, we normalized the data. The normalization method used for the brain and blood dataset was normalized by natural logarithm of (x + 1), implemented by the NormalizeData function in Seurat. After normalization, we performed several functions in Seurat to reduce dimensionality, including FindaVariableFeatures, ScaleData, and RunPCA. Curated rhesus macaque genes (Suppl. Table 2) filtered the features used for PCA. Then, we used the Seurat implementation of Harmony [[79]35] to remove the batch effect between samples processed in different batches. This was followed by running the FindNeighbor, FindClusters, and RunUMAP functions, which used the first 20 PCs (determined by the ElbowPlot) and 0.2 as resolution. The other parameters that were not mentioned were set as default. To characterize each cell cluster, we detected differentially expressed genes (DEGs) by using the FindAllMarkers function (parameters: only. pos = TRUE, min. pct = 0.25, logfc. threshold = 0.25, test. use = “Wilcox”). The clustering based on all sorted brain cells generated 12 clusters, most of which were myeloid cells. Among those 12 clusters, one cluster had a myeloid and lymphoid cell marker coexpression, indicating those cells might be doublet. Therefore, we implemented the DoubletFinder R package(version 2.0.4) [[80]36] to infer the doublet on a sample basis. We first used no ground-truth method to find the optimal pK value through the paramSweep, summarizeSweep, and find.pK functions. For the paramSweep function, we used the first 20 PCs and set ‘sct’ as false. The optimal pK with the largest value of mean-variance normalized bimodality coefficient (BCmvn) was selected. Then, we estimated the homotypic doublet proportion (nExp value), which depends on cell loading densities into the 10X Genomics device and the cell type annotations. We used our initial clustering information as cell type annotations. Given our 10X single-cell experiments targeting 8000 cells and the recovery cell number was ~ 5000 for each sample, the estimated percentage of doublet was set to 0.039 (3.9%) for this estimation. Finally, doublet inferring was implemented using the doubletFinder function (parameters: PCs = 1:20, pN = 0.25, pK = optimal pK found in each sample, nExp = nExp value estimated for each sample, reuse.pANN = False, sct = False). The clustering for sorted blood cells generated 11 clusters, most of which were lymphoid cells. Given the scope of this study, we kept the cell clusters with the expression of lymphocyte markers (e.g. CD3D, CD3G, CD3E, IL7R, TCF7) and removed other cell clusters from the downstream analyses for both brain and blood datasets. In total, we identified 26,340 brain lymphocytes and 25,808 blood lymphocytes. To cluster lymphocytes into distinct cell subsets, we followed the general procedures outlined by Smillie et al., [[81]37] which included subsetting lymphocyte clusters and performing the FindaVariableFeatures, ScaleData, RunPCA, RunHarmony, FindNeighbors, FindClusters, and RunUMAP functions for the subset lymphocyte clusters in brain and blood. The parameters set for rerunning those functions were the same as above—the subsetting and reclustering generated eight lymphocyte clusters in the brain and blood, respectively. We then performed DEG analyses for newly identified lymphocyte clusters in blood and brain by using the Wilcoxon rank sum test. The statistical thresholds were set in the same way as mentioned. We also screened the expression of markers for different lymphocyte phenotypes by using FeaturePlot and VlnPlot functions in the Seurat package. To characterize the lymphocyte by CD4 and CD8 expression, we calculated the CD4 to CD8B (more specific marker for CD8 + cells) ratio. This ratio was calculated by using the formula shown below: graphic file with name d33e426.gif The CD4 to CD8B ratio was calculated for each of the cells, embedded in the metadata of the Seurat objects, and plotted through the FeaturePlot function for visualization. Annotation validation with automatic annotation tool In addition to the manual annotation for each brain and blood lymphoid cell cluster by known markers, we also used one of the scRNA-seq automatic annotation tools, Single-cell Mayo Map (scMayoMap) [[82]38]. The DEGs of each brain or blood cluster detected by the FindAllMarkers function in the Seurat were used as input for the scMayoMap function. The tissue parameter for this function was set as blood for both brain and blood samples given that the lymphocytes in the brain are originally from the peripheral blood. The annotation results were visualized by the scMayoMap.plot function. Study of SIV-positive cells in brain and blood The SIV-positive cells were found in infected animals’ blood and brain immune cells. To understand the primary infected lymphocytes seeding SIV into the CNS, we subsetted all the brain and blood lymphocytes found with SIV mRNA transcripts for clustering. The workflow of using the subset dataset for reclustering was the same as mentioned above, and SIV gene was excluded from clustering. Furthermore, we used the first 10 PCs and 0.1 as the resolution in FindNeighbors, FindClusters, and RunUMAP functions. After clustering, we also performed trajectory analysis on SIV + cells using monocle 3 R package (version: 1.0.0) [[83]39–[84]41]. The Cluster 1 identified by clustering on SIV + cells was set as root node to order other clusters in their pseudotime. Then, we identified the genes that are differentially expressed on the different paths through the trajectory by using graph_test function in monocle3. The neighbor_graph was set to “principal_path”. The expression of top 100 genes, ranked by Moran’s I statistics, was plotted across pseudotime. Cell communication analysis between brain myeloid and lymphoid cells The cell-cell communication network’s inference, analyses, and visualization were implemented in the CellChat v2 R package (version 2.1.0) [[85]42, [86]43]. The scRNA-seq data of brain myeloid cells from same animals were obtained from a previous publication ([87]GSE253835) [[88]32]. Based on the characterizations performed in this publication, we defined Micro_0 and Micro_1 clusters as homeostatic microglia, Micro_3, Micro_5, Micro_7, and Micro_11 as inflammatory microglia, Macro_6 as CD16 + CAM, and Macro_8 as CD14 + CAM. The cell communication analysis was performed separately for infected and uninfected animals, following the vignette provided by developer. The human signaling database included in CellChat was used for analysis. Study of brain and blood lymphocyte connections We merged the brain and blood lymphocytes from both infected and control animals to understand the connection between the brain and blood dataset. We reperformed the FindaVariableFeatures, ScaleData, RunPCA, RunHarmony, FindNeighbors, FindClusters, and RunUMAP functions on the merged dataset. We found two phenotypes (CD4 + CTLs and TRM) in the brain showing a strong connection with another two phenotypes (CD4 + CTLs and CD4 + EM) in the blood by visualizing the new and old clustering information in Sankey diagram using networkD3 package (version: 3.0.4). To understand the differences between brain and blood for those cells, we performed DEG analysis for CD4 + CTLs and CD4 + EM (or CD4 + TRMs) between blood and brain using the FindMarkers function in Seurat. Then, the genes with log[2]FC > 0.1 were used for GSEA, implemented by clusterProfiler package (version: 4.10.1) [[89]44]. The gseGO function was used for enriching genes to gene ontology-biological process (GO-BP) database, and the other parameters were set as: OrgDb = org.Mmu.eg.db, minGSSize = 100, maxGSSize = 500, pvalueCutoff = 0.05, eps = 0. The ggplot2 (version: 2.3.5) was used for the visualization of GSEA results. We then performed trajectory and pseudotime analyses on integrated datasets through the monocle3 R package [[90]39–[91]41]. The trajectory graph was inferred by the learn_graph function and fitted to the UMAP generated by Seurat. The visualization of UMAP embedding with differentiation path was through the plot_cells function. The label_cell_group, label_branch_points, and label_roots were disabled for plotting. Then, the CD4 + Naïve T cell cluster was set as the root node for ordering other lymphocytes in their pseudotime, which was visualized in the boxplot by ggplot2. Then, we identified the genes that are differentially expressed on the different paths through the trajectory by using graph_test function in monocle3. The neighbor_graph was set to “principal_path”. The expression of top 100 genes, ranked by Moran’s I statistics, was plotted across pseudotime. Comparison of 12-days and 21-days acute SIV infections Four infected ([92]GSM6900144, [93]GSM6900146, [94]GSM6900148, [95]GSM6900150) and two uninfected samples ([96]GSM6900141 and [97]GSM6900143) from a publicly available dataset ([98]GSE221815) [[99]31] were aligned to our customized reference genome by 10x Genomics pipeline. Then, they were merged with our three infected and uninfected brain samples, and the aforementioned workflow in Seurat was followed to generate clusters. The markers between clusters were found through the FindAllMakers function in Seurat, which was used for characterization. To understand the upregulated and downregulated genes for brain lymphocytes in 21 days of SIV infection compared to 12 days, we generated pseudobulk for individual samples using the AggregateExpression function in the Seurat. Since Myeloid-T cells could barely be found in the 21-day dataset, they were excluded from this DEG analysis. Then, we performed DEG analysis on the pseudobulk data using DESeq2 as a statistical test method. To study the SIV + cells, cells with SIV transcript detected from 12 days of infected brain and blood samples and 21 days of infected brain samples were identified and merged for clustering and analysis. The SIV + cells from the Myeloid-T cell cluster were excluded before clustering and other analyses. Analysis of scRNA-seq data from MDM-Jurkat T cell cocultivation system All eight samples from four different groups were integrated into a single Seurat object. Low-quality cells were filtered out based on the following criteria: UMI count < 1000, gene count < 500, and mitochondrial content > 15%. After filtering, 42,390 cells remained for downstream analysis. We followed a standard scRNA-seq analysis workflow in Seurat, including normalization using the LogNormalize method, identification of highly variable genes, data scaling, and PCA using the top 2,000 most variable genes. The UMAP was performed using the first 30 PCs, and clustering was conducted with a resolution of 0.2. To correct for batch effects, we applied Harmony before running UMAP and clustering. Marker genes for each cluster were identified using the FindAllMarkers function with the same parameter settings. For further analysis, the processed Seurat object was converted to an h5ad file for visualization in the Scanpy Python package (version 1.10.3) [[100]45]. We examined the expression of macrophage and T cell markers across clusters and identified a cluster exhibiting co-expression of both, designated as the Mac_T cluster. To ensure that this identification was not biased by apoptosis-related groups, we repeated the analysis on a subset of samples excluding apoptosis groups, where we again identified the same co-expression cluster. DEG analysis was conducted to compare the Mac_T cluster with macrophage and lymphocyte clusters. DEGs with log[2]FC > 0.25 and an adjusted p-value < 0.05 were selected for pathway analysis using the compareCluster function from the clusterProfiler package. GO-BP annotations were used for enrichment analysis. Differential state analysis using pseudobulk method Differential state analysis in this study was performed using the muscat R package (version 1.16) [[101]46]. Starting from a Seurat object, we converted the data into a SingleCellExperiment format using the as.SingleCellExperiment function in Seurat. Single-cell data were then aggregated by clusters and samples using the aggregateData function in muscat for pseudobulk analysis. Differential expression analysis was conducted using a linear model with contrast information for two conditions. For the comparison between acute-infected and uninfected conditions, the uninfected group was used as the reference. Similarly, for the comparison between non-apoptosis and no-cocultivation groups, the no-cocultivation group served as the reference. Pseudobulk differential expression analysis was performed using edgeR [[102]47]. Genes with an adjusted p-value < 0.05 and an absolute log[2]FC > 0.25 were considered significantly differentially expressed. For the acute-infected vs. uninfected comparison, DEGs were further filtered to include only those expressed in at least 10% of cells in one or more groups. Visualization was carried out using built-in functions in muscat, and a multi-dimensional scaling (MDS) plot was generated using the pbMDS function. Statistics DEG analysis at the single-cell level was performed using the non-parametric Wilcoxon rank-sum test, with p-values adjusted using Bonferroni correction. In most DEG analyses, significance was defined as an adjusted p-value < 0.05 and log[2]FC > 0.25. However, in certain comparisons, a higher log[2]FC threshold was applied, as specified in the Figure legends. For sample-level DEG or differential state (DS) analysis, edgeR or DESeq2 was used with specification in method and Figure legend. Similar to the single-cell analysis, genes with an adjusted p-value < 0.05 were considered significant. Pathway enrichment analysis was performed using a hypergeometric test, with p-values adjusted for multiple comparisons across clusters or conditions. An adjusted p-value < 0.05 was considered statistically significant. Results Characterization of a cell cluster having co-expression of myeloid and lymphoid genes in scRNA-seq To examine the lymphocyte population in the brain during the initial phase of SIV infection, we isolated enriched immune cells from three acutely infected (day 12 p.i.) rhesus macaques and three uninfected animals, and performed scRNA-seq. Following bioinformatic processing, we applied the initial clustering algorithm on 181,543 brain cells, resulting in 12 clusters. Then, we characterized each of those cell clusters by screening different leukocyte markers. Most cells were identified as microglia (Suppl. Figure 1 A). Additionally, four lymphocyte clusters with high lymphocyte marker expression (Suppl. Figure 1B) were identified, containing 26,340 cells. Except for two uncharacterized cell clusters, the other cell clusters in our brain dataset could be identified as microglia, lymphocytes, or CNS-associated macrophages (CAMs) (Suppl. Figure 1 C). Interestingly, one of those cell clusters (i.e. cluster 8) had co-expression of myeloid and lymphoid cell markers, as well as a higher UMI count compared to microglia and lymphocyte clusters (Suppl. Figure 1D). This could be, for example, cells transitioning between states, or perhaps a myeloid cell fusing with or phagocytosing a lymphoid cell. Another possibility is that these are artifacts due to an initial capture of two cells (thus a doublet) instead of a single cell in the initial stage of the scRNA-seq procedure, either because the cells are highly adherent to each other or due to an artefactual capture of two independent cells. To help assess this we applied DoubletFinder [[103]36], a frequently used algorithm to identify doublets. This revealed that about half of the cells in cluster 8 could be identified as doublets (Suppl. Figure 1E). However, DoubletFinder is predicting doublets based on combining the transcriptional characteristics of pairs of cells within the dataset. Thus, it is difficult to distinguish the above possibilities. Therefore, we designed a study to validate this special observation on scRNA-seq experiment. The experimental workflow is illustrated in Fig. [104]1A. We initially hypothesized that macrophages preferentially engulf apoptotic cells via efferocytosis. In this effort, we induced apoptosis in a human T-cell line (Jurkat cells) using biological and chemical agents prior to cocultivation with monocyte-derived macrophages (MDMs). Specifically, we evaluated the efficacy of Camptothecin (Campto) and anti-CD95 in inducing apoptosis in Jurkat cells. We found that both agents significantly increased apoptosis compared to the control, with anti-CD95 demonstrating greater potency (Suppl. Figure 2 A and 2B). Extending the treatment from three to six Hours increased the proportion of apoptosis in the Jurkat cells from 16 to 32% with Campto and from 22 to 62% with anti-CD95, However, the rate of apoptosis for untreated cells remained at 4%. After confirming the effectiveness of these apoptotic agents, we cocultured the Jurkat cells with MDMs. To facilitate tracking via flow cytometry and microscopy, MDMs and Jurkats were labeled with CellTracker dyes prior to cocultivation. Given the size differences between macrophages and T cells, the two populations were readily distinguishable based on forward and side scatter properties in flow cytometry. Double-positive cells, indicative of phagocytosis, were detected exclusively within the macrophage population and not in T cells (Suppl. Figure 2 C). Interestingly, double-positive cells were observed in both apoptotic and non-apoptotic (control) groups (Fig. [105]1B), suggesting that non-apoptotic T cells could also be efficiently phagocytosed by MDMs. This observation was further corroborated by confocal microscopy (Fig. [106]1C), which showed colocalization of MDMs and cell debris from Jurkat cells in both apoptotic (Suppl. Video. 1a and 1b) and non-apoptotic conditions (Suppl. Video. 1c). The phagocytosis of jurkat cells by MDMs was further confirmed by colocalization of MDMs’ phagolysosome marker, CD68, and cell debris from Jurkat cells, which was also observed across all three groups (Suppl. Figure 2D). Fig. 1. [107]Fig. 1 [108]Open in a new tab Identification of macrophage-T-cell coexistence cells (Mac_T).(A) Workflow for generating monocyte-derived macrophages (MDM) and Jurkat cells cocultivation system. (B) Flow cytometry showing the Mac_T cells in three treatment groups (i.e. CD95, Camptothecin and No Apoptosis groups). (C) Representative figures showing macrophage-T-cell coexistence cells for three treatment groups. For both flow cytometry and confocal microscopy experiments, the MDMs were labeled with CellTracker Green and Jurkat cells were labeled with CellTracker Deep Red before the cocultivation. (D) Experimental design for scRNA-seq experiment, which involved four different groups (i.e. CD95, NoApop, Campto, and NoCoCul). (E) The UMAP plot of clusters and the normalized expression of macrophage, T cell and B cell markers. (F) The UMAP plot of 42,390 cells within nine different clusters. The annotation for each cluster was labeled. (G) The proportion of each cluster (upper panel) or phenotype (lower panel) in each experimental group. Campto: camptothecin-treated MDM-Jurkat cocultivation group. CD95: CD95-treated MDM-Jurkat cocultivation group. NoApop: non-apoptosis MDM-Jurkat cocultivation group. NoCoCul: non-cocultivation MDM-Jurkat cocultivation group Then we used scRNA-seq to confirm MDM phagocytosis of the Jurkat T. Cells from duplicate samples across four experimental conditions (Fig. [109]1D), including CD95-treated apoptotic, Campto-treated apoptotic, non-apoptotic, and a no-cocultivation control, were aggregated and clustered into nine cell clusters (Fig. [110]1E). Characterization revealed that clusters 1, 3, and 4 expressed macrophage markers, while clusters 0, 2, and 7 expressed T-cell markers. Cluster 6 was enriched for mitochondrial gene expression, cluster 8 expressed B-cell markers (likely due to contamination during monocyte isolation), and cluster 5 co-expressed macrophage and T-cell markers (Fig. [111]1E and Suppl. Table 3). Within the macrophage subsets, cluster 1 was enriched for SPP1 and lysosomal genes, while cluster 3 exhibited high expression of HLA class II molecules. Among the T-cell subsets, cluster 2 showed elevated expression of proliferation markers (e.g. MKI67, TOP2A), whereas cluster 7 upregulated activation/differentiation markers associated with CD4 + T cells (e.g. CCL5, IL32, RORA) (Suppl. Figure 2E and Suppl. Table 3). Clusters 0 and 4 lacked distinct activation markers, suggesting they represented resting T cells and macrophages, respectively. Based on these characterizations, we annotated each cell cluster showed in Fig. [112]1F and quantified the relative proportions of each cluster across treatment conditions (Fig. [113]1G). Notably, the non-apoptotic group exhibited the highest proportion (~ 16.4%) of the macrophage-T-cell co-expressing cluster (Mac_T), followed by the Campto-treated apoptotic group (~ 10.6%). In contrast, the CD95-treated and no-cocultivation groups displayed the lowest percentage (~ 5%). These results suggest that the reduced presence of Mac_T cells in the CD95-treated group may be attributed to T-cell degradation. Indeed, this group exhibited the lowest overall T-cell abundance and the highest proportion of the MT cluster (Fig. [114]1G). Furthermore, the flow cytometry shown in Suppl. Figure 2 A also suggested that CD95 had a greater ability to induce apoptosis in T cells compared to Campto. Nevertheless, the Mac_T cluster was consistently more abundant in all cocultivation conditions compared to the no-cocultivation control, indicating that scRNA-seq successfully captured macrophage-phagocytosed or efferocytosed T cells. To rule out potential biases introduced by CD95 or Campto treatment in clustering, we reanalyzed the non-apoptotic and no-cocultivation samples (Suppl. Figure 2 F). Similarly, a distinct cluster of co-expressing macrophage and T-cell markers was identified (Suppl. Figure 2G), with its relative proportion being approximately threefold higher in the non-apoptotic cocultivation group (Suppl. Figure 2 H). Chemokine genes might serve as potential markers for identifying macrophage-T-cell co-expressing cells in scRNA-seq In our characterization, we identified a Mac_T cell cluster with high expression of chemokines such as CCL2 and CCL3, which were also strongly expressed in HLAII_Mac but not in SPP1_Mac or general Mac clusters (Suppl. Figure 2E). This suggests a distinct macrophage phenotype that preferentially engulfs Jurkat cells. To further investigate this, we computed the Euclidean distance between Mac_T and other macrophage clusters in a low-dimensional space. Notably, HLAII_Mac exhibited the shortest Euclidean distance to Mac_T compared to the other two macrophage clusters (Fig. [115]2A), indicating that HLAII-expressing macrophages may preferentially engulf Jurkat cells. Similarly, to identify specific T cell phenotypes preferentially engulfed by macrophages, we calculated the Euclidean distance between Mac_T and other T cell clusters. Mac_T was closest to the activated T cell cluster (Fig. [116]2B), suggesting that the T cell component within Mac_T originates might be primarily from activated T cells. Further gene ontology (GO) analysis revealed that macrophages and T cells within the Mac_T cluster exhibited upregulated pathways related to DNA replication, mRNA splicing, and MHC class II antigen processing and presentation (Fig. [117]2C and D, Suppl. Table 4), indicating a unique immune state within this cluster. Fig. 2. [118]Fig. 2 [119]Open in a new tab Characterizations of Mac_T cluster. (A-B) The distribution of 42,390 cells on the first two principal components (PCs). The cells were colored by the cluster information, and the Euclidean distance between Mac_T cluster and (A) other macrophage clusters or (B) lymphocyte clusters were labeled and highlighted. (C) Comparison of gene ontology pathways (Biological Process, BP) between SPP1_Mac, HLAII_Mac, and Mac_T clusters. (D) Comparison of gene ontology pathways (Biological Process, BP) between Naïve_T, Proli_T, Activated_T and Mac_T clusters. (E) Multidimensional scaling (MDS) plot displaying Mac_T cells in 8 samples from four different experimental groups. The expression of Mac_T cells were averaged in each sample for pseudobulk analysis. (F) Heatmap plotting of top 20 (ranked by adjusted p value) genes in Mac_T cluster that were significantly changed in NoApop group compared to NoCoCul group. The pseudobulk comparison was performed between two animals in NoApop and NoCoCul groups using EdgeR method. The genes with adjusted p-value < 0.05 and log[2]FC > 0.25 were defined as significantly changed genes. (G) The scatter plot of the cells in Mac_T cluster regarding their expression of CCL3 and CCL4. (H) The proportion of each treatment group for CCL3 and CCL4 double negative cells in Mac_T cluster. (I) The scatter plot of the cells in the cluster of the rhesus macaque brain with co-expression of myeloid and lymphoid genes (Cluster 8 in Figure S1C) regarding their expression of CCL3 and CCL4L1. Campto: camptothecin-treated MDM-Jurkat cocultivation group. CD95: CD95-treated MDM-Jurkat cocultivation group. NoApop: non-apoptosis MDM-Jurkat cocultivation group. NoCoCul: non-cocultivation MDM-Jurkat cocultivation group Both Mac_T cells and technical doublets (macrophage-T cell aggregates from scRNA-seq artifacts) displayed co-expression of myeloid and lymphoid genes, leading to their clustering together. However, Mac_T cells likely undergo biological processes such as phagocytosis and efferocytosis, which are absent in technical doublets. Supporting this, the gene expression profiles of Mac_T cells differed significantly between cocultivation and no-cocultivation groups in low-dimensional space (Fig. [120]2E). To identify key genes driving these differences within the Mac_T cluster, we performed a pseudobulk analysis comparing Mac_T cells from the non-apoptotic and no-cocultivation groups. Notably, chemokine genes such as CXCL3, CCL3, CCL4, CCL4L2, and CXCL2 were significantly upregulated not only in the non-apoptotic group but also in the CD95 and Campto groups compared to the no-cocultivation group (Fig. [121]2F and Suppl. Table 5). These findings suggest that chemokine genes may serve as markers to distinguish Mac_T cells from technical doublets. To further validate this, we examined the expression of CCL3 and CCL4, two chemokines highly expressed in both human MDM and rhesus macaque brain myeloid cells. Within the Mac_T cluster, approximately 85% of cells co-expressed CCL3 and CCL4, while only ~ 4% lacked expression of both genes (Fig. [122]2G). Notably, most of these CCL3 and CCL4 double-negative cells originated from the no-cocultivation group (Fig. [123]2H). Similarly, in co-expression cluster from rhesus macaque sample, only ~ 8% of cells were CCL3 and CCL4 double-negative (Fig. [124]2I). This indicates that the majority of cells in this cluster are likely biologically relevant Mac_T cells, rather than technical doublets. Acute SIV infection depletes CD4 + lymphocytes and promotes the expansion of proliferating cytotoxic T cells in the brain The 26,340 cells from lymphocyte clusters and co-expression cluster (cluster 8) were re-clustered into eight different cell clusters, and the co-expression cluster (Myeloid-T cell cluster) separated from the other seven clusters from UMAP (Fig. [125]3A and B). This Myeloid-T cell cluster was also found with higher UMI, and gene counts than other clusters (Fig. [126]3C). Fig. 3. [127]Fig. 3 [128]Open in a new tab Characterization of brain lymphocytes in uninfected and acute-infected rhesus macaques. (A) The UAMP projection of 26,349 brain lymphocytes from infected and uninfected brains. The annotation for each brain lymphocyte cluster was labeled. (B) UMAP projection of normalized RNA expression of CD3D and P2RY12 genes. (C) Violin plots for UMI counts and gene counts for each brain lymphocyte cluster. (D) UMAP projection of normalized RNA expression of CD4, CD8A, CD8B genes and CD4 to CD8B ratio. (E) Violin plots for representative markers detected between CD4 + CTLs and CD8 + effector cells in the brain. (F) Heatmap for top 10 markers arranged by log[2]FC in each brain lymphocyte cluster. The scaled data was used for this plotting. (G) Violin plots for representative markers detected between CD8 + effector cells and effector memory (EM) cells in the brain. (H) Ridge plots for known markers for lymphocytes. (I) The UAMP projection of brain lymphocytes was labeled by infected status. (J) The relative proportion of each identified cluster in uninfected and acute-infected brain lymphocytes. The percentage of primarily changed cell clusters in uninfected and acute-infected conditions were labeled. The statistical test for marker detection was the Wilcoxon rank sum test with parameters as adjusted p-value < 0.05 and log[2]FC > 0.5. LOC710951: TCR α chain. LOC703029: TCR β chain. LOC720538: TCR γ chain. LOC711031: TCR δ chain Based on their gene expression profiles, we characterized the other lymphocyte phenotypes in the brain. We found that the brain lymphocytes had lower expression of the CD4 gene compared to the CD8 genes. Single cell studies of rhesus monkey lymphocytes indeed find low expression of the CD4 mRNA. By calculating the log[2] normalized CD4 to CD8B expression ratio for each single cell (Fig. [129]3D), the cells in cluster 1 and part of cells in cluster 0 were found with high CD4 to CD8B ratio, indicating they likely represent CD4 + T cells in the brain. The cells in cluster 3 and another part of cells in cluster 0 had a negative CD4 to CD8B ratio, indicating likely correspond to CD8 + T cells. Finally, we classified the brain lymphoid cells as Tissue Resident Memory (TRM) cells, γδ T cells, Effector Memory (EM) cells, CD4 + cytotoxic T lymphocytes (CTLs), NK cells, CD8 + Effector cells, and proliferating cells (Fig. [130]3A). Regarding CD4 + CTLs, they have been found to be highly enriched in the subset of CD4 + effector memory T cells expressing CD45RA [[131]48]. An analysis assessing scRNA-seq data from FACS-purified T cells has characterized analogous cells as CD4 + effector memory T cells [[132]49], and others have noted transcriptional similarities between CD4 + CTL and CD4 + effector memory T cells [[133]48, [134]50]. We will use the CD4 + CTL nomenclature here. Most lymphocytes in the brain exhibited gene expression profiles indicating cytotoxic properties, which included cluster 1 (CD4 + CTL) and cluster 3 (CD8 + Effector). To further distinguish them, we performed DEG analysis between those two clusters. We found that CD8 + effector cells were more NK-like, expressing NK cell markers (e.g. KLRC3, GNLY, FCGR3) not seen in CD4 + CTL (Fig. [135]3E). Instead, CD4 + CTL maintained IL7R expression, which should be downregulated following T cell activation [[136]51]. Additionally, compared to CD8 + effector cells in the brain, the CD4 + CTLs also had higher expression RORA, the marker for CD4 + Th17 cells [[137]52], and the CD6 gene, an essential molecule for T cell-antigen presenting cell (APC) interaction (Fig. [138]3E) [[139]53, [140]54]. Cluster 0, which contained both CD4 + and CD8 + lymphocytes, was non-cytotoxic, and its detected markers included CD69 and CD49a (ITGA1), which are known TRM T cell markers (Fig. [141]3F and Suppl. Table 6) [[142]55–[143]57]. Distinguishing CD4 + and CD8 + lymphocytes in cluster 2 and cluster 4 was challenging since double-negative cells were enriched in those clusters. However, more lymphocytes had a negative log[2] CD4 to CD8B ratio (Fig. [144]3D). Given the uncertainty for the rest of the lymphocytes’ expression of CD4 and CD8, we annotated cluster 2 and cluster 4 by their predicted biological properties. Cluster 2 had a high expression of proliferating markers (i.e. STMN1, MKI67, TOP2A), so we annotated it as a proliferating cluster. Cluster 4 had an expression of the genes related to TCR stimulation and lymphocyte cytotoxicity (Fig. [145]3F and Suppl. Table 6). However, compared to CD8 + effector cells, cluster 4 had higher GZMK expression but lower GZMA and GZMB expression (Fig. [146]3G), consistent with the granzyme gene changes during CD8 + memory T cell differentiation [[147]58]. Therefore, the evidence from scRNA-seq data indicated that cluster 4 might be effector memory (EM) T cells. Unlike lymphocyte clusters mentioned above, all the cells in clusters 5 and 7 barely expressed CD4 and CD8B (Fig. [148]3D), suggesting they were double negative cells. Cluster 7 expressed cytotoxic genes and some NK cell markers and had the lowest expression of TCR αβ chains (e.g. LOC710951 and LOC703029) and CD3 δε chains compared to other T cell clusters, so it was annotated as NK cell cluster. Cluster 5 had expressions of TCR gamma and delta chains (i.e. LOC720538 and LOC711031) and other γδ T17 cell markers, including IL23R and CCR6 (Fig. [149]3F). In addition, the CD69 and CD103 (ITGAE) playing a crucial role in lymphocytes’ initial tissue retention and prolonged residence [[150]59] were also found to be highly expressed in γδ T cells, which is consistent with the fact that the meningeal γδ T cells are thought as tissue-resident cells in the brain [[151]60]. Interestingly, most cells in γδ T cluster (cluster 7) had RNA expression of TCR αβ chains as well as γδ chains (Fig. [152]3H). Most of our manual annotations based on cell markers were further confirmed by one of the cell type annotation tools, Single-cell Mayo Map (scMayoMap) [[153]38] (Suppl. Figure 1 F). In response to acute SIV infection, the TRM (especially CD4+), CD4 + CTL, and γδ T cells were depleted in the brain relative to uninfected animals. Proliferating cytotoxic T cells, EM and Myeloid-T cell clusters increased, while the CD8 + effector and NK cell clusters barely changed in proportion at this time point (Fig. [154]3I and J). Among the three increased lymphocyte populations, the proliferating cluster had the largest increase (~ 15-fold). Acute SIV infection causes the depletion of CD4 + memory T cells and an increase of CD4 + CTLs in the peripheral blood Given that prior to entering the brain the CNS lymphocytes were blood-borne, we next studied blood lymphocytes. In the periphery, the main target for HIV/SIV is CD4 + T cells, so we focused on this cell type. Using freshly isolated PBMCs from the three acutely infected animals and one uninfected animal, CD4 + T-cells were FACS-purified and processed for scRNA-seq. The clustering on those CD4 + T-cell-sorted blood lymphocytes from the four animals generated eleven cell clusters (Suppl. Figure 3 A and Suppl. Figure 3B). Of the eleven cell clusters, eight expressed lymphocyte markers (Suppl. Figure 3 C) and three expressed monocyte markers (Suppl. Figure 3D). We subsetted the eight lymphocyte clusters and reclustered the cells. The re-clustering of lymphocytes generated eight different clusters (Fig. [155]4A). Among them, seven clusters had high expression of the CD4 gene and a positive log[2] CD4 to CD8B ratio (Fig. [156]4B), and one cluster (cluster 5) exceptionally had a negative log[2] CD4 to CD8B ratio, suggesting that this cluster contain contaminating CD8 + T cells. Additionally, a higher proportion of mitochondria genes were found as markers in cluster 6, so this cluster was denoted as MT. Besides clusters 5 and 6, the other five lymphocyte clusters were general CD4 + T cell phenotypes in the peripheral blood. Fig. 4. [157]Fig. 4 [158]Open in a new tab Characterization of T cells in CD4-sorted blood of uninfected and acute-infected rhesus macaques. (A) The UAMP projection of 25,808 CD4-sorted blood T cells from infected and uninfected animals. The annotation for each blood lymphocyte cluster was labeled. (B) UMAP projection of normalized RNA expression of CD4, CD8A, CD8B genes and CD4 to CD8B ratio. (C) UMAP projection of normalized RNA expression of known markers for different lymphocyte phenotypes and mitochondrial percentage (mitoPercent). (D) Heatmap for top 10 markers arranged by log[2]FC in each blood lymphocyte cluster. The scaled data was used for this plotting. (E) Violin plots for representative markers detected between two CD4 + CTL clusters in blood. (F) The relative proportion of each identified cluster in uninfected and acute-infected blood lymphocytes (CD4-sorted). The percentage of primarily changed cell clusters in uninfected and acute-infected conditions were labeled. The statistical test for marker detection was the Wilcoxon rank sum test with parameters as adjusted p-value < 0.05 and log[2]FC > 0.5 Most CD4 + blood lymphocytes were clustered into cluster 0, cluster 1, and cluster 3, which were non-cytotoxic T cells (Fig. [159]4C and Suppl. Table 7). Cluster 0 was found with naïve T cell markers (e.g. LEF1 and SELL), cluster 1 was found with effector memory T cell markers (e.g. ITGB1 and ICOS), and cluster 3 was characterized by regulatory T cell (Treg) markers (e.g. IL2RA and FOXP3) (Fig. [160]4D). In contrast, cluster 2, cluster 4, and cluster 7 were CD4 + CTLs (Fig. [161]4C andSuppl. Table 7), and markers for cluster 4 also included proliferating genes (STMN1, MKI67, TOP2A) (Fig. [162]4D). By comparing cluster 2 and cluster 7, we found that the cytotoxic genes were expressed at higher levels in cluster 2. In addition, cluster 2 completely lost the expression of naïve T cell genes, which still maintained expression in cluster 7 (Fig. [163]4E). Therefore, we hypothesized that cluster 7 might be in a transitional phase to become CD4 + CTLs. Those annotations were also confirmed by the cell type annotation tool, scMayoMap (Suppl. Figure 3E). During acute infection, we found that the CD4 + effector memory cells, were relatively decreased (from 32.1 to 14.5%) (Fig. [164]4F). Additionally, the percentage CD4 + Treg was also relatively decreased in acute infected animals, but with a small degree of change (from 5.8 to 4.8%). The decrease of CD4 + effector memory cells and Treg was accompanied by a dramatic increase of CD4 + CTLs (from ~ 7% to ~ 18%). However, the CD4 + naïve T cells were barely changed. Those findings were consistent with the current report that the CD4 + naïve T cells are less susceptible to HIV/SIV infection [[165]9, [166]61, [167]62]. In summary, the changes of CD4 + lymphocytes in the blood are partially consistent with what we found in the brain. The CD4 + memory T cells (EM in the blood and TRM in the brain) were the primary CD4 + lymphocytes depleted during the acute SIV infection in both peripheral and CNS. Brain lymphocytes undergo extensive activation during acute SIV infection, with effector memory T cells showing the strongest responses We examined transcriptomic changes within individual lymphocyte clusters responding to acute SIV infection, with a specific focus on brain lymphocytes. We did comparisons using statistical methods for both single-cell and bulk RNA-seq data. For the comparison at single-cell level, we observed a significant upregulation of cytotoxic, MHC class II, and interferon-inducible genes in brain lymphocytes during acute infection (Fig. [168]5A). Notably, while brain-resident CD4⁺ CTLs exhibited high baseline expression of cytotoxic genes, they did not further upregulate these genes in response to infection, unlike their counterparts in the blood (Suppl. Figure 4 A and Suppl. Figure 4B). This suggests a potential ceiling effect, as their cytotoxic gene expression was already elevated in uninfected brains. However, brain CD4⁺ CTLs did upregulate MHC class II molecules and interferon-related genes, indicating a heightened activation state. In contrast, upregulation of cytotoxic, MHC class II, and interferon-inducible genes was only observed in blood CD4⁺ CTLs (Suppl. Figure 4 C) and not in other blood CD4⁺ T-cell clusters (Suppl. Figure 4D). Further analysis of T-cell activation genes in brain (Fig. [169]5B) and blood (Fig. [170]5C) clusters showed that non-cytotoxic T cells in the brain acquired cytotoxicity during acute infection, whereas non-cytotoxic CD4⁺ T cells in the blood did not. This suggests that brain lymphocytes develop enhanced antiviral capabilities compared to those in the blood during acute SIV infection. Fig. 5. [171]Fig. 5 [172]Open in a new tab The transcriptomic changes of different brain lymphocyte phenotypes in response to acute SIV infection. (A) The gene scatter plot compared average gene expression in identified brain T cell and myeloid-T cell clusters between acute-infected and uninfected conditions. The statistical test for marker detection between lymphocytes found in infected and uninfected animals was the Wilcoxon rank sum test with parameters as adjusted p-value < 0.05 and log[2]FC > 0.5. The representative marker genes were labeled red in the gene scatter plots. (B-C) Comparison of the expression of selected cytotoxic genes, MHC class II molecules, and interferon-inducible proteins in different (B) brain and (C) blood lymphocyte clusters between uninfected and acute-infected conditions. (D) Multidimensional scaling (MDS) plot displayed brain (left) and blood (right) samples from acute and uninfected conditions. The plots were colored by cluster information. The expression of each lymphocyte cluster was averaged in each sample for pseudobulk analysis. (E) Heatmap plotting of top 20 (ranked by fold change) genes in each brain lymphocyte cluster that were significantly upregulated in acute infection group compared to uninfected conditions. (F) Heatmap plotting of top 20 (ranked by fold change) genes in each brain lymphocyte cluster that were significantly downregulated in acute infection group compared to uninfected conditions. The pseudobulk comparison was performed between three acute-infected and uninfected animals using EdgeR method. The genes with adjusted p-value < 0.05 and log[2]FC > 0.25 and expressed in an average of 10% of cells in at least 1 group were defined as significantly changed genes. (G) The UMAP projection of cells from acute-infected and uninfected brains in EM cluster and normalized CD4 to CD8B ratio (upper panel). The proportion of CD4+, CD8 + and double positive or negative cells in EM cluster for each infected and uninfected animal (lower panel) Multi-dimensional scaling (MDS) analysis at the sample level revealed a clear separation between acute-infected and uninfected animals along the second dimension (MDS2), indicating that acute infection affected multiple lymphocyte phenotypes in both the brain and blood (Fig. [173]5D). Additionally, the Myeloid-T cluster was distinct from other brain lymphocyte clusters, while blood lymphocyte cluster with high mitochondria percentage (MT cluster), non-cytotoxic, and cytotoxic T-cell clusters also separated along MDS1, reflecting their transcriptomic differences. To minimize the risk of inflated statistical significance caused by treating individual cells as independent replicates, we conducted pseudobulk comparisons between acutely infected (n = 3) and uninfected (n = 3) animals for each brain T-cell cluster by using the muscat implementation of edgeR. This comparison identified 522 significantly altered genes across different brain lymphocyte clusters (Fig. [174]5E and F and Suppl. Table 8). The EM cluster exhibited the highest number of significantly changed genes, followed by proliferating T cells and NK cells (Suppl. Figure 4E). Among these genes, ISG15, an interferon-stimulated gene induced by type I interferons during viral infection, was significantly upregulated across multiple brain lymphocyte clusters, including tissue-resident memory (TRM), γδ T cells, proliferating, and EM clusters (Fig. [175]5E and Suppl. Table 8). Notably, ISG15 has been implicated in neurocognitive disorders and proposed as a potential biomarker for disease progression in PWH [[176]63–[177]65]. Interestingly, USP18, a negative regulator of ISG15 signaling [[178]66], was also significantly upregulated in some of these clusters (γδ T cells, proliferating T cells, and EM clusters) (Fig. [179]5E). In the EM, NK cell and proliferating clusters, CD38, GZMA, STMN1, MKI67, MX1, MX2, IFI2 were significantly upregulated in acutely infected animals (Fig. [180]5E), indicating the critical roles of those three clusters in viral defense. Interestingly, we found that the CD8⁺ T cells in EM cluster were predominantly derived from acutely infected animals (Fig. [181]5G), further highlighting the profound impact of acute SIV infection on brain EM T cell activation. EM, NK cell and proliferating clusters also had more significantly downregulated genes in acute infection compared to other lymphocyte clusters. Those downregulated genes include the genes related to T cell apoptosis (e.g. SEL1L3 and BCL2) (Fig. [182]5F). To assess those transcriptomic changes in brain lymphocytes during acute infection may be influenced by indirect interactions with other immune populations, specifically brain myeloid cells, we performed cell–cell communication analysis on brain myeloid and lymphoid cell populations derived from the same animals. The clustering and annotation of brain myeloid cells were previously described [[183]32]. For the purpose of simplifying communication analysis, myeloid cell subsets were consolidated into four major groups: homeostatic microglia (HomeoMG), defined by high expression of canonical homeostatic genes and absence of activation markers; inflammatory microglia (InflamMG), characterized by upregulation of activation-associated genes and/or downregulation of homeostatic markers; and CD14⁺ and CD16⁺ CNS-associated macrophages (CD14CAM and CD16CAM), defined by elevated expression of MHC class II molecules and CD14 or CD16, respectively, with concomitant loss of microglial signature genes. Globally, cell–cell signaling between brain myeloid and lymphoid populations was attenuated under acute infection, including outgoing signals from lymphocytes to myeloid cells and from myeloid cells to lymphocytes (Fig. [184]6A). However, specific phenotypes exhibited increased activity: EM T cells showed enhanced signaling toward brain myeloid populations, while inflammatory microglia exhibited increased signaling toward brain lymphocytes. A similar pattern was observed in the myeloid T cell cluster, where total signaling was reduced in acute infection, yet incoming signals from EM cells were notably elevated (Fig. [185]6B). While the overall number of signaling interactions declined during acute infection, the strength of retained interactions was often increased, especially in EM T cells and inflammatory microglia. For example, EM T cells showed upregulated outgoing signals via the ADGRE, MIF, MPZ, CD48, CCL, LCK, ANNEXIN, PARs, PROS, and CD6 pathways (Fig. [186]6C), and received augmented input through ADGRE, MIF, MPZ, SPP1, LCK, PECAM2, RESISTIN, and ICAM (Fig. [187]6D). Inflammatory microglia similarly demonstrated increased bidirectional signaling activity under infection. Collectively, these findings suggest that transcriptional changes in brain lymphocytes during acute infection may be, at least in part, driven by inflammatory myeloid cell interactions, while reciprocal activation of myeloid populations may be modulated by EM T cell–mediated signaling. Fig. 6. [188]Fig. 6 [189]Open in a new tab Cell-cell communications between brain myeloid and lymphoid cells in infected and uninfected conditions. (A) The communication strength between brain lymphoid cells and myeloid cells. The strength of signals sent by lymphoid cells were showed on upper panel and the strength of signals sent by myeloid cells were showed on lower panel. (B) The communication strength between myeloid T cells and brain lymphoid/myeloid cells. The strength of signals sent by myeloid T cells were showed on upper panel and the strength of signals received by myeloid T cells were showed on lower panel. (C-D) Heatmaps for signals contributing the most to (C) outgoing or (D) incoming signaling of the brain myeloid and lymphoid clusters in acute infected animals (left) and uninfected animals (right). The top-colored bar plot shows the total signaling strength of each cell cluster by summarizing all signaling pathways. The right gray bar shows the total signaling strength of a signaling pathway by summarizing all cell groups Infected CD4 + cytotoxic T lymphocytes could be the primary lymphoid cells seeding the SIV in the CNS The infected CD4 + T cells in the bloodstream are one of the ‘Trojan horse’ candidates that can bring the HIV/SIV into the CNS during acute infection. However, different lymphocyte phenotypes might have different susceptibilities to HIV/SIV and varied chemotactic abilities, which determine their capabilities in seeding the virus in the brain. Among the lymphocyte clusters in the brain, the CD4 + CTL, TRM, and Myeloid-T cell clusters were the top 3 lymphocyte clusters with the highest proportion of SIV-infected cells. The percentage of SIV + cells in those three clusters was above or equal to 2%, and the SIV + cells in those three clusters contributed to 63% of SIV + cells in the brain (Fig. [190]7A). The CD8 + effector, NK, and γδ T cell clusters were the least infected as expected. For the blood lymphocytes, most of the infected cells were found in the CD4 + CTL cluster and thus comprised ~ 51% of the infected cells. While CD4 + Naïve cells comprised the second highest proportion of infected cells at 29%, they are an abundant cell type in the blood and the overall infection rate was much lower (2.2%) (Fig. [191]7B). Fig. 7. [192]Fig. 7 [193]Open in a new tab Characterization of SIV-infected cells in the brain and blood. (A-B) The UMAP projection of cells isolated from the brain and blood with the number and percentage of SIV + cells in each brain/blood cell cluster (upper panel). The pie chart shows the relative contribution of each brain/blood cell cluster to the SIV + cells (lower panel). (C-D) The heatmap for expression of genes that might contribute to the CNS infection (i.e. CD4, CCR5, ITGAL, ITGB2, ITGA4, ITGB1) in each brain/blood cell cluster. The scaled data was used for this plotting. (E) The UMAP projection for SIV-positive cells in blood and brain. (F) The UMAP projection of normalized RNA expression for known T cell, myeloid cell, and proliferating markers. (G) The UMAP projection for all infected brain and blood lymphocytes. The cells were colored by clusters (left) or pseudotime (right). The grey circles labeled on the UMAP indicate the differentiation outcomes (cell fate) for setting Cluster 1 as the root node. (H) The Sankey plot showed the connection between brain and blood lymphocyte phenotypes for infected cells. The left (red) and right (blue) cluster information are generated by clustering separately for the blood and brain datasets. The cluster information in the middle (green) is through clustering on infected cells The infection rate in different lymphocyte clusters corresponded to their expression of CD4 and CCR5 (Fig. [194]7C and D). The average expression of CD4 in CD4 + CTL, TRM, and Myeloid-T cell clusters in the brain was higher than others, which might be one of the explanations for more SIV transcripts found in them. Similar to the CD4 + CTLs in the brain, this phenotype also had highest expression of CD4 and CCR5 compared to other CD4 + T cell clusters in the blood (Fig. [195]7D). Based on above analyses, we hypothesized that CD4 + CTLs might be the primary CD4 + T cell phenotype seeding the SIV to the CNS, which was further confirmed by performing clustering only on SIV + cells in the brain and blood. Five clusters for SIV + cells were identified (Fig. [196]7E). Cluster 0 and Cluster 2 expressed cytotoxic genes, Cluster 2 also co-expressed proliferating genes, Cluster 1 expressed naïve T cell genes, Cluster 3 expressed naïve T cell and cytotoxic genes, and Cluster 4 co-expressed lymphoid and myeloid cell genes (Fig. [197]7F). The pseudotime analysis showed that the infected cells in Cluster 0, 2, 3, and 4 were differentiated from the infected cells in Cluster 1, and the infected cells in Cluster 2 was most differentiated (Fig. [198]7G). This differentiation was driven by upregulation of cytotoxic and proliferating genes, and downregulation of naïve T cell genes (Suppl. Figure 5 A). By comparing infected cell clusters in brain and blood, we found that the well-differentiated clusters (Cluster 0 and Cluster 2) were abundant in both blood and brain, on the other hand, the cluster in naïve state (Cluster 1) were barely found in the brain (Fig. [199]7H). This suggests that the well-differentiated CD4 + CTLs were the primary lymphocyte phenotype seeding the SIV in the brain. The enhanced infiltrating ability of this phenotype could be attributed to their higher expression of the integrin alpha and beta chains for two adhesion molecules (LFA-1 and VLA-4) mainly expressed on lymphocytes compared to the other lymphocyte clusters (Fig. [200]7C and D). The brain environment might promote the maturation of infiltrating lymphocytes from the bloodstream To further understand the connections between the lymphocytes in peripheral blood and CNS. We aggregated all brain and blood lymphocytes from infected and control animals and reperformed clustering, which generated 10 different cell clusters (Fig. [201]8A). Few overlaps between lymphocytes from brain and blood were found (Fig. [202]8B), which could be caused by most brain lymphocytes being CD8+ [[203]67, [204]68]. Still, we found that three clusters (i.e. Clusters 1, 3, and 4) contained the T cells from both brain and blood. Cluster 1 contained 40% of cells from the CD4 + CTL(C2) cluster in the blood and 60% of cells from the CD4 + CTL cluster in the brain, indicating most CD4 + CTLs in the brain might be derived from the CD4 + CTL(C2) cluster in the blood. Most of the lymphocytes in Cluster 3 were from CD4 + EM or CD4 + Treg in the blood, but ~ 12% of cells in this cluster were from the TRM cluster in the brain, which indicates the CD4 + EM and Treg might be the progenitor cells for the CD4 + TRM cells in the brain. Cluster 4 contained the proliferating CD4 + CTLs in the blood and the proliferating CD8+/CD4 + lymphocytes in the brain (Fig. [205]8C and D). Fig. 8. [206]Fig. 8 [207]Open in a new tab Characterization of connections between isolated blood and brain cells. (A) The UMAP projection for all 52,148 brain and blood cells. The cells are labeled by new cluster information by reclustering all 52,148 cells. (B) UMAP projection as in (A) labeled by the tissue they are isolated from. (C) UMAP projection as in (A) labeled by the previous cluster information generated by clustering on brain and blood cells separately. ) The Sankey plot showed the connection between brain and blood lymphocyte phenotypes. The left (red) and right (blue) cluster information are generated by clustering separately for the blood and brain datasets. The cluster information in the middle (green) is through clustering on the merged dataset. The bar plot showed the top 20 upregulated and downregulated pathways enriched for the (E) brain CD4 + CTLs and (F) CD4 + TRMs. The Gene Set Enrichment Analysis (GSEA) was used for analyzing the altered pathways of the CD4 + CTLs and CD4 + TRM cells in the brain compared to CD4 + CTLs and CD4 + EM cells in the blood. The pathways with p.adjust < 0.05 were selected as enriched pathways and they were then ranked by normalized enrichment score (NES) for top 20 pathways. (G) The UMAP projection for all isolated brain and blood cells. The cells were colored by clusters generated by clustering on brain and blood cells separately. The grey circles labeled on the UMAP indicate the differentiation outcomes (cell fate) for setting the CD4 + Naïve T cell as the root node. (H) The box plot showing the predicted pseudo-time for cells in different clusters of brain and blood. The boxes were ordered by the median pseudo-time value for each cluster Aggregating the blood and brain lymphocytes for analyses confirmed that CD4 + CTLs in the blood were the primary lymphocytes infiltrating the brain, regardless of infection. Additionally, CD4 + EM and Treg cluster from the blood aggregated with TRM cluster from the brain raises the possibilities that those cells might further differentiate into, or be replaced by, TRM cells in the brain. To understand How the brain environment changes the lymphocyte transcriptomic profiles, we compared the transcriptomic profile between brain and blood lymphocytes in Cluster 1 (CD4 + CTLs) and Cluster 3 (CD4 + EM and Treg in the blood and TRM in the brain). We found that the downregulated genes for TRM in the brain included ribosomal genes (Suppl. Figure 5 C). For the upregulated genes, it included the genes for lymphocyte cytotoxicity (e.g. CST7), TCR signaling (e.g. SKAP2), and lymphocyte motility (e.g. S100A4 and RGS1). For CD4 + CTLs, once they infiltrated from the blood to the brain, they also downregulated the expression of ribosomal genes and upregulated the transcription factors related to lymphocyte maturation/activation (e.g. FOS, FOSB, VAV3, ZEB2, KLRD1, and IFRD1) (Suppl. Figure 5D). This suggests that the brain environment might reshape the CD4 + CTLs by affecting their gene regulations. The gene expression alterations aligned with observed pathway changes for CD4 + CTLs and TRM cells in the brain (Suppl.Table 9). Upon entering the brain, CD4 + CTLs and CD4 + TRM cells downregulated protein and peptide biosynthesis and metabolic pathways. Conversely, they upregulated pathways associated with intracellular signal transduction, protein phosphorylation, cell differentiation, and cellular development (Fig. [208]8E and F), reflecting increased cellular activity. Then trajectory analysis was performed to predict the differentiation of different lymphocyte phenotypes in the brain and blood. We found 10 differentiation cell fate outcomes when the CD4 + naïve T cell cluster was set as the root node for predicting the differentiation path (Fig. [209]8G). Similar with what we found for infected cells, the differentiation of lymphocytes was driven by upregulation of cytotoxic and proliferating genes, and downregulation of naïve T cell genes (Suppl.Figure 5B). The pseudotime analysis revealed that the brain lymphocyte phenotypes were more differentiated than blood lymphocyte phenotypes when comparing within cytotoxic or non-cytotoxic lymphocyte populations (Fig. [210]8H). Therefore, it further implied that the same or similar lymphocyte phenotypes might be more differentiated by upregulating cytotoxic and proliferating genes and downregulating naïve T cell genes, when they migrate from the peripheral blood to the brain. 21 days of acute SIV infection led to more infected tissue resident memory T cells but fewer infected proliferating cytotoxic T cells in the brain To better understand the lymphocyte responses at different time points of acute SIV infection, we integrated a dataset [[211]31] studying responses of brain lymphocytes in 21 days of acute SIV infection. The longer infection time in their study compared to this study (12 days) allowed for a comparison of lymphocyte responses at different points of acute SIV infection. The data integration of 46,231 lymphocytes from 12 animals (three for 12-day infection, four for 21-day infection, and five for uninfected controls) followed the same pipelines as our other analyses. The batch effect between samples from different datasets was corrected (Fig. [212]9A). The clustering generated eight cell clusters. The clusters from the integrated dataset were characterized by mapping the annotations from the 12-day infection data and identifying DEGs for each new cluster (Suppl. Table 10). Both methods confirmed a consistent annotation on this new integrated dataset (Fig. [213]9B). Fig. 9. [214]Fig. 9 [215]Open in a new tab Brain lymphocyte comparison between 12 days and 21 days of acute SIV infection. (A) The UMAP projection of 46,231 T lymphocytes from two scRNA-seq datasets. The 12-day dataset included three infected animals (12 days) and three uninfected animals (ctl_12 days). The 21-day dataset included four infected animals (21 days) and two uninfected animals (ctl_21days). (B) The UMAP projection of 46,231 T lymphocytes was clustered into 8 distinctive lymphocyte clusters (left). The relative contribution of each identified cluster in uninfected (ctl_21days) and 21-day acute-infected brain lymphocytes (right). (C) The genes were upregulated and downregulated for brain lymphocytes in 21-day infection compared to 12-day infection. The lymphocytes, excluding Myeloid T cells from 12-day and 21-day infections, were aggregated for a pseudobulk analysis. The statistical test for marker detection was the DEseq2 with parameters as adjusted p-value < 0.05 and absolute log[2]FC > 2. (D) Characterization of SIV + T cells from 12-day infected brain and blood samples and 21-day infected brain samples. (E) The UMAP projection of 985 SIV + T cells from 12-days and 21-days infected brains and 12-days infected bloods (left). The relative contribution of each identified SIV + T cell cluster in blood and brain samples (right). (F) The UMAP projection of 278 SIV + T cells from 12-days and 21-days infected brains (left). The relative contribution of each identified SIV + T cell cluster in 12-day and 21-day brain samples (right) We examined the lymphocyte population changes in 21-day infection. Except for the Myeloid-T cell cluster, the other T cell phenotypes found in 12-day infection were all found in 21-day infection. Consistent with our 12-day infection (Fig. [216]3J), the CD4 + CTLs decreased, and effector memory (EM) T cells increased in the 21-day infection compared to their controls (Fig. [217]9B). However, the decrease of CD8 + effector T cells and increased TRM and γδ T cells in the 21-day infection was the opposite of 12-day infection. This could be caused by the variations between different datasets. Then, we determined the transcriptomic differences of all brain lymphocyte clusters between 12-days and 21-days SIV infection by performing DEG analysis at pseudobulk level (Fig. [218]9C). We found that the longer acute infection led to the upregulation of mitochondrial genes, indicating increased cellular respiration and possibly response to stress. In addition, the lymphocytes in 21-day infection also upregulated several chemokine receptors (e.g. CXCR5, CCR6, and CCR7), cytokines (e.g. IL23A), and transcription factors related to immune activation (e.g. FOS, FOSB, and NFKB1) compared to 12-days infection, suggesting an enhanced activation state of those cells. To validate our findings for infected cells, we reclustered the SIV + lymphocytes found in the blood (707 cells) and brain (126 cells) from the animals at 12 days p.i., and those found in the brain (152 cells) at 21 days p.i., totaling 985 cells. Because the 21-day infection dataset did not include Myeloid-T cells, the SIV + cells in this cluster from the 12-day infected animals were excluded from the analysis. The characterization showed naïve, cytotoxic, proliferating, and EM/TRM, phenotypes (Fig. [219]9D). Consistent with what we found in previous (Fig. [220]4E and Fig. [221]7E), there was a cluster co-expressed naïve T cell markers and cytotoxic genes (primarily derived from CD4 + CTL (C7) cluster), indicating a transitional state. We named this cluster as naïve-cytotoxic cluster. Notably, the cells in proliferating cluster were also CD4 + CTLs. In addition, the comparison of infected lymphocytes between blood and the brain still showed that CD4 + CTLs were still the primary phenotype presenting in both brain and blood (Fig. [222]9E). However, compared to 12-day infection, there were more SIV + EM/TRM and fewer SIV + proliferating T cells in the brain of the 21-day infection (Fig. [223]9F), indicating a different pattern of infection in the brain lymphocyte populations. Interestingly, after integrating the data from 21-day infection, EM/TRM cluster showed an elevated expression of activation and cytotoxic genes (Fig. [224]9D and Suppl. Table 11), which suggests an enhanced cytotoxicity of those cells in 21-day infection. Discussion The parenchyma of the CNS was long thought to be an immune-privileged site, shielded by tight cellular barriers from the blood making it inaccessible to circulating immune cells. While myeloid cells of various lineages are present in the brain it is increasingly recognized that lymphocytes and particularly T cells can be found in the brain parenchyma and perivascular space under physiological as well as diseased conditions. Given the overwhelming predominance of infected CD4 + T cells in the acute phase of HIV and SIV infection, we investigated their role in initiation of CNS infection. In this study we indeed found SIV-infected CD4 + T cells in the brain in the early stages of infection and identified a unique stage in which cells of a dual myeloid-T cell phenotype. We propose that this is the mechanism for CNS infection by SIV and HIV – infected T cells capable of entering the brain encounter brain macrophages or microglia, which then engulf the infiltrating T cells, and transfer the infection to these brain myeloid cells. While we focused on CD4 + cells, the primary target for SIV and HIV, most of the T cells in the brain are of the CD8 + phenotype [[225]67, [226]68], in contrast to the predominance of CD4 + T cells found in CSF [[227]69]. The predominance of CD8 + T cells in the brain parenchyma and perivascular space are likely imperative in CNS immune surveillance, particularly to viral infection. In HIV infection, the CD8 + T cells also dominate the CSF, resulting in a lower CSF CD4/CD8 T cell ratio [[228]70]. Infiltration of CD8 + T cells to the brain and CSF can contribute to the clearance of virus and infected brain cells. However, it might also trigger severe neurocognitive impairment and cerebral inflammation [[229]27, [230]28]. Although CD8 + T cells are the primary phenotype found in the brain under infection, the initiation of CNS infection is mostly caused by CD4 + T cells. Consistent with what has been reported, fewer CD4 + T cells were found in our brain lymphocyte dataset compared to CD8 + T cells, which include uninfected and 12-day SIV-infected rhesus macaque brain samples. Our scRNA-seq analysis found that those brain CD4 + T cells were primarily in two cell clusters: a tissue-resident memory (TRM) T cell cluster and a cytotoxic T cell cluster. The TRM cells in the brain and other non-lymphoid tissues were generally considered as the cells that differentiate from circulating effector/effector memory T cells that encounter antigens during infections, vaccinations, or inflammatory events [[231]71, [232]72]. The integration analysis of brain and blood lymphocytes performed in this study also showed that some of the TRM cells in the brain (likely CD4+) shared transcriptomic profiles with CD4 + EM and Treg in the blood. (Fig. [233]8D) Given their memory, TRM cells can quickly respond to reencountered antigens [[234]73, [235]74]. The TRM cells in the brain have been found to contain both CD4 + and CD8 + T cells, and they share many similarities in their surface markers [[236]56]. The inseparable CD4 + TRM and CD8 + TRM cells in the TRM cluster found in our study (Fig. [237]3A and D) indicated the transcriptomic convergence of those two brain TRM T cell phenotypes. The TRM cells were dramatically depleted in 12-day SIV infection, inconsistent with a rapid increase of TRM (CD8+) reported in virus infection [[238]75, [239]76]. This might be because most of the depleted cells in our study were CD4+, and CD8 + TRM in infected brains changed their transcriptomic profile by upregulating proliferating and cytotoxic genes, so those cells were clustered to proliferating cytotoxic T cell cluster. Indeed, compared to uninfected controls, the proliferating cytotoxic T cell cluster had an over 10-fold increase in 12-day SIV infection. The CD4 + TRM in the brain and one of its potential progenitor cells in the blood (i.e. CD4 + EM) were susceptible to SIV infection, with a high percentage of SIV + cells (Fig. [240]7A and B). The high expression of LFA-1 in CD4 + EM also suggested their potential contributions in seeding the SIV/HIV to the CNS. Interestingly, the comparison of SIV + T cells in the brain with 12-day and 21-day SIV infection showed that the longer infection time in acute SIV infection dramatically promoted the SIV expansion in TRM, indicating they might serve as one of the brain reservoirs for HIV/SIV. As another primary CD4 + lymphocyte population in the brain, CD4 + CTLs were found to decrease in the brain for both 12-day and 21-day acute SIV infection. This lymphocyte phenotype also could be found in the peripheral blood of infected and uninfected animals. Unlike its change in the brain, it was increased in the peripheral blood. This discrepancy may be attributed to the exclusion of CD8⁺ T cells and natural killer (NK) cells in the peripheral blood samples analyzed, which may have confounded the interpretation. CD4⁺ CTLs expressed high levels of genes associated with cytotoxicity and activation, complicating their distinction from CD8⁺ CTLs. DEG analysis for CD4 + and CD8 + CTLs in the brain showed that CD8 + CTLs highly expressed some NK cell markers (e.g. KLRC3, GNLY, and FCGR3), which could not be seen in CD4 + CTLs. The increased cytotoxicity in CD4 + CTLs compared to other CD4 + T cell phenotypes suggest its enhanced defending ability. Indeed, the CD4 + CTLs have been widely reported to contribute to virus clearance in numerous infectious diseases [[241]77–[242]82]. Given this, the concept of HIV-specific CD4 T cell response was proposed for controlling HIV replication in the absence of antiretroviral therapy [[243]83, [244]84]. However, given their longevity, the CD4 + CTLs were also found to serve as reservoirs for HIV-1 [[245]85]. Our study here contributes to understanding the roles of CD4 + CTLs in the context of acute SIV brain infection. More infected cells were found in CD4 + CTL cluster compared to other T cell cluster for both brain and peripheral blood, likely due to their elevated expression of CD4 and CCR5 (Fig. [246]7C and D). They also had enhanced chemotactic ability with high expression of VLA-4 and LFA-1 to seed the CNS infection. These observations support our hypothesis that CD4⁺ CTLs may act as key “Trojan horse” cells mediating viral entry into the brain during acute HIV/SIV infection. Thus, targeting chemotactic receptors such as VLA-4 and LFA-1 may represent a therapeutic strategy to prevent viral seeding of the CNS. Additionally, trajectory analysis indicated that CD4⁺ CTLs represent a terminally differentiated subset of CD4⁺ T cells, and that only highly differentiated T cells are capable of entering the brain. This suggests that therapeutic interventions aimed at maintaining CD4⁺ T cells in a less differentiated state could potentially limit their CNS infiltration and thereby reduce viral invasion during acute infection. A myeloid–T cell hybrid cluster, characterized by the co-expression of myeloid and lymphoid gene markers, was identified in the brains of both SIV-uninfected and acutely SIV-infected rhesus macaques, with a pronounced expansion during acute infection. To determine whether this cluster represents a genuine biological phenomenon rather than a technical artifact (e.g. doublets from scRNA-seq library preparation), we established an in vitro monocyte-derived macrophage (MDM) and Jurkat T cell co-culture system. Consistent with the in vivo data, cells resembling this Mac_T cluster were detected in both apoptotic and non-apoptotic co-culture conditions via scRNA-seq. Notably, a higher abundance of these hybrid cells was observed under non-apoptotic conditions, likely due to enhanced degradation of Jurkat cells during apoptosis, which may reduce detection of hybrid cell signatures. To further differentiate genuine Mac_T cells from technical doublets, we compared the transcriptomic profiles of co-cultured cells with those from monocultured controls. We found that the chemokines might serve as markers to distinguish the real Mac_T cells and technical doublets caused by scRNA-seq library preparations. Indeed, chemokines play an important role in recruiting other immune cells, activating macrophages and enhancing their phagocytic ability [[247]85]. Therefore, the upregulation of chemokines (e.g. CXCL3, CCL4, CCL3, CXCL2, CXCL1 etc.) might be the driving force for macrophages-T cell interactions. By applying this finding to the myeloid-T cell cluster in the brain of rhesus macaque, we found that over 50% of the cells in this cluster had co-expression of chemokines CCL3 and CCL4 and ~ 92% of cells had expression of at least one of those two chemokines. This suggests most cells in Myeloid-T cells are not technical artifacts but have biological importance. Interestingly, the Myeloid-T cell cluster contained 22 SIV + cells, which represents 2% of the total cells in this cluster. Given the average infection rate in brain lymphocytes is ~ 1.4%, the high infection rate in Myeloid-T cluster suggests that the interactions between infected lymphocytes and brain myeloid cells efficiently amplify brain infection. It was found that viruses could be effectively cell-to-cell transferred to macrophages from CD4 + T cells [[248]86–[249]89], and the cell-to-cell infection mechanism for HIV-1 transmission includes brain myeloid cells engulfing [[250]89] or fusing [[251]86, [252]87] with HIV-1-infected CD4 + T cells. Therefore, we hypothesized that chemokine-mediated recruitment of infected T cells by brain myeloid cells facilities viral spread through sustained cell-cell contact, enhancing viral transfer efficiency compared to cell-free infection. However, the MDM–T cell co-culture system does not fully recapitulate the complexity of interactions occurring during in vivo SIV infection. In particular, it lacks the viral context, including infected primary T cells and tissue-resident macrophages or microglia. Thus, while informative, this model is limited in its ability to reflect SIV/HIV-induced efferocytosis or phagocytosis. To better simulate the in vivo microenvironment, future studies will employ co-cultures of HIV-infected primary T cells with MDMs or microglia, followed by single-cell transcriptomic analysis. This approach will more comprehensively elucidate the role of myeloid-mediated clearance or engulfment of infected T cells in HIV-associated neuropathogenesis. Additionally, future investigations also should examine whether these myeloid–T cell interactions promote productive infection or viral latency in microglia and assess whether disrupting CCL3/CCL4-mediated chemokine signaling can dismantle this potential infectious niche. Understanding the molecular mechanisms underlying these interactions may reveal new therapeutic targets for mitigating HIV-associated neuroinflammation and viral persistence in the CNS. In addition to brain infection and the establishment of viral reservoirs, the inflammatory milieu of the brain during the acute phase of infection warrants particular attention. In this study, we identified significantly altered genes across distinct brain lymphocyte clusters. Notably, the EM cluster exhibited the greatest number of DEGs, indicating substantial transcriptomic changes in this population during acute infection. Furthermore, we observed a consistent upregulation of cytotoxic, proliferative, and antiviral genes within the EM, NK cell, and proliferating clusters, highlighting their critical roles in amplifying neuroinflammation during the acute phase. Among these findings, ISG15 and USP18 were significantly upregulated across multiple brain lymphocyte clusters. These two genes act antagonistically. ISG15 deficiency and USP18 overexpression have been associated with enhanced HIV-1 replication [[253]90]. From the perspective of neuroinflammation, a potential contributor to neurocognitive disorders, ISG15 may exacerbate disease progression by promoting type I interferon signaling [[254]63, [255]65]. Conversely, USP18, as a negative regulator of ISG15 and interferon signaling, may help maintain immune homeostasis in the brain. Despite the global upregulation of both genes in brain lymphocytes, their opposing functions present a therapeutic challenge. Tailoring intervention strategies, such as inhibiting or promoting specific molecules depending on the stage of infection, may be essential for achieving therapeutic efficacy without disrupting immune balance. This study comprehensively showed the transcriptomic profile of brain lymphocytes responding to a 12-day acute SIV infection. The overall schematic model is proposed based on the data shown in this study (Fig. [256]10). By comparing with the blood CD4 + T cells, we identified the primary CD4 + T cell phenotypes that carry the SIV/HIV to the CNS. Further characterization and analysis showed that the chemotactic ability and expression of virus entry receptors (e.g., CD4 and CCR5) in the peripheral lymphocytes were highly correlated with their ability to seed CNS infection. Our scRNA-seq data showed that LEF-1 and VLA-4 were highly expressed in the lymphocyte phenotype in both blood and brain, indicating their essential roles in mediating the HIV/SIV invasion of the brain. However, more adhesion molecules and chemokine receptors might also be involved and worth further study to benefit therapeutic development. The integration of brain and blood lymphocyte datasets partially revealed the transcriptomic alternations of some lymphocytes (CD4+) once they enter the brain. The unique brain milieu could reshape those peripheral CD4 + T cells to adapt to the brain environment. However, because the CD8 + T cells were absent in the blood samples, the connection of CD8 + T cells in the brain and blood was lacking in this study. Additionally, we found that distinguishing between brain CD4 + and CD8 + T cells was challenging using scRNA-seq data. Many brain T cells in our dataset and the 21-day infection dataset [[257]31] were neither CD4 + nor CD8+, which could also be attributed to the protein expression of those genes being more abundant than RNA expression (92). Therefore, the characterization of those brain lymphocytes might be further improved when the protein expression profile could be integrated with the RNA expression profile [[258]85]. Finally, given the advantage of large publicly available scRNA-seq data in the area of studying HIV/SIV infection, we also had a comparison of brain T cell responses at different time points of acute SIV infection. The deposition of sequence data and metadata from our studies and those of others in publicly accessible databases can further benefit science and enable the building of larger analyses with more subjects. Fig. 10. Fig. 10 [259]Open in a new tab Schematic model summarizing the interactions between brain and blood lymphocytes and their hypothesized contributions to SIV infection in the CNS. This model illustrates the proposed mechanisms by which SIV-infected lymphocytes infiltrate the brain and contribute to neuropathogenesis. In the periphery, SIV primarily infects CD4⁺ T cells. Among these, only differentiated CD4⁺ T cell subsets, characterized by the upregulation of adhesion molecules such as LFA-1 and/or VLA-6, are capable of trafficking across the blood-brain barrier and seeding the CNS. CD4⁺ cytotoxic T lymphocytes (CTLs), which co-express both LFA-1 and VLA-6, appear to be the predominant CD4⁺ T cell phenotype responsible for viral seeding in the brain. Upon entry into the CNS, CD4⁺ effector memory (EM) and regulatory T cells (Tregs) may further differentiate into tissue-resident memory T cells (TRMs) with enhanced cytotoxic potential. These infected lymphocytes can be phagocytosed by brain-resident myeloid cells, potentially facilitating viral dissemination within the myeloid compartment. In parallel, as part of the host immune response to CNS infection, EM CD4⁺ T cells, CD8⁺ effector T cells, NK cells, and TRMs upregulate cytotoxic, antiviral, and activation-related genes. While these responses may contribute to antiviral defense, they also likely amplify neuroinflammation. This illustrative figure was generated using Biorender with license [260]https://BioRender.com/a2yx9c6 Supplementary Information [261]Supplementary Material 1.^ (27.6MB, docx) [262]Supplementary Material 2.^ (15.5KB, xlsx) [263]Supplementary Material 3.^ (112.9KB, xlsx) [264]Supplementary Material 4.^ (49.4KB, xlsx) [265]Supplementary Material 5.^ (48.3KB, xlsx) [266]Supplementary Material 6.^ (323KB, xlsx) [267]Supplementary Material 7.^ (326.1KB, xlsx) [268]Supplementary Material 8.^ (354.2KB, xlsx) [269]Supplementary Material 9.^ (209.7KB, xlsx) [270]Supplementary Material 10.^ (326.8KB, xlsx) [271]Supplementary Material 11.^ (249.4KB, xlsx) [272]Supplementary Material 12.^ (202.9KB, xlsx) [273]Supplementary Material 13.^ (692.1KB, mov) [274]Supplementary Material 14.^ (829.6KB, mov) [275]Supplementary Material 15.^ (1,021KB, mov) Acknowledgements