HIV-infected female sex workers (n = 24) were recruited from a longitudinal cohort based in Nairobi, Kenya [22] , and all gave written informed consent prior to their participation in this research. Ethical approval was obtained from relevant review boards at Kenyatta National Hospital/University of Nairobi Ethics and Research Review Committee and the University of Manitoba Health Research Ethics Board, Bannantyne Campus. Clinical and demographic data were collected on subjects biannually; including CD4 T cell counts (Table 1). Antiretroviral therapy (ART) naive subjects with CD4+ counts above 400 cells/ml for over 6 years were classified as long-term non-progressors (LTNP). LTNP were followed for a mean and median years of 11.98 and 14.31 years respectively (range 6.41 - 17.43 years, n = 10). Participants who did not meet the criteria for LTNP were considered to be normal progressors (NP, n = 9), which has been demonstrated to be 3.5 years until AIDS in this cohort [23] . Additionally five subjects on antiretroviral therapy (ART) were included in the study.

A peptide library (9 mers overlapping by 8 amino acids) derived from the HIV-1 p24 clade A1 ancestral sequence (Sigma-Genosys) were pooled in a matrix format using Deconvolute This! version 1.0 [24] (courtesy of Mario Roederer, Vaccine Research Center, NIAID, NIH). Clade A1 is the predominant circulating clade in this study population [25] - [27] . Peptide pools were used at 2 µg/ml/peptide, and stimulations were accompanied by 2 positive controls Cytomegolovirus, Epstein-Barr virus and Influenza virus peptides (CEF, 32 peptides/pool, 2 µg/ml/peptide; AnaSpec) and Staphylococcus aureus enterotoxin B (SEB, 0.1 mg/ml; Sigma-Aldrich) and duplicate negative controls consisting of media alone. Monoclonal antibodies for flowcytometry analyses included: CD3-AmCyan, CD8-APCCy7, CCR7-PeCy7, IFNγ-FITC, CD107a-PeCy5, MIP-1β-PE, IL2-APC (BD Biosciences), TNFα-Pacific Blue, CD27-Alexa Fluor 700 (eBiosciences) and CD45RA-ECD (Beckman Coulter).

Fresh PBMC were isolated by density gradient centrifugation, re-suspended in RPMI 10% fetal bovine serum (R-10), and stimulated with the HIV-1 p24 peptide pools described above, with each peptide at a concentration of 2 μg/ml/peptide. Pre-titrated amounts of CD107a were added prior to stimulation, as described [14] . After one hour at 37˚C 5% CO₂, protein secretion was inhibited by the addition of titrated amounts of Golgi-Plug and Golgi-Stop (BD Biosciences), and incubation was continued for another 13 hours. Cells were then washed, stained for surface marker expression, permeabilized (BD Biosciences), and stained for intracellular markers.

Cells were analyzed on a LSRII flowcytometer (BD Biosciences). Between 30,000 and 100,000 events were collected within the lymphocyte gate per sample. Data analyses were performed using FlowJo version 9.2 (TreeStar). Boolean gates were applied to the 3 surface markers and the 5 functions and was assessed using SPICE 5.1 (courtesy of Mario Roederer, Vaccine Research Center, NIAID, NIH).

HIV-specific CD8+ T cell responses were considered positive for a given parameter if they were ≥2-fold higher than the mean of their respective negative controls. Data are reported after background subtraction. Polyfunctional responses were also background subtracted and a lower threshold corresponding to the 90th percentile of distribution of negative values was built for each cytokine pattern and values below this threshold were set to 0 [14] . Statistical analyses were performed using Graph Pad Prism 5.0 and SPICE 5.1. Phenotype comparisons were determined using Wilcoxon matched-pairs and one-way ANOVA Kruskal-Wallis, Post test Dunn’s multiple caparisons were also done. Functional correlations were Spearman’s rank correlation. Due to the large number of correlations involved, we highlight and report only the most significant and strongest correlations (Those with a threshold for significance of less than p = 0.001 (Spearman’s rank correlation)).

A number of memory T cell phenotypes have been described, with the dominant paradigm based on CCR7, CD45RA, and CD27 expression [4] [7] [28] . These markers can be used to define 8 phenotypes including naive, central memory, effector memory, transitional memory, and terminally differentiated effector subsets (Table 2) as well as undefined combinations of these markers that have not been examined to date. Following stimulation with HIV-1 p24 peptide pools and a Cytomegalovirus, Epstein barr virus and Influenza virus (CEF) peptide pool we were able to characterize the memory phenotypes of both HIV and CEF specific CD8+ T cells (responding with any single or combination of functions) in HIV-infected subjects using these surface markers. Representative flowcytometry gating is shown in Figure 1.

Overnight HIV p24 and CEF-specific responses were measured in chronically HIV infected subjects (n = 24) using multiple immunological readouts including IFNg, MIP-1b, TNFa, CD107a, and IL2 responses. Published previously in Richmond et al. 2012 J. Virology, HIV and CEF-specific responses were detected in the majority of HIV+ subjects [29] . Responding cells were assigned to the 8 phenotypes defined in Table 2 to determine the

^a[4] [30] [52] -[56] ; ^bNo phenotypic category has been ascribed to these combinations of markers.

memory phenotypes of HIV and CEF-specific CD8+ T cells. Overall, HIV and CEF-specific antigen responding CD8+ T cells were evenly distributed between 5 of the 8 memory subsets; CCR7+CD27+CD45RA+ (naive), CCR7−CD27−CD45RA+ (T_TD), and CCR7−CD27−CD45RA− (T_EM), including a substantial number of responses in two novel phenotypes CCR7+CD27−CD45RA+ (Undefined #2), CCR7+CD27-CD45RA− (Undefined #3), (Range 14.4% - 20.5%, Figure 2(a)) while the remaining three categories CCR7-CD27+CD45RA+ (Undefined #1), CCR7+CD27+CD45RA− (T_CM) and CCR7−CD27+CD45RA− (T_TM) were observed less frequently (<10%). We then compared the distributions of the various phenotypes between HIV specific and CEF-specific CD8+ T cells. HIV-specific CD8+ T cells were more likely to be CCR7+CD27−CD45RA+ (Undefiend #2) compared to CEF-specific cells (p = 0.0012, mean 17.27% versus 14.68%, respectively, Wilcoxon matched pairs). This data is in agreement with previous studies that show that different chronic viral infections induce CD8+ T cells with distinct memory phenotypes and functionality [30] [31] .

To more clearly understand phenotypic differences observed between HIV and CEF specific CD8 T cells, we compared the expression of each surface marker individually on responding cells (Figure 2(b)). Our data show that CD27 and CCR7 are relatively equally distributed on HIV and CEF specific CD8+ T cells. Both HIV and CEF-specific cells are typically CD27 negative (67.17% and 67.09%, respectively). There was a modest increase in expression of CD45RA on HIV-specific CD8+ cells compared to CEF-specific CD8+ T cells (54.65% and 50.96%, respectively, p = 0.008 Wilcoxon matched-pairs). These data show that depending on the phenotype CD8+ T cells responding to CEF and HIV antigens may have unique memory phenotypes. However, taken together, both categorical and individual marker data show that there is relatively a homogeneous distribution of virus specific T cell responses among all the different phenotypic categories of T cells as defined by CCR7, CD27 and CD45RA, including two previously undefined phenotypes.

We next determined the surface memory phenotypes of HIV-specific CD8+ T cells expressing each functional parameter individually, independent of expression any other functional readouts (IFNg, MIP-1b, TNFa, CD107a, and IL2). CD8+ T cells positive for each parameter were stratified across the 8 phenotypic categories. Here we observed several functional responses that appear to correlate with the phenotype measured (Figure 3). For example and surprisingly, cells defined as T_EM were the least likely to be IFNγ positive (p < 0.0001, Kruskal-Wallis, Post test Dunn’s multiple comparison), possibly due to the fact that cells responding with IFNγ tended to have higher CCR7 expression, than cells responding with other functional parameters (p < 0.0001,data not shown). Each of the phenotypes had a particular functional profile associated with it; the majority of “Undefined #2”

(CCR7+CD27−CD45RA+) were CD107a and IFNγ positive, while T_CM and Undefined #3 (CCR7+CD27− CD45RA−) were the predominantly expressing IFNγ. The majority of T_TD expressed higher MIP-1b+, T_TM were more often IL2 positive, and T_EM were most likely to be CD107a and IL2 positive. Interestingly, 10% - 20% of responding cells regardless of function could be phenotypically described as naive. Thus, effector function as defined by cytokine responses to HIV appear to be completely independent of, and not predicted by, surface memory phenotype.

To further explore the link between functional parameters and surface phenotypic categories, we next sought to determine if any of the functions correlated with a specific surface phenotype. We correlated the percentage of responding cells for each function with the proportion of overall responding cells contained within each of the 8 phenotypic categories. All functions (CD107a, IFNγ, MIB-1γ and TNFγ, excluding IL2) correlated with one or more of the eight memory phenotype categories (Table 3), Both the percentage of cells responding with IFNγ and TNFα had similar correlation patterns, consistent with our previous data which found that the expression of IFNγ and TNFα are strongly correlated in normal progressing HIV+ subjects [29] . Both were positively associated with the Undefined #2 and T_CM phenotypes (All p < 0.0001, Spearmen rank correlation), but were inversely correlated to T_TM, Undefined #3 and T_EM phenotype (p < 0.0001, p < 0.0001 and p = 0.0002, respectively). The percentage of cells responding with CD107a was also inversely correlated to the T_EM phenotype (p < 0.001), but positively correlated with Undefined #1 (p < 0.001). Conversely, the percentage of cells responding with MIP-1β had no similarity to any of the other functions, only correlating with the Undefined #3 category (p = 0.0005), again consistent with our previous data, which found no correlation between MIP-1β responses and other functions [29] . These data suggest that while some functional attributes do weakly associate with specific “memory phenotypes”, this is parameter-specific. These data suggest that the allocation of antigen-specific cells to functional categories on the basis of surface marker expression is exceedingly complex, and varies greatly depending on the functional attribute examined. Thus defining function upon surface phenotype may not be predictable using the markers examined here.

The number of functions expressed simultaneously by a single antigen specific cell is increasingly being considered as a correlate of protective immunity in a number of infections [14] [32] - [35] . To determine the surface phenotypes of polyfunctional cells, we stratified our data by CD8+ T cells concurrently expressing increasing numbers of functions (5+, 4+, 3+, 2+, and 1+). Polyfunctional responses were observed in all eight memory phenotypic categories, regardless of the number of functions expressed. Figure 4 shows, for each polyfunctional level, the relative proportion of surface phenotypes that account for the response. The percentage of cells

^agray and orange shading indicates positive and negative correlations, respectively.

responding with each level of polyfunctionality was, than correlated with the proportion of responding cells in each of the 8 phenotypic categories (Table 4). Monofunctional responding cells were associated with the phenotypic categories Undefined #1, Undefined #2 and T_CM (all p < 0.0001), but were inversely correlated to T_TM, Undefined #3 and T_EM (all p < 0.0001). Similarly, dual-functional responding cells were inversely correlated to T_TM and T_EM (both p < 0.0001). Cells expressing 3 and 4 functions were positively correlated with the phenotypic category Undefined #3 (p = 0.0004 and p = 0.0007, respectively) while cells expressing 4 and 5 functions were correlated to the phenotype T_TM (p < 0.0001 and p = 0.0003, respectively). These data suggest that with increasing expression of multiple functions, the memory surface phenotype is variable. Although most of our data suggests that memory surface markers are unable to predict all functionally active CD8+ T cells, there was a trend towards polyfunctionality associating with the transitional memory phenotypes, T_TM and Undefiend #3.

We finally determined whether HIV and CEF-specific CD8+ T cells differ in their phenotypic distributions on the basis of HIV disease status in this cohort. Participants were separated into normal progressors (NP), long- term non-progressors (LTNP) and patients on ART, to determine phenotypic differences related to disease progression. HIV and CEF-specific CD8+ T cells were assigned to the 8 surface phenotypes (as above). Phenotypic differences were observed in the proportion of responding CD8+ T cells between LTNP, ART or NP subjects, with the LTNPs having a higher proportion of naive, and a lower proportion of T_TD cells, than NPs and ARTs (Figure 5(a), Figure 5(b), HIV, p < 0.0001 and p = 0.0094; and CEF, p = 0.0012 and p = 0.0074, respectively, Kruskal-Wallis, Post test Dunn’s multiple comparison). These data suggest that the phenotypic changes on CD8+ T cells following stimulation also differs between clinical groups and viral-specific populations, perhaps due to different activation/stimulation histories.