All study participants provided written informed consent.

The study was approved by the Sud-Mediterranee-1 Ethics Committee and the French Health Products Safety Agency and complied with the Helsinki Declaration.

Human immunodeficiency virus-1-infected subjects with PHI were included in a multicenter phase 3 randomized trial (ANRS-147 OPTIPRIM) (www.ClinicalTrials.gov, number NCT01033760). The endpoint was the impact of intensive vs. standard cART at month 24 on blood HIV-DNA levels. The results of this study have been published (21). Antoine Chéret was the Principal Investigator, Laurence Meyer the Methodological Investigator, and Christine Rouzioux the Virologist Investigator. We proposed a sub-study where the participants would give blood samples at day 0 (before cART initiation) and month 3. This sub-study, in which Daniel Olive and Françoise Gondois-Rey were the investigators, was designed to investigate parameters of innate immunity linked to cART efficacy. Among 90 patients in the main study, 30 patients were randomly included in this sub-study. This work shows original data on NK cells, DC, CMV, and plasma cytokines, and uses information from the main study. Patient characteristics are listed in Table 1.

Fifteen healthy donor samples were obtained from the French Blood Bank (EFS, Etablissement Français du Sang) as controls. PBMC samples were frozen and kept in liquid nitrogen until tested.

Peripheral blood mononuclear cells were stained with multiparametric panels containing 9 or 12 fluorescent markers, respectively, designed to investigate NK and DC populations. The NK panel contained NKG2A-PacBlue (clone Z199, home-made), live–dead Aqua (Life Technology), CD57-FITC (Beckman Coulter; 1/30), NKG2C-PE (R&D; 1/40), CD14-PC5 (Beckman Coulter; 1/30), CD19-PC5 (Beckman Coulter; 1/30), CD56-PC7 (Beckman Coulter; 1/30), CD3-AF700 (BD Biosciences; 1/40), and CD16-APCH7 (BD Biosciences; 1/40). The DC panel contained live–dead Aqua (Life Technology), BDCA2-FITC (Miltényi; 1/30), CD123-PercpCy5.5 (BD Biosciences; 1/20), HLA-DR-ECD (Beckmann Coulter; 1/40), CD3-PC5 (BD Biosciences; 1/40), CD56-PC5 (Beckman Coulter; 1/30), CD19-PC5 (Beckman Coulter; 1/40), CD33-PC7 (BD Biosciences; 1/40), CD14-APCH7 (BD biosciences; 1/40), and CD16-AF700 (BD Biosciences; 1/40). Cells were incubated for 20 min at RT with reagents pre-mixed in PBS, washed, and then fixed with 4% PFA. Data were acquired on a LSRII-SORP (BD Biosciences) equipped with four lasers (405 nm/100 mW, 488 nm/100 mW, 560 nm/50 mW, and 630 nm/40 mW). Photo-multiplicators were set using unstained and fully stained samples and linked to the cytometer standardization using the acquisition setting tool. Also, 7.74 × 10⁶ ± 3 × 10⁵ events were recorded. Compensations were performed with beads individually stained with corresponding reagents.

Data were exported and analyzed with FlowJo (version 9-2, MacOS X). NK cells were gated as CD3⁻CD14⁻CD19⁻CD56⁺CD16⁺, CD3⁻CD14⁻CD19⁻CD56⁺CD16⁻, or CD3⁻CD14⁻CD19⁻CD56⁻CD16⁺ (Figure S1 in Supplementary Material). Contaminating non-classical monocytes represented less than 0.11% of the gated events (not shown). The remaining data (3.5 × 10⁵ ± 2.5 × 10⁴ events) were exported to create new files further subjected to automated gating. DC’s were defined as CD3⁻CD14⁻CD19⁻CD56⁻CD16⁻HLA-DR⁺live cells. In the DC gate, the pDC were gated as CD33^lowBDCA2⁺CD123⁺ cells, the mDC as CD33^highBDCA2⁻CD123⁻ cells.

The flowClust implementation version 3.4.11on R version 3.1.2 under Linux Cent OS 6 was applied on the following parameters: CD56-PC7, CD16-APCH7, NKG2A-PacBlue, NKG2C-PE, CD57-FITC, to compute clusters as described (22). The number of clusters that fit the data optimally was estimated by computing flowClust on a predefined range and comparing them with the statistical criteria BIC and ICL (not shown). This number was estimated to be above 20; therefore, 27 clusters were computed in the events within the NK gate of the 60 samples (30 patients, T0, and M3) and one healthy donor. Two samples failed to be computed.

MFI values and event counts of the 1,593 clusters generated after computation were exported in an Excel table.

MeV (version 4.9.0, https://sourceforge.net/projects/mev-tm4/) (23) was used to visualize and group multiparametric clusters using hierarchical clustering (HCL) of their centers of CD56, CD16, CD57, NKG2A, and NKG2C MFI. Euclidean distance and average linkage were chosen. Prior to MeV, centers were rescaled to adjust the 5 (respectively, 95) percentile of each dimension to −3 (respectively, +3) and normalized. The tree was cut interactively using objective MeV tools and color interpretation of the heatmap to define populations as homogeneous groups of clusters. Names were interactively given according to the comparisons of signatures between groups and the expression of known NK cell markers. The blocks were saved and imported in a spreadsheet program, leading to a matrix with a population identifier column associated to the initial count of events. The percentages of each population of each patient sample were summarized using pivot tables.

Human immunodeficiency virus-RNA was quantified in plasma by real-time RT-PCR with the Cobas TaqMan HIV1 v2.0 assay (Roche Diagnostics). Threshold values were arbitrarily given to samples below the threshold of the assay (20 RNA copies/mL). Total HIV-DNA was quantified by ultra-sensitive real-time PCR in PBMC using the Generic HIV-DNA assay from BioCentric (Bandol, France) as described (24).

Statistical graphics were performed with Prism 6 software. The Kruskall–Wallis test followed by multiple comparison Dunn’s posttest were used to compare variables between groups. Correlations were evaluated by using simple linear regression analysis and Spearman’s rank correlation test.

We investigated the phenotypes of differentiation of NK cells with multiparametric cytometry using CD56, CD16, NKG2A, CD57, and NKG2C, known to delineate sequential stages of NK cell maturation (2, 4). In order to discover unexpected populations, multi-stained samples were analyzed through unsupervised computation of clusters, using a method previously validated (22). Computed clusters were visualized in a MeV heatmap according to their normalized MFI for the five markers and merged to identify populations as homogeneous groups of clusters (Figure 1). The tree was interactively cut to summarize the results in 19 groups, using objective tools of MeV and a subjective overview of the heatmap (Figure 1). Six of them including only a few clusters were excluded because of their rare representation. Thirteen groups were named according to comparisons of marker MFI and homogeneity with known NK population signatures, a process similar to manual gating of cytometry data visualized in dot plots (Figure 1). Accordingly, two groups of clusters expressing the highest levels of CD56, of NKG2A, no or low levels of CD16, a typical CD56^bright signature, were thus named. Seven populations expressing medium levels of CD56 and high levels of CD16 were identified as CD56^dim. They included several groups differentially expressing NKG2A and CD57, two markers linked to immaturity or maturity, respectively, and NKG2C, a NK cell receptor expanded during CMV infections. Four populations expressing low levels of CD56 or CD16 previously described in HIV-infected patients as dysfunctional NK populations were identified (15). Finally, phenotypes of NK cells usually found in human PBMC, including populations specifically expanded during HIV infection, and new phenotypes characterized by unexpected combinations of markers were found in patients’ samples using the automatic clustering and interactive merging approach. It should be emphasized that other analyses of the same data, based on the choice of higher or lower numbers of populations could have also been pertinent. As an example, the few clusters characterized by CD56^brightCD16^negNKG2C⁺ visible at the bottom of the tree were included in the CD56^bright pool while other cuts of the tree could have separated this original combination of markers.

We next determined how these NK populations were individually distributed among the 30 primary HIV-infected patients. The blocks of clusters were imported into a spreadsheet program, leading to a matrix with a population name, associated with initial counts of events and patient number. The frequencies of the 12 populations within the total NK cells of each patient were calculated and summarized using pivot tables (Table S1 in Supplementary Material). One population (CD56^negCD16⁺CD57⁺), although present in different patients, represented less than 0.5% of total NK cells and was excluded from the analysis. Frequencies of the 12 remaining populations were visualized in a MeV heatmap and used to cluster patients (Figure 2A). Three groups of patients, named X, Y, and Z appeared on the map. Group X was characterized by the highest frequencies of CD56^dimNKG2A⁺ and CD56^negCD16⁺NKG2A⁺ NK cells. Group Y showed high frequencies of CD56^dimNKGC2⁺CD57⁺and CD56^dimCD57⁺ NK cells. Group Z contained mainly a population of CD56^dim CD57⁻NKG2A⁻ NKG2C⁻ (Figure 2A).

We checked whether the frequencies of the most frequent NK cell populations directing the clustering (the four populations on top of the list) were significantly different between the three groups (Figure 2B). Group X showed significant higher frequencies of CD56^dimNKG2A⁺ than Y and Z (34% vs., respectively, 13.8 and 14.3%). Group Y showed significant higher frequencies of CD56^dimCD57⁺ NK cells than X and Z (21% vs., respectively, 2.9 and 7.2%), and higher frequencies of CD56^dimNKG2C⁺CD57⁺ than X and Z (17.8% vs., respectively, 2.9 and 0.9%). Group Z showed 42.6% of CD56^dim NKG2A⁻NKG2C⁻CD57⁻ while this phenotype represented only 12 or 16% of NK cells in, respectively, X and Y. Accordingly, the three groups of patients were significantly characterized by, respectively, high proportion of CD56^dimNKG2A⁺ for group X, high proportion of CD56^dimCD57⁺ NK cells for group Y, and high proportion of CD56^dim NKG2A⁻NKG2C⁻CD57⁻ for group Z.

To evaluate the relationship between NK profiles of patients and HIV infection, the clinical and virological characteristics of patients within groups were compared at T0 (Table 2). Primary infection was defined by detectable plasma HIV-RNA and incomplete HIV-1 western blot, acute infection by one band or fewer (21) and early infection by more than one band. Four out of 14 patients from group X, 4 out of 10 patients of group Y, and all 4 patients of group Z were acutely infected (Figure 3A). The mean age of patients from each group was not significantly different (37, 40, and 42 years for, respectively, X, Y, and Z), nor was the mean of estimated time since infection and enrollment (36.5, 35.3, and 34.7 days for, respectively, X, Y, and Z) (Table 2). T-CD4 counts were similar for all groups, although group Z patients showed a trend to lower levels (Table 2). T-CD8 counts were significantly higher in X than Y and Z (mean of 2,148 counts/μL vs. 1,308 and 1,415 for, respectively, Y and Z) (Table 2).

A striking difference was observed when the viral load was compared between the three patient groups. Patients from groups X and Z displayed significantly higher mean HIV-RNA levels than patients from group Y (respectively, 5.77 log₁₀ HIV-RNA copies/mL and 6.05 log₁₀ HIV-RNA copies/mL vs. 4.88 log₁₀ HIV-RNA copies/mL in group Y), while HIV-DNA were not significantly different despite a trend to lower levels for Y (Figure 3B). Thus, the major differences observed between the three groups of patients at inclusion highlighted the lower viral load of group Y, the group defined by an expansion of CD57⁺ NK cells.

We next addressed the question of whether patients of group X, Y, and Z were different with respect to other major disease progression markers. We analyzed plasma inflammatory markers (IL-6, IP-10) (25, 26) and also focused on markers of innate immune activation and exhaustion, including CD38 expression on NK cells, CD86 expression on monocytes, PDL-1 expression on mDC (27), and pDC and mDC frequencies (28). These markers were all described to be linked to viral load and disease progression (29). They were compared between the groups of patients and some of them were compared to a group of 15 healthy donors (Figure 4).

Patients of group X displayed higher IL-6 and IP-10 plasma levels than patients of group Y (Figure 4A). CD38 expression on NK cells was significantly increased in all patients groups as compared to healthy donors. CD86 on monocytes and PDL-1 on mDC were significantly increased in patients X as compared to healthy donors, while patients Y showed a trend to lower increase and values of PDL-1 not different from that of healthy donors (Figure 4B). All patients groups showed significant decrease of pDC as compared to healthy donors, but patients Y showed a clear trend to higher values of pDC frequencies (Figure 4C). Only patients X and Z showed significant decrease of mDC frequencies while patients Y exhibited values similar to healthy donors (median of, respectively, 0.3 vs. 0.33%) (Figure 4C). Therefore, in addition to exhibit the lowest activation, Y patients displayed low exhaustion of DC, as shown by a trend to higher frequencies of pDC and frequencies of mDC not different from that of healthy donors (Figure 4C). As expected according to their high viremia levels, patients from groups X and Z displayed higher immune activation and exhaustion while patients Y exhibited a better immune status in PHI.

We then assessed the impact of the different virological and immunological status at inclusion on early cART outcome. While all patients showed a 3-log decrease of viral load after 3 months of cART (Table 2), we searched for those patients having reached a viral load below 50 copies/mL at M3. Notwithstanding the cART regimen, 6 out of 14 (57%) patients X, 1 out of 4 (25%) patients Z, and 7 out of 10 (70%) patients Y reached a viral load below 50 copies/mL at M3. Taken together, only 38% of patients lacking significant frequencies of CD57⁺ NK cells (X + Z) reached a threshold of 50 log₁₀ HIV-RNA copies/mL at M3 whereas 70% of patients with high frequencies of CD57⁺ NK cells (Y) could reach it.

In order to evaluate the kinetics of NK cell maturation at a short time after infection, we compared the frequencies of CD57⁺ NK cells at inclusion and after 3 months of cART in groups of patients X, Y, and Z (Figure 5). All CD57⁺ NK cells were summed. HIV-RNA showed a mean decrease of 2 logs in both groups after 1 month of treatment, one additional log being lost during the third month (left graph). In the meantime, the frequency of CD57 NK cells remained unchanged in patients from group Y (47.5% at T0 vs. 48.9% at M3), while the frequencies of CD57 populations of patients from group X and Z significantly increased, or alternatively the proportion of immature NK cells decreased (13.9% at T0 vs. 20.5% at M3 for X; 11.7 vs. 19.4% for Z). Therefore, even if the frequencies of CD57 NK cells increased during the 3 months of cART, the values reached remained far from those of patients from group Y at T0 (47.5%).