To assess whether HIV differentially depletes different subsets of tissue CD4 T cells, we measured HIV-associated cell depletion in several discrete T cell subsets using flow cytometry first ( Table S1 ). Fresh HLAC were prepared as previously described (22), either mock infected or infected with an X4-tropic HIV.GFP reporter virus, and cultured for 6 days before analysis ( Figure 1A ). As previously reported, we observed a marked loss of CD4 T cells (defined as CD3+CD8-) in the infected culture (frequency of 6.89%) as compared to the uninfected control (frequency of 32.1%) ( Figure 1B ). In contrast and as expected, CD8 T cell numbers (defined as CD3+CD8+) did not decrease in infected cultures. Normalization of the leftover live CD4 T cell counts in the infected culture to the CD8 T cell counts (details described in Materials and Methods) confirmed significant depletion of the CD4 T cells ( Figure 1C ). Interestingly, HIV-associated CD4 T cell depletion was more pronounced in the CD4 T memory (Tm) relative to in the CD4 T naïve (Tn) cells. Furthermore, within the memory compartment, T follicular helper (Tfh) cells were preferentially killed over either effector memory (Tem) or central memory (Tcm) cells ( Figures 1C ; S1 ). These FACS results suggest that while HIV infection is associated with T cell depletion by HIV in multiple CD4 T subsets, and that the levels of depletion differ within these subsets.

To study HIV-associated depletion in greater depth, and to take advantage of the PP-SLIDE bioinformatics approach that corrects for virus-induced remodeling of cellular phenotypes, we designed a new 38-parameter CyTOF panel able to distinguish a wide range of T cell phenotypic subsets ( Table S2 ). Even if some parameters are altered by infection, the high number of CyTOF parameters allows efficient backtracking to the original cell population via PP-SLIDE. As recently described (34), PP-SLIDE uses the 40 CyTOF parameters to match each cell in the infected culture to its “k-nearest neighbor” (kNN) cell in the uninfected control population ( Figure 2A , and Materials and Methods). HIV infection is known to remodel productively-infected cells (“infected”), but non-productively-infected (“bystander”) cells could also be remodeled by the inflammatory environment of infected cells. To examine killing, we infected HLAC cells with HIV-GFP for 6 days followed by CyTOF and PP-SLIDE analysis. Using CD4 T cell markers and HIV reporter (GFP) expression, we selected infected (GFP+) and bystander (GFP-) CD4 T cells in the infected HLAC population, as well as the CD4 T cells present in the uninfected control ( Figures 2B ; S3 ). We then visualized the phenotypes of CD4 T cells using t-distributed stochastic neighbor embedding (tSNE), a dimension reduction visualization method (37) ( Figure 2C ). Remodeling of both infected (pink dots) and bystander (purple dots) CD4 T cells from the infected culture was suggested by the fact that many of these cells did not map to regions of the tSNE plot occupied by CD4 T cells from the uninfected sample (gray dots) ( Figure 2C , left panels). In contrast, after application of PP-SLIDE, their “nearest neighbors” (“kNN infected” and “kNN bystander”) localized within the regions occupied by the uninfected CD4 T cells in the tSNE plot ( Figure 2C , right panels, aqua, blue and gray dots). As the kNN infected and kNN bystander cells harbor the predicted features of the original infected and bystander CD4 T cells, prior to remodeling, for the remainder of the study we simply refer to these cells as “infected” and “bystander” cells.

Although our flow cytometry data suggested that HIV-associated killing occurred preferentially among memory CD4 T cells (Tm cells), this might have been an artifact due to the remodeling of the surviving bystander cells into naïve-like CD4 T cells (Tn cells). To address this possibility, we classified the HLAC cells into the main immune subsets (B, CD8 T, Tn, and Tm cells) by applying the FlowSOM clustering approach (38) to the PP-SLIDE corrected CyTOF data ( Figures 3A ; S4 ). We found that both Tm and Tn cells were killed by HIV but at different levels. Compared to the uninfected culture, the infected culture had lost 84% of its Tm cells and 68% of its Tn cells ( Figure 3A , red and aqua). The higher loss of Tm over Tn cells was statistically significant (n=6 donors, Figure 3B ), confirming preferential depletion of memory over naïve CD4 T cells.

Preferential killing of the Tm subset could result from higher permissivity to productive infection leading to higher viral-induced cytotoxicity. To test if Tm cells were also preferentially infected by HIV, we measured the level of productive infection (% productively infected) in Tm and Tn cells ( Figure 3C ). We found a higher proportion of productively-infected cells among Tm than Tn cells; in fact, the majority of the Tn cells (> 95%) were resistant to HIV infection. Our observation that Tn cells are depleted but poorly infected suggests that many Tn cells die as bystander cells.

As Tm cells were preferentially killed compared to Tn cells, we focused the rest of our analysis on Tm cells. We characterized killing within different Tm subsets and investigated whether the higher HIV-associated killing of Tm subsets correlated with their permissivity to productive infection. First, the Tm cells were subjected to FlowSOM clustering ( Figure 4A ). Tm subsets that were preferentially killed (red) and -infected (aqua) were identified by their higher level of death or infection, respectively, as compared to those found in total Tm cells (details described in Materials and Methods, Figure S5 ). Interestingly, although there was overlap, cells preferentially killed resided in distinct areas of the tSNE relative to those preferentially infected, suggesting they represent different subsets ( Figure 4B , dotted circles). Next, we looked into each of the preferentially-killed Tm subsets (clusters 1, 2, and 3), and quantitatively assessed their levels of HIV-associated depletion and infection ( Figures 4C–E ). Among these three preferentially-killed clusters, cluster 1 was preferentially depleted but was highly resistant to productive infection, suggesting that these cells likely died as bystander cells ( Figures 4D, E ). In contrast, clusters 2 and 3 were preferentially depleted and were highly permissive to productive infection, suggesting death in these two clusters might be attributed at least in part to apoptosis triggered by a productive infection. In addition to clusters 2 and 3, cluster 4 was also preferentially infected ( Figures 4E, F, H ). However, unlike cluster 2 and 3, cluster 4 was not preferentially killed ( Figure 4G ). Taken together, these findings suggest that Tm subsets are differentially depleted by HIV. Among the preferentially-killed Tm subsets, some are resistant to productive infection (cluster 1) while others are highly permissive to productive infection (cluster 2, 3). Moreover, high permissivity to infection is not always associated with high killing of Tm cells, as exemplified by cluster 4 (Table S3).

To further characterize the phenotypic features of preferentially-killed and preferentially-infected Tm subsets, we assessed expression levels of various antigens within our CyTOF panel ( Figure 5 ). We found that all preferentially-killed subsets (clusters 1, 2, and 3) expressed high levels of the two chemokine receptors CXCR5 and CXCR4 ( Figure 5A ). While CXCR5 defines Tfh-like cells, CXCR4 is the co-receptor used by our reporter virus, suggesting that efficient entry of HIV into cells may underlie preferential HIV-associated cell death. In contrast, unique features of the preferentially-infected subsets (clusters 2, 3, and 4) included low expression levels of CCR7 and CD62L, markers of Tcm cells ( Figure 5B ). These results suggest that Tm cells with a Tem phenotype (CCR7-CD62L-) (33) are preferentially infected. The preferentially-infected clusters also expressed high levels of the exhaustion markers PD1 and CTLA4 ( Figure 5C ). As these antigens are also markers of activated cells (39, 40), we assessed the activation status of the preferentially-infected subsets. We observed that the early activation marker CD69 was high on all the preferentially-infected clusters, while CD25 and HLADR, which are upregulated at later stages of T cell activation, were preferentially expressed only in cluster 4 ( Figure 5D ).

Interestingly, cluster 4 was the only subset that was preferentially infected but spared from HIV-associated killing. This suggests that preferentially-infected cells resistant to HIV-induced killing exhibit the following features: 1) low expression of the CXCR4 co-receptor, 2) a Tem phenotype (CCR7-CD62L-), and 3) high expression of activation markers (PD1, CTLA4, CD69, CD25, and HLADR) ( Figures 5A–D ). In contrast, among the four analyzed Tm clusters, cluster 1 was the only one preferentially susceptible to HIV-associated killing but not productive HIV infection, suggesting its propensity to be killed by bystander mechanisms ( Figures 4D, E ). The phenotypes associated with cluster 1 were: 1) high CXCR4 expression, and 2) a Tcm phenotype (CCR7+CD62L+), 3) low expression of exhaustion markers (PD1-CTLA4-), and 4) low expression of activation markers (CD69-CD25-HLADR-) ( Figures 5A–D ).

Further mining of the phenotyping data revealed that various combinations of CD127, CD25, PD1, CD57, and CXCR5 could be used to uniquely define the four subsets ( Figure S7 ; Table S3 ). In particular, we found cluster 1, the only cluster preferentially killed by bystander mechanisms, could be defined as CD127+CD25- Tm cells. Cluster 4, which was highly permissive but not preferentially killed, could be defined as CD25+PD1+CD57- Tm cells. Clusters 2 and 3 were characterized by CD57+PD1+ Tm cells and CXCR5+PD1+CD25-CD57-CD127- Tm cells, respectively, and both of these subsets were both preferentially killed and infected. These identified markers enable characterization of preferentially killed/infected subsets without a need for 40 parameter clustering.

To refine our understanding of what distinguishes the different Tm cell subsets, we analyzed their transcriptional profiles. Taking advantage of our ability to identify preferentially killed/infected subsets using only a limited number of surface markers, we then implemented Antibody-Seq, which isolates cells with DNA oligo-barcoded antibodies directed as their surface antigens and subjects them to single-cell RNA-seq (41, 42). We focused this analysis on the two extreme categories: the preferentially-killed but not infected subset (cluster 1-like), which could be defined as CD127+CD25- Tm cells; and the preferentially-infected but not killed subset (cluster 4-like), which could be defined as CD25+PD1+CD57- Tm cells. Consistent with our CyTOF data ( Figure 5A ), the cluster 1-like cells expressed high levels of CXCR4 protein ( Figure 6A ). In contrast, the cluster 4-like cells expressed lower levels of CXCR4; this level is presumably sufficient to support productive infection but not viral-induced killing, as previously suggested (24). In addition, similar to our CyTOF result ( Figure 5D , right), we found cluster 1-like cells expressed lower levels of HLADR, while cluster 4-like cells expressed high levels ( Figure 6B , left). In addition, the mRNA levels of OX40, another activation marker, was expressed in a similar pattern of low expression in cluster 1-like cells and high expression in cluster 4-like cells ( Figure 6B , right).

We next delved into transcriptomic features pertinent to HIV permissivity and cell death. Cluster 1-like cells preferentially expressed genes for the HIV restriction factors SERINC5, SAMHD1, APOBEC3G, MX2, TRIM32, ISG15 (43–48), which may explain their low permissivity to productive infection ( Figure 6C ). In contrast, cluster 4-like cells expressed lower level of these restriction factor genes, which may explain why these cells allow the virus to complete its life cycle.

Interestingly, cluster 1-like cells highly expressed the genes for caspase 1, caspase 4, and gasdermin D, which are involved in inflammatory pyroptosis. The gene for caspase 3, classically associated with apoptosis but more recently also shown to be linked to pyroptotic death (49, 50), was also upregulated in these cells ( Figure 6D ). The preferential expression of these cell death-associated genes in cluster 1-like cells over cluster 4-like cells suggests that these cells are primed for pyroptotic death ( Table S3 ).

In summary, these transcriptomics data demonstrate that the killing of bystander Tm cells is associated with high expression of intracellular restriction factors and cell death factors. In contrast, cells expressing low levels of viral restriction factors and cell death factors were more highly permissive to HIV infection but relatively spared from virus-induced cell death.

Human tonsils were obtained from the Cooperative Human Tissue Network (CHTN) during routine tonsillectomies, mainly for sleep disorders. These tissues were processed as previously described (18). Briefly, HLAC single-cell suspensions were created by dissection and then by pressing the Tonsil tissue through a 40-µm mesh. Live lymphocytes were then isolated from the single cell suspensions by Ficoll density gradient centrifugation. HLAC cells were cultured in tonsil culture media that consisted of RPMI 1640 supplemented with 15% fetal bovine serum (FBS) (Corning), 100 mg/mL gentamicin (Gibco), 200 mg/mL ampicillin (Sigma-Aldrich), 1 mM sodium pyruvate (Sigma-Aldrich), 1% non-essential amino acids (Mediatech), 1% Glutamax (ThermoFisher), and 1% Fungizone (Invitrogen).

Details about the antibodies used in flow cytometry staining are listed in supplemental table 1 and for CyTOF in supplemental table 2. All CyTOF antibodies were purchased from Fluidigm. HLADR Qdot(112Cd) was purchased pre-conjugated. All other CyTOF antibodies were purchased unconjugated and then conjugated with elemental isotopes in house using MAXPAR^® X8 Ab Label Kits from Fluidigm.

The pNLENG1-IRES-GFP clone was derived from NL4-3 as previously described (59). The pro-viral expression vector DNA encoding the pNLENG1-IRES-GFP reporter virus was transfected into 293T cells using the Promega Fugene HD transfection reagent (catalog no. E2311) and cultured at 37°C. Media was replaced after 16 hours. Culture supernatants were collected 24 and 48 hours post-transfection. Virions were concentrated by ultracentrifugation at 20,000 rpm (32198 g) for 2 hours on a Beckman Coulter Optima XE-90. Gag-p24 levels of the HIV.GFP viral stocks were quantitated by enzyme-linked immunosorbent assay (ELISA) using the Lenti-X™ p24 Rapid Titer Kit from Clontech (catalog no. 632200).

One million HLAC cells were mixed with HIV-1.GFP (NLENG1-IRES-GFP) or mock control (culture media) in 100 µl total volume and cultured in V-bottom wells of a 96-well plate. To achieve different levels of T cell depletion, we applied 3 different HIV amounts (25, 50, and 100 ng p24 gag of HIV) and treated them as technical repeats in the current study. Cells were centrifuged at 25°C at 1200 x g for 2 hours and then cultured at 37°C as a pellet (23). Both infected and uninfected (mock) HLAC cells were collected and processed for FACS or CyTOF staining followed by calculation of % live and % infection as described in the following sections. A total of 6 donors were included in the current study and technical repeats from each donor with more than of 10% CD4 T cell depletion were analyzed.

Cells collected from HLAC cultures were processed for fluorescence-activated cell sorting (FACS) staining using a live-dead cell discriminator dye (Zombie Aqua) and stained with fluorescently labeled antibodies specific for the cell markers in supplemental table 1. Data were collected on a BD LSRFortessa™ X-20 Cell Analyzer flow cytometer and analyzed with FlowJo software from BD bioscience. At least 100,000 events per sample were collected, and at least 50,000 live cells were included in the subsequent data analysis. Gating strategies for each cell subset are shown in Supplementary Figure S1 (Cells, Singlet, Live, CD8 T, CD4 T, Tn, Tm, Tem, Tcm, and Tfh).

CyTOF staining was conducted as previously described (32). Briefly, cells collected from HLAC cultures were first stained for cell surface markers using antibodies conjugated with elemental isotopes listed in supplemental table 2. For live-dead staining, cells were treated with 139In-loaded-maleimide-DOTA (Macrocyclics) at 5 µg/ml on ice for 30 minutes, and cisplatin (Sigma-Aldrich) at 25 µM at room temperature for 60 seconds. Cells were then immediately fixed with 2% paraformaldehyde overnight at 4°C. Cells from multiple samples were barcoded using a Cell-ID™ 20-Plex Pd Barcoding Kit from Fluidigm, and combined before being permeabilized using the eBioscience™ Foxp3/Transcription Factor Staining Buffer Set. Cells were then stained for intracellular antigens including GFP, Gag, Foxp3, cleaved caspase-3, CTLA4, SAMHD1, and IFI16. CyTOF data were collected from a Fluidigm CyTOF2 Helios Mass Cytometer, normalized, and de-barcoded by CyTOF^® Software from Fluidigm. Expression data for all parameters in each cell were next analyzed in the form of Flow Cytometry Standard (FCS) files in FlowJo™ software from BD bioscience. Data were pre-gated on Cells, Intact cells, Singlets, and Live cells before further analysis ( Figure S2 ). A range of 10,000-130,000 cells (Live singlets post-CD8 normalization) per sample were included in the CyTOF analyses. tSNE and FlowSOM analysis were conducted in the Cytobank platform from Cytobank, Inc.

PP-SLIDE was previously introduced and validated as an approach to trace HIV-remodeled cells back to their original pre-infected state (31–34). The detailed description and R script were made available in our previous publication [Neidleman et al. eLife 2020 (32)]. In the current study, we implemented PP-SLIDE to predict the pre-infection cell marker expression of all cells in the HIV-exposed HLAC culture. As an example of the single-cell analysis using PP-SLIDE, for Infe-cell #1, we calculated the Euclidian distances between Infe-cell #1 and each cell from the uninfected culture. The cell in the uninfected culture found to have the shortest Euclidian distance with Infe-cell #1 is identified as the predicted precursor of Infe-cell #1 (kNN_{Infe-cell #1}). The cell marker expression of the kNN cells was used for phenotyping the infected and bystander cells in all our CyTOF data.

To evaluate the level of productive infection in specific cell subsets, we first gated on the subset of interest in FlowJo and then measured the cell number in this subset in uninfected and infected samples. The level of HIV-productive infection in a selected subset, for example, subset S1, is calculated by the equation below:

% infection (S1): percent of productively infected cells in subset S1

% infection (S1) = number of S1 cells in infected sample/number of S1 cells in uninfected sample x 100

Similarly, the level of productive HIV infection in the Tm population is calculated as: number of productively infected cells in the infected sample/number of Tm cells in the uninfected sample.

Tm subsets with a higher proportion of productively infected cells (% infection) than total Tm cells are referred to as “preferentially-infected subsets”.

To evaluate the level of HIV-associated killing in specific cell subsets, we first gated on the subset of interest in FlowJo and then measured the following cell numbers (for example, subset S1):

cNum_UnS1: Cell number of S1 subset from uninfected sample

cNum_InfeS1: Cell number of S1 subset from infected sample

cNum_UnCD8: Cell number of CD8 T cells from uninfected sample

cNum_InfeCD8: Cell number of CD8 T cells from infected sample

The level of HIV-associated killing is indicated by % killing and is calculated using the following equation:

% killing (S1): percent of HIV-associated killing of cells in subsets of interest S1 compared to uninfected sample and normalized with CD8 T cell number.

% killing (S1) = (cNum_InfeS1 ÷ cNum_UnS1) X (cNum_UnCD8 ÷ cNum_InfeCD8) X100

Tm subsets with a higher level of HIV-associated killing than that of total Tm cells are defined as “preferentially-killed subsets”.

T cells were isolated from HLAC using Immunomagnetic negative selection (StemCell Technologies), and dead cells were removed using a dead cells removal kit (Miltenyi Biotec). The cells were stained with a panel of TotalSeq-A human antibodies (BioLegend) according to manufacturer’s protocol. Stained cells were then loaded onto a Chromium Next GEM chip G and both ADT (Antibody Derived Tag) and GEX (Gene Expression) libraries were generated using Chromium Next GEM Single Cell 3’ Reagent Kits v3.1 (10X genomics) for next generation sequencing. A range of 11,000-13,000 events per sample were collected for the subsequent single-cell RNAseq analysis. The data was preprocessed in Cell Ranger for alignment against the human genome and further analyzed in SeqGeq ^®.