This study included samples from three sources. We obtained data and additional peripheral blood samples from a previously published cohort of 740 healthy individuals aged 40–97 that approximately mirrored San Francisco Bay Area ethnic demographics (24). Additional healthy individuals between the age of 18 and 80 years from the same draw area were recruited. Deidentified samples from HLA-A*02⁺ individuals of various ages and with positive CMV serology were purchased from the Stanford Blood Center (Palo Alto, CA, USA). The study was in accordance with the Declaration of Helsinki, approved by the Stanford Institutional Review Board, and all participants gave written informed consent.

For cellular phenotyping of tetramer-specific cells, we used CD3-APC/Cy7, CD8-qDot605, CD45RA-PE/Cy7, CCR7-PerCp/Cy5.5, CD28-PE, CD85j (ILT-2)-APC, and tetramer–Pacific Blue. Antibodies were purchased from BD Bioscience, Biolegend, or eBioscience. HLA-A*0201 monomers loaded with peptides for CMV pp65 (NLVPMVATV) or EBV BRLF1 (YVLDHLIVV) were tetramerized and labeled with streptavidin–Pacific Blue.

Total T cells were isolated by negative selection using human T cell RosetteSep enrichment kit (StemCell Technologies) from platelet donor apheresis lymphocytes of HLA-A2 donors who are CMV seropositive. T cells were stained with CD4, CD8, pp65 HLA-A*0201 tetramer, and CD85j antibodies. CD85j⁺ and CD85j⁻ pp65-HLA-A*0201 tetramer⁺ CD8 T cells were sorted using a FACSAria (BD Bioscience) and split into two to four replicates with 4,000–5,000 cells per replicate. Total RNA was extracted from each T cell replicate using RNeasy Plus Micro kit (Qiagen), followed by generation of cDNA using SuperScript VILO master mix (Invitrogen). The amplification and sequencing of TRB gene libraries followed the protocol as previously described (25).

The sequences were mapped to human TRB reference sequences as described in detail previously (25, 26). Clonotypes were defined as sequences with the same TRBV and TRBJ gene segments and identical CDR3 amino acid sequences. In addition, any clonotype that was only found in one replicate library was filtered out of the analysis. The clonality index for each population was calculated using the lymphclon package (https://arxiv.org/abs/1408.1149) (25).

PBMCs were left unstimulated or stimulated for 18 h with CMV peptide pools in the presence of brefeldin A and monensin (BD Bioscience). For CMV-specific stimulation, two peptide super pools, each made up of overlapping peptide pools of four different antigens, were used. The immediate early (IE) pool consisted of IE-1, IE-2, US3, and UL36. The late pool consisted of pp65, UL32, UL48AB, and UL55 (gB), based on previously described work (27). Following stimulation, cells were resuspended in CyFACS buffer (1× PBS with 0.1% BSA, 2 mM EDTA, and 0.5% sodium azide) and stained with isotope-tagged antibodies before being acquired on the CyTOF. For a detailed protocol, see http://iti.stanford.edu/himc/protocols.html (CyTOF ICS protocol) and Table S1 in Supplementary Material.

Data acquired from CyTOF were initially analyzed using FlowJo v10.1 (FlowJo Inc.). CD3⁺CD19⁻CD8⁺CD4⁻ cells expressing CD107a or one of the following cytokines, IFNγ, TNFα, IL-2, GM-CSF, or MIP1β, after stimulation with IE or late pool were considered “CMV-responsive” CD8 T cells. The CMV-responsive cells for 30 individuals were concatenated, and cluster analysis was performed using X-shift (28). For final clustering, basic phenotypic (CD45RA, CCR7, CD28, CD27, and CD127) and the six preselected response factors were excluded.

Statistical analyses, including Spearman correlation coefficients and non-parametric testing, were performed using GraphPad Prism v6 (GraphPad, San Diego, CA, USA). All p-values were derived using two-tailed tests and p < 0.05 were considered significant.

We first characterized the relationship of CD85j⁺ cells to age and to specific CD8 T cell subsets in a cohort of 210 healthy individuals that has been previously described (24). CD85j⁺ CD8 T cells increased with age (Figure 1A). Similarly, terminal differentiated effector CD8 T cells (termed “TEMRA”) increased with age, with a slope and correlation coefficient similar to that of CD85j⁺ cells (Figure 1B). Effector memory (EM) and naïve (N) CD8 cell frequencies also correlated with age (EM: p = 0.007, r = 0.19; N: p < 0.0001, r = −0.5) (Figure 1B; Figure S1A in Supplementary Material). As TEMRA and EM populations both showed similar positive correlations with age as the CD85j⁺ population, we next asked which CD8 T cell subset, TEMRA or EM, most closely correlated with CD85j⁺ cell frequencies. CD85j⁺ cells positively correlated with TEMRAs (Figure 1C). No correlation between EM and CD85j⁺ cell frequencies was found. Furthermore, CD85j⁺ cells negatively correlated with naïve CD8 T cells (N: p < 0.0001, r = −0.405) (Figure S1B in Supplementary Material). We also observed increased frequencies of CD85j⁺ cells and TEMRA cells in CMV-positive individuals and in males (Figures S1C,D in Supplementary Material).

To confirm that CD85j is expressed on TEMRA and whether increased CD85j⁺ cells with age is a result of increased frequencies of TEMRAs with age, we analyzed CD85j expression on T cell subsets in a subcohort of 20 mid-age (40–50 years) and 20 older (>60 years) individuals. Indeed, CD85j is expressed by TEMRAs, with a small fraction of CD85j⁺ EM T cells (Figure 1D). Central memory and naïve cells had very few CD85j⁺ cells. Of note, CD85j-intermediate staining is almost exclusively observed within the effector T cell populations. Naïve and CM populations lack this population and the low CD85j staining observed in naïve and CM cells is seen in all populations of cells. With age, there was a trend for increased percentage of CD85j⁺ TEMRA and EM, however, the difference did not reach statistical significance in our small subcohort (Figure 1E). We also compared the frequencies of CD85j⁺ cells within T cell subsets by gender and CMV status within this subcohort. CMV-positive individuals had higher frequencies of CD85j⁺ TEMRAs and CD85j⁺ EM CD8 T cells compared with CMV-negative individuals (Figure 1F). Gender also has a modest effect of the frequencies of CD85j⁺ TEMRAs, with males exhibiting slightly higher frequencies than females (Figure 1G). Age only had a minor effect on CD85j expression by effector subsets, not reaching significance. On the other hand, CMV infection caused a pronounced increase in CD85j⁺ TEMRAs, regardless of age. Thus, mainly terminal differentiated effector CD8 T cells express CD85j and multiple factors including differentiation state, CMV infection, gender, and age influence the frequencies of CD85j⁺ T cells and effector T cell subsets.

CMV⁺ individuals expressed the highest frequencies of CD85j⁺ TEMRAs, thus we further investigated the expression of CD85j on CMV- and EBV-specific CD8 T cells from older CMV⁺ individuals. Directly ex vivo, both CMV (pp65) and EBV (BRLF1) tetramer-positive CD8 T cells were detectable (Figure 2A) but CMV-specific CD8 T cells were present at much higher frequencies than EBV-specific cells (median 1.2 vs. 0.29%, respectively) (Figure 2B). Expression of CD85j was detectable on CMV- and EBV-specific CD8 T cells, in particular on CD28⁻ effector cells (Figure 2C). Surprisingly, the expression of CD85j was similar on CMV- and EBV-specific cells (Figure 2D). Moreover, CD85j expression on CMV- and EBV-specific cells from the same individuals positively correlated (Figure 2E), suggesting that the probability of CD85j expression is not influenced by the nature of antigenic stimulation or the extent of clonal expansion, but by response patterns unique to individuals.

To further understand why the frequencies of CD85j⁺ cells varied between individuals, irrespective of whether specific for CMV or EBV, we characterized the phenotypic composition of CD85j⁺ CMV- and EBV-specific CD8 T cell populations. Expression of CD85j on CD8 T cell subsets was also confirmed in total CD8 T cells and CMV-specific CD8 T cells, demonstrating again robust expression on effector cells (Figure 2F). Both CMV- and EBV-specific populations of CD85j⁺ cells were predominated by effector populations (TEMRAs and EM cells), although CD85j⁺ CMV-specific cells had a higher median frequency of TEMRAs than EBV (Figure 2G). Similar to results from total CD8 T cells (Figure 1), the frequency of CD85j⁺ CMV- and EBV-specific cells positively also correlated with the frequency of TEMRA⁺ CMV- and EBV-specific cells (Figure 2H). However, individuals with high frequencies of CMV-specific CD8 T cells, which is indicative of memory inflation, exhibited increased percentage of cells expression CD85j and of the TEMRA subset (Figures 2I,J), suggesting that CD85j is gained during differentiation and retained during expansion of antigen-specific terminal-differentiated effector cells.

If CD85j expression is a late event in T effector cell expansion and differentiation, CD85j⁺ cells would be biased for larger clonal populations of antigen-specific T cells. Thus, we compared the TCR repertoires between CD85j⁺ and CD85j⁻ CMV-specific (pp65 tetramer-positive) CD8 T cells. The repertoires in individual donors largely overlapped indicating that both subsets derived from the same progenitor cells (Figure 3A). The total number of unique TRB sequences (TCR richness) was similar between the two populations, with median counts of 213 (range: 113–1,361) and 185 (range: 70–781) for CD85j⁺ and CD85j⁻ populations, respectively (Figure 3B). TRB sequences unique for one subset were all highly infrequent and, therefore, likely reflected the low probability of reidentification rather than uniqueness for one particular subset. Sizes of T cell clonotypes in both subsets showed a similar logarithmic distribution with a few clones dominant and no depletion of infrequent clones in either subset. Accordingly, clonality indices, a measure of clonal expansion, were not different (Figure 3C). In addition, there was a strong correlation between individual TCR clone frequencies in CD85j⁺ and CD85j⁻ populations (Figure 3D). Clonality indices also positively correlated with frequencies of CMV-specific CD8 T cells (Figure 3E). Thus, clonality reflects memory inflation but this clonal expansion occurs in the CD85j⁺ as well as CD85j⁻ subsets and CD85j expression does not appear to increase with clonal size or to have major influence on selection of the antigen-specific repertoire.

In initial studies using flow cytometry, stimulation with immediate early or late CMV antigens revealed that CD85j⁺ CD8 T cells produced significant levels of IFNγ (Figure S2 in Supplementary Material). To further interrogate the functional role of CD85j in antigen-specific responses, we utilized mass cytometry to determine simultaneous expression changes of 30 different parameters on CD8 T cells after CMV-specific peptide stimulation with immediate early or late antigen peptide super pools. These parameters included 5 T cell phenotyping markers, 7 cytokines, 2 cytotoxic factors, and 12 markers related to activation, exhaustion, and/or senescence. We termed all CD8 T cells expressing CD107a or one of the following cytokines, IFNγ, TNFα, IL-2, GM-CSF, and MIP1β, after stimulation as “CMV-responsive.”

From our global clustering analysis, we found five distinct clusters of CMV-responsive cells (Figure 4A). The most prominent cluster, cluster 1, included 70.3% of all CMV-responsive cells (Figure 4B) and displayed no significant difference in cell frequencies between the two different peptide stimulations (Figure 4C). Cluster 4 was the only cluster that showed differences based on peptide stimulation, where most cells within this cluster were found with late, but not immediate early, peptide stimulation. CD85j expression differentiated into three groups; CD85j-negative, -intermediate, and -high (Figure 4D). At a single-cell expression level, clusters 3 and 5 were CD85j-neg/low, clusters 1 and 4 were CD85j-intermediate, and cluster 2 was CD85j-high (Figure 4E). Overall, CD85j^intermediate and CD85j⁻ clusters made up 81.0 and 14.7% of the CMV-responsive population, respectively. The smallest fraction of CMV-responsive cells was CD85j^high (cluster 2: 4.3% of all cells). The analysis of functional markers shows that CD85j^high cells expressed high levels of MIP1β and CD33. Further analysis revealed that this MIP1β⁺CD33⁺CD85j^high population was already present in unstimulated cells (Figures S3A,B in Supplementary Material). Therefore, it is undetermined whether these CD85j^high cells are truly CMV-specific T cells or a population of myeloid cells.

Median expression of (1) phenotypic markers, (2) response cytokines, and (3) other markers were individually compared between the five clusters using unbiased hierarchal clustering (Figure 4F). CD85j^intermediate clusters separated from CD85j⁻ and CD85j^high clusters in all comparisons. CD85j^intermediate clusters (clusters 1 and 4) strongly separated from CD85j⁻ clusters by phenotypic but not stimulation-induced cytokine expression. Similar to previous analysis of CD85j⁺ cell phenotypes, we found that CD85j^intermediate clusters contained primarily terminally differentiated (CD45RA⁺CCR7⁻CD28⁻) CD8 T cells, whereas CD85j⁻ cells were CD27⁺ memory or naïve-like populations. Functionally, both CD85j^intermediate and CD85j⁻ cells expressed high levels of IFNγ, TNFα, and CD107a. However, the CD85j^intermediate population uniquely coexpressed high levels of CD57, granzyme B, and perforin, which are common markers of senescent cells or meditators of CTL responses (Figure 4F; Figure S4 in Supplementary Material). A subset of CD85j^intermediate cells from cluster 4 also coexpressed CD56 and NKG2C. Little to no expression of PD-1, an exhaustion marker, was detected on these cells. Median expression values and p-values for individual marker comparison between the different clusters are provided in Table S2 in Supplementary Material. Single-cell expression analysis of all CMV-responsive CD8 T cells demonstrated CD85j, CD57, and granzyme B expression positively correlated and clustered with NK or innate markers, CD56, perforin, and NKG2C (Figure 4G). In addition, hand-gating of CMV-responsive CD8 T cells confirmed that the majority of CD85j⁺ cells coexpress CD57 and granzyme B (Figure 4H), and coexpression is maintained across multiple donors and CMV antigen responses (Figures 4I,J). CD85j⁺CD57⁺granzymeB⁺ CD8 T cells were also found in the unstimulated population and single-cell analysis again revealed that CD57 and granzyme B positively correlated with CD85j in the resting CD8 T cell population (Figure S3 in Supplementary Material). Thus, CD85j⁺ CMV-responsive CD8 T cells are phenotypically and functionally similar to that of terminally differentiated, not exhausted, CD8 T cells.

“Senescent” CD8 T cells are classically characterized by a reduction in proliferative capacity but maintained cytokine production, unlike exhausted cells, which also lost the ability to produce effector cytokines. Thus, we determined the potential role of CD85j in antigen-specific cytokine responses and proliferation in CMV-infected, older individuals. Consistent with the functional data from CyTOF analysis (Figure 4), stimulation of PBMCs with immediate early (IE-1) and late (pp65) CMV antigen peptide pools both induced simultaneous production of IFNγ and TNFα by a subset of CD8 T cells (Figure 5A). However, blocking CD85j did not affect the frequencies of CD8 cells producing IFNγ and TNFα upon pp65 and IE-1 stimulation (Figures 5B,C). Thus, CD85j does not appear to play a role in the functional ability of CD8 T cells to produce cytokines in response to antigen.

Stimulation with pp65 peptide pool induced proliferation (CFSE^low cells) of a low-frequent T cell population (Figures 6A,B). Unlike cytokine production, CD85j blocking demonstrated a significant increase in the frequencies of proliferating CD8 T cells. CD85j blocking alone without stimulation was not sufficient to cause proliferation (Figure 6A; Figure S5A in Supplementary Material). To establish whether this inhibitory effect was specifically blocking CD85j engagement on CD8 T cells and not on CD85j-expressing antigen-presenting cells, we developed a cell culture system using immobilized pp65-loaded HLA-A*0201 tetramers to stimulate purified CD8 T cell in vitro. As observed with total PBMCs, CD85j blocking enhanced proliferation of CMV-specific CD8 T cells (Figure 6C; Figure S5B in Supplementary Material). Thus, engagement of CD85j on CD8 T cells reduced proliferation in response to antigen, indicating that CD85j may be a checkpoint regulator inhibiting clonal expansion of virus-specific T cells during aging.