From August 2017 and November 2019, we recruited 134 adults PWH without diabetes (fasting blood glucose (FBG) <100 mg/dl and/or hemoglobin A1c (HbA1c) <5.7%), with pre-diabetes (FBG 100-125 mg/dl and/or HbA1c 5.7-6.4%) or with diabetes (FBG ≥ 126 mg/dl and/or HbA1c ≥6.5% or on anti-diabetic medications) to the HIV, Adipose Tissue Immunology and Metabolism (HATIM) study from the Vanderbilt Comprehensive Care Clinic, an academic, urban HIV treatment facility (17). All participants were on ART combination therapy for ≥18 months, with a minimum of 12 months of sustained plasma viral suppression, a CD4⁺ T cell count >350 cells/μl, and no known inflammatory or rheumatologic conditions. Exclusion criteria were self-reported heavy alcohol use (>11 drinks/week), known liver cirrhosis, active hepatitis B or C, cocaine or amphetamine use, and use of corticosteroids or growth hormones. Anthropometric measurements including waist circumference, height, weight, and body mass index (BMI) were obtained on the day of recruitment (Table 1). Diabetic PWH were older, with significantly fewer smokers. Participants provided written informed consent, and the study was approved by the Vanderbilt University Institutional Review Board. The study is registered at ClinicalTrials.gov (NCT04451980).

Subcutaneous adipose tissue was obtained from participants by liposuction and the stromal vascular fraction (SVF) was processed within 30 minutes to 1 hour of the procedure as previously published in detail (17). Peripheral blood mononuclear cells (PBMCs) were processed by Ficoll gradient. PBMCs and SVF from all participants were cryopreserved and subsequent assays were performed at a later date in batches. We also re-analyzed data single-cell metabolic profiling of T cells obtained from healthy human donors at the Stanford Blood Center, according to the guidelines of the Stanford Institutional Review Board (29).

We performed non-contrast computed tomography (CT) imaging within 1 week of blood collection and anthropometric measurements. This was performed using a Siemens Somatom Force multidetector scanner (Erlangen, Germany). Total coronary arterial calcium (Agatston units, Au) was measured in the left anterior descending (LAD), left main (LM), left circumflex (LCX), and right coronary artery (RCA). For our analysis, coronary arterial calcium (CAC) was treated as a categorical variable (presence or absence of coronary CAC). The mean coronary cross-sectional area (external diameter of the outer walls) was measured at three equidistant points of the LAD. The mean coronary cross-sectional area (corCSA) was derived from the mean of three points and used as a surrogate for arterial remodeling. Non-alcoholic fatty liver disease was defined by liver attenuation, which was averaged from nine total regions using the open-source OsiriX software field (30). Perivascular adipose tissue (PAT) volume (adipose tissue around the LM coronary, LAD, circumflex, and RCA) and epicardial adipose were measured as previously described (30).

A multiparameter flow cytometry antibody panel was used to stain PBMCs (17, 18, 26). The panel used to define CGC⁺ cells included anti-CD3, CD4, CD8, CCR7, CD45RO, GPR56, CX3CR1, CD57, CD14, CD19, and LIVE/DEAD Aqua (Supplemental Table 1). We included Class I and Class II CMV tetramers with this panel to identify virus-specific T cells. CMV Class I (pp65 [HLA-A02 NLVPMVATV (NLV)]) and Class II tetramers (gB [HLA DR1:07 DYSNTHSTRYV (DYS), pp65 [HLA DR1:03 EFFWDANDIYRIF (EFF)], IE2 [HLA DR1:03 TRRGRVKIDEVSRMF (TRR), IE1 [HLA DR1:03 VKQIKVRVDMVRHRI (VKQ)] and pp65 [HLADQB1*06:02 LLQTGIHVRVSQPSL (LLQ)] were obtained from the NIH tetramer facility supported by contract 75N93020D00005 from the National Institute of Allergy and Infectious Diseases. The analysis was performed using the BD FACS Aria II flow cytometer. Bulk and single-cell sorting was performed using a 70μm nozzle into 96 well plates or Eppendorf tubes, respectively, as previously published (17). An additional panel with fluorescently tagged antibodies was used to further characterize KLRG1, CD27, and CD28 expression on CGC⁺ T cells (anti-GPR56, CCR7, CD38, KLRG1, CD14, CX3CR1, CD45RO, CXCR3, PD1, CD27, CD57, CD3, Live/Dead stan, CD8, CD4, CD28, and CXCR5) (Supplemental Table 1). These samples were run on a Cytek/Aurora. We used Cytobank to analyze the flow cytometry and mass cytometry data (31). The gating strategy used to define the immune subsets is shown in the Supplemental Material (Supplemental Figure 1).

Plasma samples were aliquoted at 25 µl and spiked with 5 µL of metabolomics internal standards solution. Extraction of metabolites was performed by protein precipitation by adding 200 µL of 8:1:1 Acetonitrile: Methanol: Acetone (Fisher Scientific, San Jose, CA) to each sample. Samples were mixed thoroughly, incubated at 4°C for 30 min to allow protein precipitation, and centrifuged at 20,000xg to pellet the proteins. After centrifugation, 190 µl supernatant was transferred into a clean microcentrifuge tube and dried under a gentle stream of nitrogen at 30°C (Organomation Associates, Inc., Berlin, MA). Samples were reconstituted with 25 µL of injection standards solution, mixed, and incubated at 4°C for 10-15 min. Reconstituted samples were centrifuged at 20,000xg and supernatants were transferred into LC-vials for LC-MS analysis.

LC-MS untargeted metabolomics was performed on a Thermo Q-Exactive Orbitrap mass spectrometer equipped with a Dionex UPLC system (Thermo, San Jose, CA). Separation was achieved on an ACE 18-pfp 100 x 2.1 mm, 2 µm column (Mac-Mod Analytical, Inc., Chadsford, PA) with mobile phase A as 0.1% formic acid in water and mobile phase B as acetonitrile (Fisher Scientific, San Jose, CA). The gradient was run at a flow rate of 350 µL/min and consisted of: 0-3 min, 0% B; 3-13 min, 80% B, 13-16 min, 80% B, 16-16.5 min, 0% B. The total run time was 20.5 min. The column temperature was set at 25°C. The injection volume was 4 µL for negative and 2 µL for positive polarity. All samples were analyzed in positive and negative heated electrospray ionization with a mass resolution of 35,000 at m/z 200 as separate injections. The heated-electrospray conditions are 350°C capillary temperature, 3.5 kV capillary voltage, 50 sheath, and 10 arbitrary units of auxiliary gas. LC-MS injection was done following a sequence of 3 blanks, neat QC, pooled QC, 10 randomized samples, blank, neat QC, pooled QC, 10 randomized samples, and so on.

The percent relative standard deviation of internal standard peak areas was calculated to evaluate extraction and injection reproducibility. The raw files were then converted to mzXML using MS Convert (ProteoWizard, Palo Alto, CA). Mzmine 2 was used to identify features, deisotope, align features and perform a gap filling to fill in any features that may have been missed in the first alignment algorithm. The data were searched against an internal retention time metabolite library. All adducts and complexes were identified and removed from the data set.

We used cytometry by time of flight (CyTOF) to further define cell surface markers expressed on CGC⁺ T cells. In brief, cryopreserved PBMCs were thawed and treated with Nuclease S7. After two washes, the cells were stained with LIVE/DEAD Cisplatin stain for 3 minutes, followed by quenching, and then stained with a master mix of CyTOF antibodies against surface markers. Sixteen percent PFA was used to fix cells for 15 minutes at room temperature. After one wash, we resuspended cells in 1mL cold methanol, and caps were sealed with parafilm before incubating overnight at -20°C. On the day the cells were to be analyzed, we washed them with 1X PBS/1% BSA. They were then stained with the intracellular marker CTLA4 for 20 minutes at room temperature. This was followed by staining with a 25μM DNA intercalator (Ir) in the presence of 1.6% PFA for 20 minutes at room temperature and then transferred to 4C until analyzed. Just before analysis on Helios, we washed cells with PBS followed by a wash with Millipore H2O. We resuspended 500,000 cells/ml (in minimum 500uL) ddH2O for the CyTOF run. We added 1/10th volume of equilibration beads to the cells and filtered the cells immediately before running. The Cytof panel included antibodies that define T cells (CD3, CD4, CD8) and memory subsets (CD45RA, CD45RO, CCR7).

For the metabolic profiling, cryopreserved PBMCs from healthy donors were thawed in a cell culture medium (CCM; RPMI 1640 containing 10% FBS, and GlutaMAX; Thermo Fisher Scientific) supplemented with 1:10000x Benzonase (Sigma-Aldrich). Cells from different donors were live cell barcoded using CD45 antibodies as previously described (32), then washed and combined for downstream cell staining. Cells were suspended in TruStain FcX Fc blocker (BioLegend) for 10 min at RT and washed in cell staining media (CSM: PBS with 0.5% BSA and 0.02% sodium azide) before staining. Surface staining was performed in CSM for 30 min at RT. Cells were resuspended in monoisotopic cisplatin-195 for 5 min to label non-viable cells (Fluidigm, 0.5 uM final concentration in PBS). Cells were washed in CSM and fixed and permeabilized using the Foxp3/Transcription Factor Staining Buffer Set (eBiosciences). Intracellular staining was performed in a permeabilization buffer for 30 min at RT. Cells were then washed and resuspended in intercalator solution (1.6% PFA in PBS and 0.5 mM rhodium-intercalator (Fluidigm)) for 1 hr at RT. Cells were washed and resuspended in CSM + 10% DMSO and cryopreserved. Before the acquisition, cells were thawed in CSM and washed twice in Cell Acquisition Solution (CAS; Fluidigm). All samples were filtered through a 35 mm nylon mesh cell strainer, resuspended in CAS supplemented with 1x EQ four-element calibration beads (Fluidigm), and acquired on a Helios mass cytometer (Fluidigm).

CGC⁺ and non-CGC⁺ CD4+ and CD8+ T cells were sorted from PBMCs stained with the multiparameter flow cytometry panel. Greater than 90% purity was confirmed by spot checks of sorted cells. The CGC⁺ T cells were expanded using the ImmunoCult™ Human CD3/CD28 T Cell Activator (Stem Cell Technologies, #10991). Cells were expanded per the manufacturer’s protocol with the replacement of media supplemented with human interleukin (IL)-2 (10 to 50ng/ml) every 2-3 days, depending on the density of the cells. Cells were expanded past 14 days and were re-stimulated once more by the addition of CD3/CD28.

CMV IgG levels were measured in the plasma by ELISA per the manufacturer’s protocol (Genway, # GWB-BQK12C).

Single-cell T cell receptor (TCR) sequencing was performed as published (17). In brief, we stained PBMCs and index-sorted CGC⁺CD4⁺ T cells by flow cytometry into 96- well plates containing 3µL of lysis buffer with a ribonuclease inhibitor (33, 34). We used uniquely tagged primers (TSOend primer and the constant region primers, TCRA or TCRB) for reverse transcription, which tags the cDNA with well-specific barcodes coupled with a unique molecular identifier (UMI) to allow for multiplexing. Samples from each well were then pooled and amplified using the KAPA HiFi HotStart ReadyMix (Roche, Basel, Switzerland). Nested polymerase chain reactions were performed to target the TCR region specifically. We purified the PCR products using Agencourt AMPure XP (Beckman Coulter, CA, UWA) and indexed libraries were created for sequencing using Truseq adapters. The prepared libraries were quantified using the KAPA Universal qPCR Library Quantification Kit (Kapa Biosystems Inc., MA, USA). The products were sequenced on an Illumina MiSeq using a 2×300bp paired-end chemistry kit (Illumina Inc., CA, USA). Reads were quality-filtered and passed through a demultiplexing tool to assign reads to individual wells and mapped to the TCRB and TCRA loci. We used the MIXCR software package to assign TCR clonotypes. We used the visual genomics analysis studio (VGAS), an in-house program for visualizing and analyzing TCR data (http://www.iiid.com.au/software/vgas).

Our group previously sorted CGC⁺CD4⁺ and CGC⁺CD8⁺ T cells, as well as other memory T cells, for RNA transcriptomic analysis as published (17). Here, we performed a secondary differential expression analysis to assess differences in 475 genes involved in metabolic pathways between cell types (Supplemental Table 3).

We measured interleukin (IL)-4, IL-10, IL-6, and highly sensitive reactive protein (hs-CRP) in plasma using a multiplex assay (Meso Scale Diagnostics, Rockville, MD) as previously published (35).

We analyzed the metabolic profile of CGC⁺CD4⁺ T cells and CGC⁺CD8⁺ T cells ex vivo using the SCENITH assay as published (36). In brief, cryopreserved PBMCs were rapidly thawed and resuspended in RPMI media supplemented with 10% fetal bovine serum. After two washes, the cells were stained with antibodies against CCR7 and CX3CR1 at 37°C for 15 minutes. The cells were added in duplicate per condition to 96-well plates at about 1 million cells per well in 180ul R10 media. They were rested for 15-30 minutes at 37°C. 2-Deoxy-D-glucose (2mM), oligomycin (3mM), and DGO (1mM 2DG and 1.5mM oligomycin) were added to the cells, which were incubated at 37°C for 30 minutes. This was followed by the addition of puromycin (10μM) to each well, incubated at 37°C. We included samples without puromycin as controls. The cells were incubated at 37°C for 30 minutes. Cells were spun down and washed with PBS twice. These were then stained with surface antibodies for 15 minutes at room temperature. The cells were washed twice and fixed using the Foxp3/Transcription factor fixation fix/perm solution (20 minutes) and then washed with the Foxp3/Transcription factor permeabilization buffer. This buffer was used to dilute the anti-puromycin antibody. Cells were incubated with the anti-puromycin antibody for 20 minutes at room temperature. Cells were then washed with PBS and immediately analyzed by flow cytometry.

We sorted CGC⁺ and non-CGC⁺ CD4⁺ and CD8⁺ T cells and expanded them as above to obtain enough cells for the Seahorse assay. After 12 days of expansion, 100,000 cells per well were plated on Cell-Tak (Corning) coated plates in Seahorse XF Base Medium (Agilent, 102353-100) supplemented with 1 mM L-glutamine and 1 mM pyruvate at pH 7.4. Seeded cells were centrifuged at 200 g without break and incubated for 1 h at 37° C in a non-CO (2) chamber. ECAR measurements were taken using the Agilent Seahorse XF96 analyzer under basal conditions and after consecutive injections with 10 mM Glucose, 1.5 uM Oligomycin, and 50 mM 2-deoxy-glucose (2-DG). Control wells with assay medium lacking cells were used for background measurements.

Continuous variables/clinical demographics are presented as median values [25^th and 75^th percentiles], and statistical analyses comparing the three metabolic groups were performed using the Kruskal-Wallis test. Differences between categorical variables, represented as proportions, were analyzed using the Chi-squared test. TCR and RNA transcriptomic analyses were performed using the visual genomics analytics studio tool (VGAS) (37). Differential gene expression between CGC⁺ T cells and non-CGC⁺ T cells was performed using Kruskal Wallis tests, and adjustment for multiple corrections using the Benjamini Hochberg (BH) method. The top differentially expressed genes, p-value < 0.1, were included in the KEGG pathway and Gene Ontology pathway enrichment analysis using Enrichr and Appster (38–40). The relationship between CGC⁺ T cells and plasma metabolites was analyzed using Spearman’s rank correlation. Metabolomic pathways represented by metabolites correlated with CGC⁺ T cells were analyzed using MetaboAnalyst 5.0. Over Representation Analysis (ORA) of the significant plasma metabolites was performed using the hypergeometric test. One-tailed adjusted p values are provided after correcting for multiple testing (FDR). We used the Kruskal-Wallis test to analyze differences in the proportions of immune cell subsets between two groups/treatments, and Wilcoxon tests when there were more than two groups. Relationships between immune subsets and anthropometric or clinical laboratory measurements were determined using Spearman’s rank correlation analysis. We also measured relationships between the immune subsets and other factors after adjustment for potential confounders using partial Spearman’s rank analysis. Statistical analysis was performed using R version 4.1.0 (41) and Prism version 9.

The data presented in the study are deposited in the NIH GeneExpression Omnibus repository, accession number GenBank: GSE159759. Differential gene expression was performed using custom software, Visual genomics analysis studio (VGAS) (37).

We previously defined CMV-specific CD4⁺ T cells that bind the DYSNTHSTRYV epitope (DYS, HLA-DR7, glycoprotein B (gB)) as CGC⁺, however the extent to which this subset of cells is CMV-specific is unknown (18). We characterized CGC⁺ T cells using additional markers that have been associated with CMV-specific T cells. Two-dimensional plots show CX3CR1, GPR56, and CD57 expression on CD8⁺ T cells in two HIV-negative donors (CMV-negative and CMV-positive) and four CMV-positive PWH (Figure 1A). We used the Uniform Manifold Approximation and Projection (UMAP) algorithm to visualize clusters of CD8⁺ T cells related by marker expression. The CGC⁺ cluster on CD8⁺ T cells had variable CD57 expression (Figure 1A, UMAPs) and expressed the killer-cell lectin-like receptor G1 (KLRG1, a marker associated with senescence) (Figure 1B). Additional markers included in the panel defined the CGC⁺ CD8⁺ T cell cluster as largely made up of T effector memory RA-revertant (TEMRA) cells (CD45RO^- CCR7^-), CD28^-, CD27^+/- and CD38⁺ (Supplemental Figures 2A, B).

We characterized the CMV-specificity of CD8⁺ T cells from a subset of HLA-typed individuals (Supplemental Table 2). In participant #1, we evaluated CMV-specific CD8⁺ T cells in the peripheral blood and adipose tissue using a class I tetramer against the HLA-A*02-01 binding CMV 65 kilodalton phosphoprotein (pp65) epitope_495-503 NLVPMVATV (NLV). NLV tetramer⁺ cells constituted 2.0% of total CD3⁺ T cells and 3.8% of total CD8⁺ T cells in the peripheral blood of participant #1 (Figures 2A, B). Two-dimensional flow cytometry plots show that NLV-tetramer⁺ CD8⁺ T cells co-express CX3CR1 and GPR56 with variable expression of CD57 (Figure 2C). 65.7% of the NLV-tetramer⁺ CD8⁺ T cells are TEMRA and the rest effector memory T cells (TEM) (Figure 2D). NLV tetramer⁺ cells were present in the cluster of CX3CR1⁺ and GPR56⁺ cells with variable expression of CD57 (Figure 2E). We gated on the CGC⁺ CD8⁺ cluster and show the expression of the NLV tetramer, CX3CR1, GPR56 and CD57 (Figure 2F). The coloring channel highlights NLV-tetramer⁺ cells (bright red) and shows that 8.4% of the CGC⁺ CD8⁺ T cell cluster from participant #1 express TCRs that bind the NLV tetramer. For this participant #1, we also analyzed NLV tetramer⁺ CD8⁺ T cells in the adipose tissue, given the contribution of adipose tissue inflammation to the development of cardiometabolic disease. 1.2% of total CD3⁺ T cells and 3.4% of total CD8⁺ T cells expressed the NLV TCR (Figures 2G, H). These NLV-specific CD8⁺ T cells present in the adipose tissue also co-expressed CX3CR1 and GPR56, with variable CD57⁺ expression. Like the matched peripheral blood, 54.1% of the NLV-tetramer⁺ T cells in adipose were TEMRA and the rest were TEM (Figures 2I, J). UMAPs show that NLV-tetramer⁺ CD8⁺ T cells present in the adipose tissue from participant #1 cluster with CGC⁺ CD8⁺ T cells, with a proportion of ~ 6.0% (Figures 2K, L).

To assess the heterogeneity among individuals, we analyzed CMV-specific responses in the peripheral blood of additional donors. Participant #2 had NLV⁺ tetramer cells that constituted 0.6% of total CD3⁺ T cells (Figure 2M) and 1.4% of total CD8⁺ T cells (Figure 2N). In this participant, NLV-specific CD8⁺ T cells were CX3CR1⁺ and GPR56⁺, with much less CD57 expression (Figure 2O). There was also a higher proportion of TEM cells among the gated NLV tetramer⁺ cells (Figure 2P). NLV tetramer-specific cells in participant #2 did not form a tight cluster as seen with participant #1 (Figure 2Q), and 1.9% of CGC⁺ CD8⁺ T cells were NLV-tetramer specific (Figure 2R). Two additional HLA-A*02:01 PWH were also evaluated: participant #3 with 2.1% NLV tetramer⁺ cells as a proportion of total CD3⁺ T cells (Supplemental Figures 3A-F) and participant #4 with 0.8% NLV tetramer⁺ cells as a proportion of total CD3⁺ T cells (Supplemental Figures 3G-L). In summary, despite heterogeneity in immune cell markers, NLV-tetramer⁺ CD8⁺ T cells are largely present within the CGC⁺ cluster.

Like CGC⁺ CD8⁺ T cells, we also characterized markers expressed on CGC⁺ CD4⁺ T cells. The two-dimensional plots show CX3CR1 and GPR56 expression on the y and x-axis and highlight CD57 expression on the z-channel (Figure 3A). The CMV-negative donor had very few CGC⁺ CD4⁺ T cells, unlike CGC⁺ CD8⁺ T cells. The UMAP shows the separation of the CGC⁺ CD4⁺ T cell cluster from the rest of the CD4⁺ T cells. In addition, CGC⁺ CD4⁺ T cells also express KLRG1, which may be less variable than CD57 (Figure 3B). With additional markers, we can define CGC+ CD4+ T cells as TEM (CD45RO⁺ CCR7^-) and TEMRA (CD45RO^- CCR7^-) cells that are CD28^+/-, CD27^-, PD1^+/- and CD38^+/-(Supplemental Figure 5). We used MHC Class II tetramers to identify CD4⁺ T cells recognizing two immunodominant CMV epitopes: DYS and LLQTGIHVRVSQPSL (LLQ, HLA-DQ06:02, pp65 protein) as previously published (10, 13). Participants with HLA-DR7 and HLA-DQ6 were selected (Supplemental Table 2). 12.1% of the total CD4⁺ T cells in participant #2 were DYS tetramer⁺ (Figure 4A). Two-dimensional flow cytometry plots show that DYS-tetramer⁺ CD4⁺ T cells also express CX3CR1 and GPR56 with variable CD57 (Figure 4B). 88.7% of the DYS-tetramer⁺ CD4⁺ T cells were TEM, and the rest were TEMRA (Figure 4C). Visualization using the UMAP technique showed that the majority of the DYS tetramer⁺ cells were within the CGC⁺ CD4⁺ cluster (Figures 4D, E). Analysis of the DYS tetramer on the CGC⁺ CD4⁺ cluster showed that 42.0% of CGC⁺ CD4⁺ T cells in participant #1 had TCRs that recognized the DYS epitope (Figure 4F). In participant #5 (Supplemental Table 2, HLA-DQ6+) LLQ-specific CD4⁺ T cells comprised 0.28% of CD4⁺ T cells (Figure 4G). LLQ tetramer⁺ T cells co-expressed CX3CR1, and GPR56 with variable CD57 expression (Figure 4H). 93.6% tetramer⁺ cells fell within the CGC⁺ cluster (Figures 4I-K). 1.3% of CGC⁺CD4⁺ T cells in participant #5 were specific for the LLQ epitope (Figure 4L). In summary, the majority of the CD4⁺ T cells with TCRs that recognize two different immunodominant CMV epitopes are CGC⁺. CMV-specific CD4⁺ T cells in PWH are significantly inflated compared to matched HIV-negative controls (13). These data suggest that in PWH without evidence of acute CMV infection at the time of the study, a large proportion of CGC⁺ T cells may be CMV-specific.

To characterize CMV-specific T cells with TCRs to less immunodominant epitopes with a lower frequency of tetramer⁺ T cells in PBMCs, we flow-sorted and expanded CGC⁺ and non-CGC⁺ T cells as depicted in the schematic (Figure 5A). We made attempts to expand CGC⁺ CD8⁺ T cells but could not analyze this population due to a high proportion of cell death. (Photos of expanded cell subsets on day 10 of culture are shown in Figure 5B). The morphology of the expanded CGC⁺ CD4⁺ T cells in some participants was distinct from the non-CGC⁺CD4⁺ and CD8⁺ T cells, with satellite clusters that we speculate may represent clonal expansion. Control MHC class II tetramers (HLA-DR3 and HLA-DQ6) with the CLIP peptides were also used to stain CGC⁺ CD4⁺ T cells from expanded cell lines (Figure 5C). We used the CMV tetramers to identify CD4⁺ T cells with TCRs to less dominant epitopes (TRRGRVKIDEVSRMF (TRR, HLA-DRB1*03:01, IE2 protein) and VKQIKVRVDMVRHRI (VKQ, HLA-DRB1*03:01, IE1 protein)). We measured tetramer⁺ CD4⁺ T cells (DYS, TRR, and VKQ) after a 10-day expansion of sorted CGC⁺ CD4⁺ T cells and compared them to unsorted PBMCs (Figure 5D). For some of the less dominant epitopes, we had improved detection after expansion in culture (TRR and VKQ). Sorted non-CGC⁺ CD4⁺ T cells expanded in culture for 10 days also had some CMV-specific CD4⁺ T cells by tetramer analysis (Figure 5E). Expanded CGC⁺ CD4⁺ cells maintained CX3CR1, GPR56, and CD57 expression on the tetramer⁺ cells (Supplemental Figures 5A, C, E). The tetramer⁺ CD4⁺ T cells that we detected in the expanded T cells from the non-CGC⁺ CD4⁺ sort also expressed CX3CR1 and GPR56 compared to the other cells in the same pool. This may suggest that some CGC⁺ T cells were among the non-CGC⁺ T cells obtained by sorting before expansion, or that CGC⁺ CD4⁺ T cells may be derived from non-CGC⁺ T cells (Supplemental Figures 5B, D, F). In general, CGC⁺ CD4⁺ T cells maintained higher levels of CX3CR1, GPR56, and CD57 in culture (Supplemental Figures 5G, H). Notably, there was no significant difference in the mean fluorescence intensity of the CX3CR1 and GPR56 between the tetramer⁺ cells in the CGC⁺ CD4⁺ population versus tetramer⁺ cells in the non-CGC⁺ CD4⁺ expanded T cells, while CD57 expression trended towards being higher in the expanded CGC⁺ CD4⁺ T cells (Supplemental Figure 5I). Taken together, CGC⁺ CD4⁺ T cells as we have defined them can proliferate in culture after CD3/CD28 stimulation with IL-2 supplementation. This is different from previous studies that failed to show the proliferation of CD57⁺ CD4⁺ T cells after stimulation with PHA (42) or HIV antigens (43). In our studies, the CGC⁺ CD4⁺ T cell cluster appears to be driven by CX3CR1 and GPR56, which includes CD4⁺ T cells with variable CD57⁺ T cell expression that may undergo several rounds of replication. In summary, CMV-tetramer⁺ CD4⁺ T cells in PWH are mainly CX3CR1⁺ GPR56⁺ with variable expression of CD57, while similar cells present in the non-CGC⁺ cells have low expression of CD57.

While we observed that CGC⁺ CD4⁺ T cells can recognize several CMV epitopes, it remains unclear the extent to which the CGC⁺ T cell cluster of cells is CMV-specific as a whole. To understand whether starting with CGC⁺ CD4⁺ T cells can help define CMV-specific T cells agnostic to HLA typing, we sorted single CGC⁺ CD4⁺ T cells into 96-well plates from two participants (#4 and #6, both HLA-DR7) as shown (Figure 6A). Two-dimensional plots show that 5.5% of CD4⁺ T cells in participant #4 are DYS tetramer⁺ (Figures 6B, C) and largely TEM cells (Figure 6D). UMAPs show DYS tetramer⁺ T cells within the CGC⁺ T cell cluster (Figures 6E, F). Paired αβ TCR sequences from the sorted CGC⁺ CD4⁺ T cells are shown in Circos plots (Figure 6G). Clonal TCRs that did not have paired αβ pairs are not shown on the Circos plots. Out of a total of 41 TCRs with paired αβ pairs, 26.8% had the CDR3 (CASSGGTGGGADTQYF). Other clonal TCRs are shown. Participant #6 was selected because of known DYS-specific CD4⁺ TCRs identified by bulk sequencing apriori (13). We also sorted CGC⁺CD4⁺ T cells from participant #6 (Figures 6H-M). Out of a total of 70 TCR sequences with matched αβ TCR pairs, 5 of the top 10 clones among the CGC⁺ CD4⁺ T cells (n=21) had been previously identified as CMV-DYS specific by sequencing TCRs from DYS tetramer⁺ CD4⁺ T cells (13). This suggests CGC⁺ CD4⁺ T cells in PWH co-infected with CMV may be largely CMV-specific.

The frequency of peripheral blood CGC⁺ CD4⁺ T cells from participants in the full cohort of PWH was positively correlated with plasma anti-CMV IgG titers (ρ=0.21, p=0.03), while CGC⁺ CD8⁺ T cells were not significant (ρ=0.18, p=0.07) (Supplemental Figures 6A, B). Finally, CGC⁺ CD4⁺ T cells and CGC⁺ CD8⁺ T cells were strongly correlated with each other (Supplemental Figure 6C).

To investigate the extent to which CGC⁺ CD4⁺ and CGC⁺ CD8⁺ T cells differ in cardiometabolic disease conditions (diabetes, pre-diabetes, hypertension, coronary arterial calcium, nonalcoholic fatty liver disease, and pericardial fat volume), we measured the frequencies of CGC⁺ T cells in PBMCs as a proportion of total CD4⁺ and CD8⁺ T cells. We found similar proportions of CGC⁺ CD4⁺ and CGC⁺ CD8⁺ T cells in participants with and without hypertension (Figure 7A), while CGC⁺ CD4⁺ T cells were higher in participants with coronary arterial calcium (CAC) on CT imaging (p=0.0009), (Figure 7B), and non-alcoholic fatty liver disease (NAFLD, defined as absolute liver attenuation less than 58 Hounsfield units [HU] on CT imaging [p=0.04; Figure 7C]). Both CGC⁺ CD4⁺ and CGC⁺CD8⁺ T cells were significantly higher among participants with prediabetes and diabetes as compared to non-diabetics (Figure 7D). Notably, the frequency of CGC⁺ CD8⁺ T cells was positively correlated with pericardial fat volume as measured by CT imaging (p=0.02), which was mainly driven by diabetic participants (p=0.0008) (Figure 7E). A similar relationship was not observed for CGC⁺ CD4⁺ T cells and pericardial fat volume (p=0.3). Collectively, these results indicate that there are more circulating CGC⁺ CD4⁺ and CGC⁺ CD8⁺ T cells in persons with HIV with subclinical atherosclerosis, NAFLD, pre-diabetes, and diabetes.

Prior murine studies showed the adoptive transfer of peripheral senescent CD8⁺ T cells (CD8⁺ CD44⁺ CD153⁺) from spleens of mice on a high-fat chow diet to mice on a normal chow diet was followed by insulin resistance in the recipient animals, suggesting a role for circulating T cells in the development of metabolic dysregulation (44). This concept is further supported by epidemiologic analyses in PWH showing an association between a higher circulating CD4⁺ TEMRA cell frequency and the subsequent development of diabetes mellitus (45). In light of these findings, we assessed the relationship between peripheral CGC⁺ CD4⁺ T cells (as a proportion of total CD4⁺ T cells) and CGC⁺ CD8⁺ T cells (as a proportion of total CD8⁺ T cells) with several cardiometabolic disease risk factors. CGC⁺ CD4⁺ T cells were correlated with waist circumference, and there was a trend towards a correlation with age and BMI (Figures 8A–C). These cells were most strongly correlated with fasting blood glucose in non-diabetic and pre-diabetic individuals (p=0.002) (Figure 8D). The correlation with fasting blood glucose was stratified by the metabolic group because of the effect of diabetes medications on glucose levels (all diabetic participants were receiving anti-diabetes medications and had a fasting blood glucose range of 73-416 mg/dL). We did not observe a significant relationship between CGC⁺ CD4⁺ T cells and fasting blood glucose in diabetic PWH (=-0.13, p=0.47). T cells can mediate inflammatory and anti-inflammatory effects through plasma cytokines. Therefore, we also assessed whether CGC⁺ T cells were related to inflammatory markers and anti-inflammatory cytokines important in cardiometabolic disease (35). The frequency of CGC⁺ CD4⁺ T cells correlated with hsCRP (p=0.01), IL-4 (p=0.04), IL-10 (p=0.04) but not IL-6 (p=0.5) (Figures 8E-H). On the other hand, CGC⁺ CD8⁺ T cells were not correlated with age, BMI, or waist circumference, and had a modest correlation with fasting blood glucose in non-diabetic/pre-diabetic PWH (r=0.27, p=0.009), and a trend toward significance in diabetic PWH (r=0.33, p=0.05) (Figures 8I-L) CGC⁺ CD8⁺ T cells were also correlated with hsCRP (ρ=0.30, p=0.005) (Figure 8M) but not IL-4, IL-10, or IL-6 (Figures 8N-P). Both CGC⁺ CD4⁺ T cells (ρ=0.23, p=0.01) and CGC⁺ CD8⁺ T cells (ρ=0.27, p=0.003) were correlated with hemoglobin A1C in a partial Spearman’s correlation analysis adjusted for age, sex, body mass index (BMI), hypertension, statin use, and smoking status (Figure 9, top). CGC⁺ CD4⁺ T cells had a stronger association with hemoglobin A1C when BMI was removed from the model (ρ=0.27, p=0.004), while the relationship between CGC⁺ CD8⁺ T cells and hemoglobin A1C attenuated (ρ=0.24, p=0.01) (Figure 9, bottom).

Both fasting blood glucose and HbA1c can be influenced by multiple factors, including the time of day, hydration status, medications, red blood cell survival, genetics, and vitamin deficiencies, among others. Plasma metabolites offer an additional profile of metabolic status, and changes in metabolites are frequently present before the development of overt disease (46). Therefore, we next assessed whether the observed association between CGC⁺ T cells and blood glucose measurements was also present for other plasma metabolites. We found that frequencies of CGC⁺ CD4⁺ T cells and CGC⁺ CD8⁺ T cells were positively correlated with plasma levels of branch chain amino acids (isoleucine and norleucine), carbohydrate metabolites (glucose/fructose, glucosamine, galactosamine, and fumarate), acetoacetate, and 2-beta-hydroxybutyric acid among others (Figure 10A). On the other hand, phosphocholine, L-Anserine, 3-Methylhistamine, Sulfino-L-Alanine, and Hydroxy-L-Tryptophan were inversely correlated with the frequencies of CGC⁺ CD4⁺ and CGC⁺ CD8⁺ T cells. Most of these correlations were not present for total memory CD4⁺ or CD8⁺ T cells (Supplemental Figures 7A, B). The frequency of CGC⁺ CD4⁺ cells positively correlated with metabolites that enriched for the starch and sucrose metabolism pathways (FDR 0.02) (Figure 10B), and negatively with other metabolites from the taurine and hypotaurine metabolism pathways (FDR 0.09, data not shown). Metabolites that were positively correlated with CGC⁺ CD8⁺ cells enriched for the pentose and glucuronate conversions, starch, and sucrose metabolism pathways but were not statistically significant (FDR 0.16) (Figure 10C). Similarly, metabolites that were negatively correlated with CGC⁺ CD8⁺ enriched for taurine and hypotaurine metabolism pathways but they were not significant (FDR 0.33, data not shown). Taken together, these results demonstrate that CGC⁺ T cells are related to starch and sucrose metabolites in plasma, as well as branched-chain amino acids (BCAA) which have been linked with the development of diabetes and cardiovascular disease (46–48).

CD4⁺ TEMRA cells are senescent and bioenergetically flexible compared to CD8⁺ TEMRA cells, partly due to their ability to effectively engage both glucose metabolism and oxidative phosphorylation (49). One possible explanation is that CD8⁺ TEMRA cells may have dysfunctional mitochondria (49). Given the observed relationship between the frequency of CGC⁺ T cells with starch and sucrose metabolites, we studied their metabolic profile as a way to understand their functional capacity in PWH. We analyzed the mitochondrial and glucose dependence of T cells using SCENITH, Single Cell ENergetIc metabolism by profiIing Translation in inHibition. Puromycin uptake by CGC⁺ CD4⁺ and CGC⁺ CD8⁺ T cells after incubation with inhibitors was measured by geometric mean fluorescence (Figures 11A, C). In general, naïve cells (CD4⁺ and CD8⁺) had lower levels of puromycin uptake compared to T cell memory subsets (Figures 11A, C). We found that unstimulated CD4⁺ memory (TCM, TEM, TEMRA, and CGC⁺) T cell subsets were predominantly mitochondrial-dependent (Figure 11B). CD4⁺ naïve T cells had higher mitochondrial dependence compared to CD4⁺ TEM (p=0.006). Although not significant, CGC⁺ CD4⁺ T cells had lower mitochondrial dependence and higher glycolytic capacity when compared to CD4⁺ naïve T cells (p=0.05) (Figure 11B). CD8⁺ T cell memory subsets also had higher mitochondrial dependence. However, CD8⁺ TEM (p=0.02) and TEMRA (p=0.01) cells showed significantly higher glycolytic dependence than central memory CD8⁺ T cells (TCM) (Figure 11D). Unstimulated CGC⁺ CD8⁺ T cells also had higher glucose dependence than CD8⁺ TCM but this was not significant (p=0.07) (Figure 11D). CGC⁺ CD4⁺ T cells that we expanded in culture using CD3/CD28/CD49d did not utilize glycolysis as effectively as non-CGC⁺ T cells (Supplemental Figures 8A, B). Notably, expanded CGC⁺ CD8⁺ T cells utilized glycolysis more than CGC⁺ CD4⁺ T cells (Supplemental Figure 8C). Taken together, since T cells are known to be bioenergetically flexible and can undergo metabolic reprogramming to utilize the more abundant nutrition sources in the local environment, it is possible that CGC⁺ T cells can meet their bioenergetic demands using abundant nutrient sources other than glucose.

Transcriptionally, unstimulated CGC⁺ CD4⁺ T cells were enriched for cytotoxic RNA transcripts (GNLY, CD244, GZMH, CTLA4) and were deficient in transcripts that form the aerobic/mitochondrial electron transport chain [SDHC (complex II), UQCR10 (complex III), COX5B (complex IV)] when compared to non-CGC⁺ CD4⁺ memory T cells (Figure 11E and Supplemental Tables 4, 5). CGC⁺ CD8⁺ T cells, on the other hand, displayed higher CPT1A (fatty acid β oxidation), GRIN1 (glutamate receptor), and lower COX4I1 and COX7C transcripts than non-CGC⁺ CD8⁺ T cells (Figure 11F and Supplemental Tables 6, 7). Some studies have suggested that long-chain fatty acid oxidation modulated by CPT1A is important for CD8⁺ T cell memory development, while knockout studies in mice suggest that CPT1A is dispensable (50).

We performed a direct comparison of CGC⁺ CD4⁺ and CGC⁺ CD8⁺ T cell transcriptomes to try and explain why CGC⁺ CD4⁺ T cells appeared to have a stronger relationship with cardiometabolic disease conditions and risk factors. We found that CGC⁺ CD8⁺ T cells were enriched for genes involved in mitochondrial ATP synthesis coupled electron transport [SDHC, UQCR10, COX5B] and fatty-acyl-CoA biosynthetic process [TECR] (Supplemental Figures 8A-C and Supplemental Table 6). Notably, three genes, SDHC, UQCR10, and COX5B, were consistently lower in CGC⁺ CD4⁺ T cells as compared to other memory CD4⁺ T cells and CGC⁺ CD8⁺ T cells. While CGC⁺ CD4⁺ T cells had higher expression of several genes including SLC16A1, BCL2, BIRC3, ICOS, CTLA4, and IDO1 which enriched several pathways including the pyruvate metabolic process. Taken together, the RNA transcriptomes of both CGC⁺ CD4⁺ and CGC⁺ CD8⁺ T cells show lower expression of transcripts that encode for mitochondrial complexes than non-CGC⁺ T cells. However, a direct comparison between CGC⁺ CD8⁺ and CGC⁺ CD4⁺ T cells suggests differences in bioenergetic sustenance. Notably, CGC⁺ T cells are largely TEM and TEMRA, compared to other memory cells which would include cells that are TCM, TEM, and TEMRA. Differences that may be driven by TCM in the non-CGC⁺ memory T cells have not been accounted for in this analysis.

Although CGC⁺ CD4⁺ and CGC⁺ CD8⁺ T cells were modestly correlated with fasting blood glucose, we did not find significant differences in glucose dependence compared to other memory T cells. Transcriptional analysis suggested that CGC⁺ CD8⁺ T cells may rely on long-chain fatty acid oxidation mediated by CPT1A. A previous study showed that PD-1 ligation of T cells induced CPT1A expression leading to an increased rate of fatty acid β oxidation (FAO) while limiting glycolysis (51). Some CGC⁺ CD4⁺ T cells and CGC⁺ CD8⁺ T cells express PD-1 (Supplemental Figures 2, 4), and our differential gene expression analysis suggests that CGC⁺ CD8⁺ T cells may use FAO as an energy source. We leveraged a separate cohort of HIV-negative donors that had undergone single-cell metabolic profiling of CD4⁺ and CD8⁺ T cells by mass cytometry (29). In this prior cohort, study, we defined CGC⁺ T cells based on the expression of killer cell lectin-like receptor G1 (KLRG1), and lack of expression of CD27 and CD28 (Figure 12A). Select previously validated metabolic antibodies (29) were used to characterize metabolic proteins/enzymes in CGC⁺ CD4⁺ and CGC⁺ CD8⁺ T cells. These included metabolic proteins involved in fatty acid metabolism (CPT1A), the tricarboxylic acid cycle and the electron transfer chain (ATP synthase (ATP5A)), mitochondrial expression (voltage-dependent ion channel 1 (VDAC1)), glycolysis and fermentation (hexokinase 2 (HK2)), signaling and transcription (ribosomal protein S6, pS6)) and amino acid metabolism (CD98) (Figure 12B). Of all the metabolic proteins tested, CPT1A expression was higher on CGC⁺ CD4⁺ T cells than all other CD4 subsets (Figure 12C), and when compared to all other non-CGC⁺ CD4⁺ T cells, both CPT1A and CD98 were higher while pS6 was lower (Figure 12D). CGC⁺ CD8⁺ T cells also had higher CPT1A than other CD8⁺ T cell subsets, while pS6 was significantly lower (Figures 12E, F). Taken together, higher levels of CPT1A in both CGC⁺ CD4⁺ and CGC⁺ CD8⁺ T cells implicate FAO as a potential source of energy in CGC⁺ T cells. Lower pS6 compared to other T cells may indicate that CGC⁺ cells have a lower basal level of translation than other memory subsets.