To define CD8 T-cell markers associated with immunopathogenesis in HIV-2-infected individuals, blood samples were obtained from study participants belonging to a well-defined, occupational cohort of police officers in Guinea-Bissau ( Table 1 ) (41, 42). In addition to HIV-2 infected individuals, individuals diagnosed with HIV-1 or HIV-1/HIV-2 dual (HIV-D) infections, all either treatment-naïve or not successfully treated [with VL > 1000 RNA copies/ml (43)], were included. For comparison, successfully treated HIV-1-infected participants and HIV-seronegative individuals within the same cohort were included. Informed consent was obtained from the participants and the National Ethical Committee, Ministry of Public Health in Guinea-Bissau, and the Ethical Committee at Lund University, approved the study.

All samples were collected during a 14-day time period on site in Guinea-Bissau and shipped to Sweden. Plasma was collected using EDTA vacutainer tubes (BD Biosciences) and whole blood using Cyto-Chex BCT tubes (Streck), in which cells are preserved while inadequate for functional analysis (44). HIV status was determined as previously described (45), and confirmed using the Geenius HIV-1/2 confirmatory assay (Bio-Rad). Absolute numbers of CD4 T-cells/µl and percentage of CD4 T-cells of lymphocytes (CD4%) were determined using the FACSPresto instrument (BD Biosciences). We used CD4% as an additional marker of immunodeficiency due to findings that CD4% is a more stable disease marker than absolute CD4 T-cell count in settings with elevated pathogenic burden and comorbidities (17, 23, 46), the close correlation to markers of T-cell exhaustion in HIV-1 infection (47), and on the previous use of this marker in the studied cohort (48).

With minor modifications, HIV-1 and HIV-2 VL were determined by in-house quantitative PCR (qPCR) protocols as described (49). Briefly, viral RNA was extracted using miRNeasy micro Kit (Qiagen) and TaqMan qRT-PCR was performed using the Superscript III Platinum One Step qRT-PCR kit (ThermoFisher Scientific). The VL quantification limit was 75 RNA copies/ml plasma for HIV-1 or HIV-2 mono-infected, and 135 RNA copies/ml plasma for HIV-D-infected.

Directly upon arrival from Guinea-Bissau all whole blood samples, collected in Cyto-Chex BCT tubes (Streck), were processed and analyzed on a BD Fortessa instrument (BD Biosciences) as previously described (22). Likewise, it was confirmed that the expression patterns of markers were equivalent between Cyto-Chex-stabilized blood and fresh blood samples (data not shown). In brief, the blood samples were incubated in Lysis buffer (BD Biosciences) at a ratio of 1:6 for 5 min before centrifugation and washing in PBS/FCS (2%). After treatment with DNase (6 U/ml), the cells were extracellularly stained in PBS/EDTA (2mM) for 10 min at 37°C and further 20 min at room temperature. Permeabilization was performed with the FoxP3 kit (eBiosciences) prior to 60 min intracellular staining. Antibodies used for staining of extracellular (2B4, CCR7, CD3, CD4, CD8, CD14, CD19, CD38, CD45RO, CD226, CXCR5, HLA-DR, PD-1 and TIGIT) and intracellular (Eomes) antigens, clones, fluorophores and suppliers are listed in Supplementary Table S1 . The cells were resuspended in Cytofix Buffer (BD Biosciences) and analyzed on a BD Fortessa.

Plasma concentrations of interferon gamma-induced protein 10 (IP-10, also known as CXCL10), soluble CD14 (sCD14), and β2-microglobulin (B2M), which are plasma markers previously linked to chronic immune activation and inflammation, viremia and disease progression in HIV-1 and HIV-2 infections (50–52), were analyzed in plasma of HIV-2-infected individuals using the Magnetic Luminex assay (R&D Systems Inc.) on the Bio-Plex 200 platform (Bio-Rad Laboratories Inc.) according to manufacturer’s instructions. Plasma was diluted 1:2, 1:600 and 1:4000 for quantification of IP-10, sCD14 and B2M, respectively.

The compensated flow cytometry dataset was minimized by isolating CD8 T-cells (CD3+CD8+) out of single CD14-CD19- lymphocytes using the FlowJo 2 LLC software (version 10.5.3). For FlowSOM analysis, the dataset was limited to HIV-infected individuals. One HIV-1 and two HIV-2 aviremic individuals were subsequently excluded due to technical issues and quality of flow cytometry data. CD8 T-cell populations from the remaining individuals were then clustered using the FlowSOM algorithm (53) in R (54). Data for all cells (n = 3,232,181) across all samples (n = 45) were loaded, logicle transformed and scaled using the FlowSOM() function. A 10x10 grid was used for clustering based on the expression of ten markers: CD45RO, CCR7, HLA-DR, CD38, 2B4, PD-1, TIGIT, Eomes, CD226 and CXCR5. We set rlen = 100 and compensate = FALSE, as compensation had already been applied. We finally used maxMeta = 45, such that the final number of meta-clusters was determined by the FlowSOM() function itself.

This was followed by statistical analyses of differences between groups using a negative binomial generalized linear model from the edgeR package (55). Group-wise comparisons were individually FDR-controlled using the Benjamini-Hochberg method. Heatmaps showing cluster distribution per sample and median expression of the clustering markers were generated using ComplexHeatmap (56). The UMAP was based on scaled expression values of the ten clustering markers and generated using the uwot package (57). It is based on a sampling of 50,000 cells per HIV status category, with approximately equal sampling across samples within each of the categories. The ggplot2 package (58) was used for both the UMAP and density plots, which depict the expression distributions for each clustering marker across all cells in a given cluster.

For subsequent analysis using GraphPad PRISM, we used a manually gated dataset including data from all study participants. Here, memory CD8 T-cells were identified from CD3+CD8+, single CD14-CD19- lymphocytes, followed by exclusion of naïve CCR7+CD45RO- T cells. The frequencies of the analyzed cell populations obtained by Boolean gating were guided by “fluorescence minus one” (FMO) controls in FlowJo. Differences between study participants were analyzed using Mann-Whitney U test, comparing two groups, and Kruskal Wallis test with Dunn´s post-test, comparing more than two groups. Spearman rank correlation analysis was used to test correlations between CD8 T-cell populations and CD4 levels, VL and plasma inflammation markers. Manually gated co-expression patterns were analyzed using permutation tests provided in SPICE (version 6.0) (59).

To characterize CD8 T-cell phenotypes in HIV-2-infected individuals, participants belonging to an occupational cohort in Guinea-Bissau (41, 42) were enrolled ( Table 1 ). The HIV-2-infected participants (n=23), all either treatment naive or unsuccessfully treated, were further subdivided into viremic (n= 9) and aviremic (n=14) individuals, based on the plasma VL quantification level of 75 RNA copies/ml. In addition, HIV-1 (n=12) or HIV-D-infected (n=5), either treatment naive or unsuccessfully treated, as well as successfully treated HIV-1-infected (n=8) with VL < 1000 RNA copies/ml, and HIV seronegative (n=27) individuals belonging to the same cohort were studied. Both median CD4% and absolute CD4 T-cell counts were found to be higher in the seronegative individuals as compared to the viremic HIV-2 individuals (p<0.001), and in the aviremic HIV-2 compared to the viremic HIV-2 group (p=0.007 and p=0.024, respectively), whereas no statistical difference were observed between the HIV-2 aviremic and the seronegative participants ( Figures S1A, B ). A statistical difference in CD4 T-cell levels, with lower median CD4% and absolute CD4 T-cell count, was however observed between the treated HIV-1 group as compared to the seronegative participants (p=0.019 and p=0.031, respectively; Figures S1C, D ). Additional characteristics of the study participants are depicted in Table 1 .

It has previously been demonstrated that HIV-1 specific CD8 T-cells display an exhausted phenotype that is only partially restored by ART (29, 30). We have recently linked this phenotype to the TIGIT/CD226 axis in HIV-infection (30). To which extent the TIGIT/CD226 axis is affected by HIV-2 infection, particularly in relation to viremia has not been investigated.

First, unsupervised clustering analysis using FlowSOM was used to enable visualization of CD8 T-cell population expressing any combination of the assessed markers that delineated the HIV groups. The analysis revealed nine distinct CD8 T-cell clusters ( Figures 1A, B ). One of these clusters, cluster 5, showed statistical differences between aviremic and viremic HIV-2-infected study participants ( Figure 1A ). Cluster 5 consisted of late differentiated (CD45RO+, CCR7int/-) CD8 T-cells expressing a combination of activation (CD38+, HLA-DRint/high), inhibitory receptors (2B4+PD-1+TIGIT+), as well as Eomes ( Figure 1A and Figure S2 ). Moreover, this cluster comprised a mixed population of CD226+ and CD226- cells ( Figure S2 ). In HIV-1 infection, the combined expression of CD38, HLA-DR, and PD-1 has been shown to delineate T-cell pathogenesis distinguishing infected from healthy individuals (47). Similarly, although expression of Eomes is associated with memory formation, the severity of CD8 T-cell exhaustion has been linked to the expression of Eomes (60). In chronic viral infections, high expression of Eomes is a hallmark of T-cell exhaustion (29, 60, 61).

In HIV-2 infection, proportions of cluster 5 not only differentiated the viremic and aviremic individuals (p= 0.034; Figure 2A ) but correlated also with disease progression in the form of declining CD4% (r = -0.603, p = 0.004; Figure 2B ) and CD4 count (r = -0.580, p = 0.006; Figure S3 ), as well as elevated VL (r = 0.512, p = 0.018; Figure 2C ). Moreover, increased frequencies of cluster 5 correlated with elevated plasma levels of inflammation markers, including B2M (r = 0.719, p<0.001), sCD14 (r = 0.775, p<0.001) and IP-10 (r = 0.806, p<0.001) ( Figures 2D–F ). Thus, our results indicate that cluster 5 represents CD8 T-cells with an exhausted phenotype that is clearly linked to immunopathogenesis in HIV-2 infection.

Furthermore, in treatment naïve or not successfully treated individuals, cluster 5 was more frequent in the participants with HIV-1, compared to HIV-2, infection (p = 0.027), while the frequency of these cells in HIV-D-infected participants was not different, when compared to the other groups ( Figure 2A ). Moreover, along level of viremia, lower frequencies of cluster 5 were observed in successfully treated HIV-1 infection as compared to HIV-1-infected not receiving such treatment (p = 0.003) ( Figure 2A ).

Of note, both cluster 1 and cluster 4 differentiated the aviremic and viremic HIV-2-infected groups in direct comparisons (p = 0.049 and p= 0.049 respectively) but did not differ significantly in the multiple FlowSOM comparison including all HIV-infected groups (data not shown). Cluster 1 represents a population negative for the investigated T-cell activation and immune checkpoint markers, while cluster 4, like cluster 5, consists of a population of cells expressing activation and checkpoint markers ( Figure 1A and Figure S2 ).

In summary, unsupervised multidimensional clustering suggests that late differentiated and exhausted CD8 T-cells differentiate viremic and aviremic HIV-2 infections and are elevated with CD4 T-cell depletion and systemic inflammation.

To evaluate the impact of aviremic HIV-2 infection on T-cell pathogenesis, we next compared the HIV-2 groups in relation to the seronegative individuals. Guided by the results obtained by the unsupervised clustering analysis and previous knowledge in the HIV-field linking the combined expression of activation and inhibitory receptors to T-cell exhaustion, senescence and health (22, 30, 47), we investigated CD8 T-cell populations expressing different combinations of CD38, HLA-DR, 2B4, PD-1, and TIGIT. According to the permutation analysis statistically significant differences were found between the seronegative control group and both the aviremic and viremic HIV-2-infected group (p = 0.050 and p < 0.001) ( Figure 3A ). More in depth analyses revealed that differences between the HIV-2-infected groups and seronegative individuals could be traced back to four different population. First, a cell population positive for all five analyzed markers (CD38+HLA-DR+2B4+PD-1+TIGIT+) (p=0.042; Figure 3B ), best resembling the expression pattern of cluster 5 and separating both the aviremic and viremic HIV-2 groups from the seronegative participants. Similarly, the cell population solely lacking CD38, also separated the aviremic HIV-2 from the seronegative individuals (p=0.031; Figure 3C ), but not the seronegative from the viremic HIV-2. Based on the co-expression of PD-1 and TIGIT that is linked to increased inhibitory function (62, 63), and impaired HIV-1 specific T-cell responses (30), these populations resemble exhausted T-cells. The third population, a cell population lacking PD-1 (CD38+HLA-DR+2B4+PD-1-TIGIT+) was pronounced in the group of viremic HIV-2 infected participants (p<0.001; Figure 3D ) but did not distinguish the aviremic HIV-2 from the seronegative. In chronic viral infection, expression levels of PD-1 can determine whether T-cell exhaustion is reversible, as is the case for cells with low, but not high levels of PD-1 (64, 65). In addition, expression of PD-1 together with CD38 and CD101 are known to be permanently dysfunctional populations (66), suggesting that the populations described may represent different stages of T-cell exhaustion (67, 68), needing further attention. Moreover, a cell population only expressing 2B4 (CD38-HLA-DR-2B4+PD-1-TIGIT-) was less frequent in viremic HIV-2-infected as compared to the seronegative individuals (p = 0.022; Figure 3E ). 2B4 can act both as a stimulatory and inhibitory receptor (69), and synergize with CD226 in a co-stimulatory capacity (70) and was further evaluated together with CD226, PD-1 and TIGIT.

When performing similar comparisons including the HIV-1 groups, the permutation analysis showed that frequencies of CD8 T-cells expressing activation and exhaustion markers differed comparing the seronegative group and the viremic HIV-1 group (p<0.001), while no such difference was seen between seronegative group and the treated HIV-1 group ( Figure S4A ). Detailed analyses further revealed that none of the activated and exhausted CD8 T-cell populations, found to be elevated in the treatment naïve aviremic HIV-2 group ( Figures 3B, C ), were increased in the successfully treated HIV-1-infected individuals compared to the seronegative group ( Figures S4B, C ).

Taken together, these findings suggest that frequency of activated and exhausted cells within the memory CD8 T-cell compartment are elevated in HIV-2-infected individuals, independent of viremia, while the level of these cells are normalized in persons with successfully treated HIV-1-infection.

CD226 is a marker conferring receptiveness for co-stimulation that has been shown to compete with the immune inhibitory receptor TIGIT for binding to the shared ligand (71). Thus, since cluster 5 included a mix of cells expressing, or not expressing, CD226 ( Figure 1 ), we next analyzed expression patterns of CD226 together with 2B4, PD-1 and TIGIT on memory CD8 T-cells. Significant differences between the seronegative control group and both the HIV-2 aviremic and viremic group were revealed by the permutation test (p=0.007 and p<0.001, respectively; Figure S5A ). Furthermore, the permutation test also showed that the frequencies of CD8 T cells expressing these markers differed comparing the seronegative participants and the viremic HIV-1 group (p<0.001), and also comparing the treated and viremic HIV-1 groups (p=0.045) ( Figure S5B ).

More specifically, the frequency of 2B4+PD-1-TIGIT-CD226+ CD8 T-cells was lower among memory CD8 T-cells of HIV-2 aviremic (p = 0.021) individuals when compared to seronegative individuals ( Figure 4A ). In contrast, the frequency of highly exhausted (2B4+PD-1+TIGIT+CD226-) CD8 T-cells increased gradually in relation to HIV-2 viremia and was significantly higher in both the HIV-2 aviremic and viremic groups compared to the seronegative controls (p = 0.043 and p < 0.001 respectively) ( Figure 4A ).

To focus on the TIGIT/CD226/PVR network more specifically, we limited the analysis to the expression of TIGIT and CD226. As exemplified in Figure 4B , a visually striking skewing of TIGIT and CD226 expression patterns was noted when comparing individuals of the three study groups. The population frequencies of TIGIT+CD226+ double positive and TIGIT-CD226- double negative CD8 T-cells both differed between the viremic HIV-2-infected and seronegative individuals (p = 0.003 and p = 0.010, respectively) ( Figure 4C ). In addition, the frequency of the stimulation-receptive TIGIT-CD226+ CD8 T-cells was decreased in both aviremic (p = 0.003) and viremic (p = 0.004) HIV-2-infected individuals when compared to the seronegative group. Conversely, exhausted and stimulation-unreceptive TIGIT+CD226- CD8 T-cells were increased in both aviremic (p = 0.012) and viremic (p = 0.015) individuals compared to the seronegative controls ( Figure 4C ). Of note, we found no significant differences in the CD8 T-cell differentiation phenotypes of TIGIT+CD226- and TIGIT-CD226+ CD8 T-cells ( Figure S5C ) between the groups.

In summary, the expression pattern of TIGIT and CD226 suggests that both aviremic and viremic HIV-2 infection skews the expression of the regulatory TIGIT/CD226/PVR receptor network on CD8 T-cells towards exhaustion, away from the stimulation-receptive state seen in the HIV seronegative population.