Participants were identified from the Melbourne HIV Cohort, a prospective study of HIV-positive and HIV-seronegative men who self-report having sex with men. Participants in the Melbourne HIV Cohort were reviewed annually to assess co-morbidities. Peripheral blood mononuclear cells (PBMC) and plasma were prepared and archived from each visit. Baseline samples were analyzed from 20 cART-naïve HIV⁺ MSM and 15 HIV⁻ MSM matched for age with the HIV⁺ MSM at the baseline time-point. HIV⁺ individuals were recruited when they were cART-naïve and followed-up every 3 months for 12 months following cART-initiation, then annually thereafter. Of the 20 HIV⁺ MSM, one initiated a cART regimen consisting of efavirenz, festinavir and lamivudine. The other 19 participants received a cART regimen of tenofovir and emtricitabine, plus either efavirenz (n = 6), rilpivirine (n = 5), raltegravir (n = 3), ritonavir + atazanavir (n = 4), or ritonavir + neviripine (n = 1); two individuals had their regimen altered (from raltegravir to dolutegravir and from efavirenz to raltegravir) during the follow up period. At the time of the study, 10 of the HIV⁺ MSM had reached the 24-month post-cART initiation time-point and were included in the analysis. Exclusion criteria included co-morbid disease (e.g., cardiovascular disease, diabetes) and current use of statins, steroids, or other anti-inflammatory medications. For selected experiments, an additional 14 HIV⁻ men of a similar age were recruited from the general community. Ethical approval for this study was obtained from the Alfred Hospital Research and Ethics Committee.

Cells and plasma were prepared from whole blood collected into acid citrate dextrose tubes. PBMC were collected following Ficoll density gradient centrifugation of blood and stored in liquid N₂. Cells were stained with LIVE/DEAD^® fixable dead cell stain (ThermoFisher Scientific, Waltham, MA, USA) prior to immunophenotyping. Expression of surface receptors on NK cells, monocytes, and T cells were detected by staining with the following antibodies: CD56 APC (clone NKH-1) from Beckman Coulter (Brea, CA); CD14-V500 (clone M5E2), CD16 PE-Cy7 (clone 3G8), CD3 PerCP-Cy5.5 (clone UCHT1), CD38-PE (clone HB7), HLA-DR FITC or APC-H7 (clone G46-6), CD4 PE-Cy7 (clone RPA-TA), CD8 APC-H7 (clone SK1), all from BD Biosciences (San Jose, CA, USA); CD3 BV510 (clone OKT3), CD56 AF700 (clone HCD 56), CD57 Pacific Blue (clone HCD57), CD355 PerCP-Cy5.5 (NKp46, clone 9E2), CD337 AlexaFluor 647 (NKp30, clone CD337), all from Biolegend (San Diego, CA, USA). Expression of intracellular FcRγ was detected after labeling of surface antigens and following permeabilization with Perm/Wash buffer 1 (BD Biosciences), then staining with anti-FcRγ FITC (FcεR1, γ subunit, rabbit polyclonal, Millipore, Darmstadt, Germany). The specificity of the polyclonal anti-FcRγ FITC antibody has been previously demonstrated (3). Cells were acquired on a Fortessa LSR flow cytometer (BD Biosciences) and data analyzed using FlowJo software (version 10, FlowJo LLC, Ashland, OR, USA). Gating strategies for each cell type are depicted in Figure S1 in Supplementary Material.

Plasma concentrations of soluble CD163 (sCD163) and CXCL10 were measured using commercial ELISA kits as per manufacturer’s instructions (Macro 163, IQ Products, Groningen, Netherlands and DIP-100, Quantikine ELISA, R&D Systems, Minneapolis, MN, USA, respectively). IgG reactive with HCMV was quantified using HCMV lysate, HCMV glycoprotein B (gB), and HCMV IE-1 antigens as previously described (3). CMV seropositivity was defined as >2 SD above the mean antibody levels for HCMV lysate derived for a set of 11 samples that had been deemed seronegative by the ARCHITECT CMV IgG assay (Abbott Diagnostics, IL). Data are presented in arbitrary units (AU) defined relative to a standard plasma pool run on each plate. Cross-reactivity of the CMV ELISA with other herpes viruses has not been formally assessed. Samples were measured over a range of dilutions to ensure accurate quantitation in the high range.

Cross-sectional comparisons of marker levels between groups were made using Mann–Whitney U-test, while differences between baseline and post-cART time points in HIV⁺ MSM were made using Wilcoxon matched pairs signed rank test (GraphPad Prism software, version 6.05). Multilevel modeling was used to estimate latent growth-curve models exploring the subject-specific nature of the association between each marker and time. Latent growth-curve models were also estimated on the natural log of each immunological marker and post-estimation non-linear equations using exponentiated model coefficients were estimated to provide proportional rates of immunological change at specific time-points. Latent growth-curve models comprised two-levels, HIV⁺ individuals at level-2 (i.e., random intercept and coefficient for time) and their marker responses over-time at level-1 (see Supplementary Eq. 1 in Supplementary Material). Latent growth-curve modeling was extended to incorporate more complex bivariate outcome models (i.e., two outcomes), enabling simultaneous estimation of log-marker rates of change and post-estimation inference comparing rates of change between different markers (i.e., modeling the comparative mean per-cent change between two markers over-time, and accounting for the correlation between individuals’ responses to these two markers across time). In these models, typically, random effects for heterogeneity in both participants’ baseline marker levels (i.e., random intercept) and the nature of any change in marker level over time (i.e., random coefficient/slope) were estimated, with an unstructured covariance estimated between each random effect. A limitation of these analyses was that for markers measured using a binomial statistical understanding (i.e., outcomes where cell proportions were determined) the distributional assumptions of the linear mixed models used in longitudinal modeling were not entirely met. Nested model-based likelihood-ratio statistics were used to provide statistical inference for model fit when relaxing model constraints (random effects and the functional form of fixed effects for time). To assess the fit of the estimated latent growth-curve models, diagnostic plots comparing participants observed marker levels with Bayesian model-based (best linear unbiased predictions) predicted levels over time were produced and inspected. Contemporaneous (i.e., both outcome and factor variable responses from the same time-period were regressed) unadjusted longitudinal associations between selected factors and participant CD56^dim FcRγ⁻ NK cell proportions were estimated using multilevel modeling. In these multilevel models, factors were estimated as time-varying fixed effects with a random intercept (level-2) to account for the dependency in the data given an individual’s repeated marker level measurement over-time. Statistical inference was assessed at the 5% level. Stata version 13.1 statistical package (StataCorp LP, College Station, TX, USA) was used for multilevel modeling and bivariate latent growth-curve modeling was undertaken using the user-written Stata program generalized linear latent mixed modeling (gllamm) [(33)].

In many developed countries, the HIV epidemic is concentrated in at-risk populations such as MSM, yet demographic and clinical differences that exist between these populations and the general community [i.e., prevalence of smoking, sexually transmitted infections, HCMV seropositivity (34)] are rarely considered in immunological studies. Given the association between HCMV infection and adaptive-like NK cell expansion in HIV seronegative individuals shown by ourselves and others (3, 22), we first asked whether proportions of FcRγ⁻ NK cells were influenced by MSM status. We compared samples from 14 non-MSM males (median age [IQR] 31.0 [30.0–39.0] years, 78.6% or 11/14 CMV-seropositive), and 14 MSM of a comparable age (35.0 [29.3–43.5] years, 85.7% or 12/14 CMV-seropositive) who were all HIV seronegative. MSM had higher levels of IgG antibodies reactive with the HCMV envelope protein gB (Figure 1A, p = 0.047) but also showed significantly increased proportions of FcRγ⁻ NK cells (median [IQR] 14.4% [4.8–17.8] for MSM vs 4.0% [1.8–6.8] for non-MSM; p = 0.006, Figure 1B). These differences persisted when only CMV seropositive individuals were compared (p = 0.012 and 0.037 for gB antibodies and FcRγ⁻ NK cells, respectively, not shown). These findings confirmed the importance of controlling for MSM sexual exposure and HCMV burden in our subsequent analyses of NK cell immunology in HIV infection.

To assess the impact of untreated HIV infection on FcRγ⁻ NK cell expansion in an appropriately controlled study population, we analyzed baseline samples from cART-naïve HIV⁺ and HIV⁻ MSM of similar age (n = 20 and 15, respectively) from the Melbourne HIV cohort. Demographic characteristics and relevant clinical parameters are detailed in Table 1. Proportions of FcRγ⁻ NK cells were expanded in HIV⁺ MSM, with a median of 28.6% (IQR: 24.6–38.3%) compared to 14.4% (4.8–17.8%) in HIV⁻ MSM (p < 0.0001, Figure 2A). The phenotype of these cells in the two groups was similar; CD56^dim FcRγ⁻ NK cells had very low expression of the natural cytotoxicity receptors NKp30 and NKp46 and heightened expression of the maturation/differentiation marker CD57, which was statistically significant in HIV⁺ individuals; however, there was substantial inter-individual variation in the pattern of CD57 expression on FcRγ⁻ and FcRγ⁺ NK cells (Figure S2A–C in Supplementary Material). To further explore the phenotype of these cells and investigate whether they were the result of proliferative expansion, we analyzed expression of the chemokine receptor CXCR6 [indicative of tissue-homing NK cells (35)], and a proliferation marker (Ki-67) on FcRγ⁻ and FcRγ⁺ CD56^dim NK cells in a subset of HIV⁺ donors. Compared to FcRγ⁺ CD56^dim NK cells, FcRγ⁻ cells had significantly lower expression of CXCR6 (0.7 vs 5.8%; p = 0.004, data not shown) while both populations showed equal, low expression of Ki-67 (0.5 vs 1.0%, p = 0.945, data not shown). These data indicate that the proportion of adaptive-like FcRγ⁻ NK cells is expanded by HIV infection (in addition to the effects of MSM-related factors) and that this cell population does not appear to be the result of heightened proliferation or show increased expression of tissue-homing receptors but may represent a more mature NK cell subset.

Given our findings regarding the influence of MSM-related factors on FcRγ⁻ NK cell expansion, it was important to confirm the effect of HIV on NK cell activation and other immune parameters in an adequately controlled cohort. Compared to HIV⁻ MSM, viremic, cART-naive HIV⁺ MSM had significantly increased levels of NK cell activation as indicated by proportions of either HLA-DR⁺/CD38⁺ (median [IQR] 4.5% [2.5–5.9%] in HIV⁺ MSM vs 2.3% [1.4–3.0%] for HIV⁻ MSM, p = 0.015) or CD69⁺ NK cells (7.4% [4.5–10.8%] vs 3.2% [2.2–6.2%], p = 0.004) (Figures 2B,C, respectively). In addition to the increase in HCMV antibody levels associated with MSM status shown in Figure 1, viremic HIV infection was associated with a further increase in antibody levels to both whole HCMV lysate (p = 0.046, Figure 2D) and the gB antigen (p = 0.011, Figure 2E), while antibody levels to HCMV IE-1 did not differ with HIV status (data not shown).

Analysis of T cell and monocyte activation markers confirmed the well-established effect of viremic HIV infection on heightened CD4⁺ and CD8⁺ T cell activation (as assessed by HLA-DR/CD38 coexpression; p < 0.0001 for both vs HIV⁻ MSM, Figures 2F,G), expansion of inflammatory intermediate CD14⁺⁺CD16⁺ monocytes (p < 0.0001 vs HIV⁻ MSM, Figure 2H), and concomitant reduction in classical CD14⁺⁺CD16⁻ monocyte proportion (p = 0.008, data not shown). Taken together, data from this unique MSM cohort confirm that untreated HIV infection is associated with increased NK cell activation and generalized adaptive and innate immune activation, and indicate that this occurs in addition to the effects related to MSM status shown in Figure 1. Although HCMV seropositivity was near-ubiquitous in this cohort and not significantly different between HIV⁺ and HIV⁻ MSM (Table 1), a sub-analysis of only HCMV⁺ MSM indicated the proportion of activated HLA-DR⁺/CD38⁺, CD69⁺, and FcRγ⁻ NK cells was significantly higher in HIV⁺ vs HIV⁻ MSM (p = 0.032, 0.012, and 0.001, respectively, data not shown), indicating an effect of HIV independent of HCMV serostatus.

We undertook longitudinal analyses of pre- and post-cART samples from HIV⁺ MSM to determine the extent to which cART was able to reverse HIV-associated defects to NK cells as compared to other cellular/immunological compartments. Sixty-five percent, 95%, and 100% of individuals achieved undetectable viral load (<20 copies/mL) after 6, 12, and 24 months of cART, respectively (Table 1), and significant increases in CD4⁺ T cell counts observed at all post-cART time-points (p < 0.05 for all vs baseline, Table 1) confirmed the efficacy of cART in this cohort. The relatively high nadir CD4⁺ T cell count of 385 [329–656] (median [IQR]) and baseline CD4⁺ T cell count of 452 [382–711] cells/μL are typical of a contemporary HIV cohort where individuals initiate cART at higher CD4⁺ T cell counts prior to experiencing significant immunological damage.

Descriptive statistical analysis revealed that cART had no impact on proportions of FcRγ⁻ NK cells, which remained similar to pre-cART levels at 6, 12, and 24 months post-cART initiation (Figure 2A). Levels of activated HLA-DR⁺/CD38⁺ NK cells were also unaltered by cART and remained similar to levels in viremic, cART-naïve individuals at all follow-up time-points (Figure 2B). However, the proportion of NK cells expressing the early activation marker CD69 at 6, 12, and 24 months post-cART time-points were significantly lower than pre-cART levels (p = 0.004, 0.036, and 0.008, respectively; Figure 2C). Viral suppression associated with cART did not alter levels of HCMV antibodies to either whole lysate or gB antigens (Figures 2D,E, respectively). The lack of an effect of cART on NK cell dysfunction and HCMV antibody levels was in contrast to its effect on T cells and monocytes, where HIV-related CD8⁺ T cell activation and intermediate monocyte expansion were reversed to levels observed in uninfected MSM within 6 months of cART initiation (Figures 2G,H, respectively), while CD4⁺ T cell activation was resolved within 24 months of cART (Figure 2F). These descriptive analyses indicate cART is less effective at decreasing activation of NK cells compared to T cells and monocytes and demonstrate the persistence of NK cell dysfunction and elevated HCMV antibodies in HIV⁺ individuals despite 24 months of viral suppression.

To fully and quantitatively model the differential effect of cART on individual immune cell types indicated by the above descriptive analyses, we employed a mixed effects modeling framework to compare the rate of decline of HIV-related NK cell activation with other cellular compartments. This analytical approach has the additional benefit of accounting for the inherent variation in each individual’s initial immunological status and the way they subsequently respond to therapy, permitting a more accurate modeling of the change to specific immunological parameters over time in response to cART than is achievable with simple descriptive statistical analyses.

Latent growth-curve modeling confirmed cART had no significant effect on proportions of activated HLA-DR⁺/CD38⁺ NK cells [Wald χ²(1) = 2.3, p = 0.129] or FcRγ⁻ NK cells [Wald χ²(1) = 2.5, p = 0.115], while the HIV-related increase in activated CD69⁺ NK cell proportion declined significantly over time on cART [Wald χ²(1) = 9.1, p = 0.003, Table 2]. Our modeling also confirmed that T cell activation [Wald χ²(2) = 32.5 and Wald χ²(2) = 68.1 for CD4⁺ and CD8⁺ T cells, respectively, p < 0.001 for both, Table 2] and inflammatory intermediate monocyte subset expansion [Wald χ²(2) = 61.2, p < 0.001] were significantly reduced over time on cART.

To compare the rate at which cART reversed CD69⁺ NK cell, T cell and monocyte activation, the proportion of pre-cART immune activation, which remained after 6 and 12 months on cART, was calculated for each cell type (Table 3). After 12 months of cART, 60% (95% CI:53–67%) of activated CD4⁺ and 30% (21–39%) of activated CD8⁺ T cells remained, indicating a respective 40 and 70% reduction in T cell activation. Similarly, 12 months of cART was associated with a 53% (44–62%) and 85% (54–138%) reversal of the HIV-related alterations to intermediate and classical monocyte subset proportions, respectively. In contrast, reversal of activated CD69⁺ NK cells was substantially slower, with only 20% (13–27%) of HIV-related activation reversed following 12 months of cART.

To determine whether differences in the rates at which cART reversed HIV-related activation of NK cells as compared to T cells and monocytes were significant, we estimated bivariate latent-growth curve models regressing each of the marker outcomes on respective functions of time simultaneously and allowing each individual’s treatment responses in marker levels to correlate. This revealed significant differences in the rate of reversal of CD69⁺ NK cell activation as compared to inflammatory intermediate monocyte expansion [6 months: Wald χ²(1) = 42.6; 12 months: Wald χ²(1) = 28.0; p < 0.001 for both, Figure 3A]. Similarly, the proportion of activated CD69⁺ NK cells declined significantly more slowly after cART initiation than activated (HLA-DR⁺/CD38⁺) CD4⁺ T-cells [6 months: Wald χ²(1) = 13.7; 12 months: Wald χ²(1) = 16.1, p < 0.001 for both, Figure 3B] or CD8⁺ T-cells [6 months: Wald χ²(1) = 58.3; 12 months Wald χ²(1) = 47.8, p < 0.001 for both, Figure 3C]. These analyses indicate that cART has a differential effect on different arms of the immune system, and robustly confirm that NK cell activation resolves more slowly than T cell or monocyte dysfunction following cART initiation.

Given the persistence of both FcRγ⁻ NK cells and elevated HCMV antibody levels in cART-experienced HIV⁺ MSM, and our previous observation of an association between these two parameters in HIV-uninfected individuals (3), we extended the modeling to investigate associations between FcRγ⁻ NK cells and elevated HCMV antibody levels in HIV⁺ MSM. Contemporaneous time-varying associations between individuals’ FcRγ⁻ NK cell levels and other key markers were estimated from longitudinal data obtained up to 12 months post-cART initiation using linear mixed modeling, which permitted the analysis of repeated measures data from the same individuals. This analysis indicated that FcRγ⁻ NK cell levels were not associated with levels of antibody to either HCMV lysate [Wald χ²(1) = 0.23, p = 0.631] or gB [Wald χ²(1) = 0.89, p = 0.345, Table 4]. Significantly, there was no association between the proportions of FcRγ⁻ NK cells and NK cell activation (either HLA-DR⁺/CD38⁺ or CD69⁺ NK cells, p = 0.705 and 0.182, respectively, Table 4), nor with any other cellular or soluble immune activation marker measured. This lack of association was also observed when only baseline samples from cART naïve individuals were analyzed (data not shown). These findings are consistent with our previous cross-sectional study that indicated an association between HCMV antibody levels and adaptive-like FcRγ⁻ NK cells in HIV⁻ but not HIV⁺ individuals (3).

In contrast to levels of FcRγ⁻ NK cells, NK cell activation, measured as either HLA-DR⁺/CD38⁺ or CD69⁺ NK cells, was significantly associated with CD4⁺ T cell activation [Wald χ²(1) = 5.31, p = 0.021 and Wald χ²(1) = 9.55, p = 0.002, respectively] and the proportion of intermediate [Wald χ²(1) = 4.29, p = 0.038 and Wald χ²(1) = 6.02, p = 0.014, respectively] and classical [Wald χ²(1) = 5.71, p = 0.017 and Wald χ²(1) = 4.36, p = 0.037, respectively] monocyte subsets (Table S1 in Supplementary Material). Activated CD69⁺ NK cells were also significantly associated with CD8⁺ T cell activation [Wald χ²(1) = 9.24, p = 0.002] and plasma levels of CXCL10 [Wald χ²(1) = 6.66, p = 0.010], whilst neither HLA-DR⁺/CD38⁺ or CD69⁺ activated NK cells were associated with HCMV antibody levels (p > 0.05 for all). These findings suggest that whilst NK cell activation is associated with other markers of adaptive and innate immune activation in HIV⁺ individuals, the expansion of adaptive-like FcRγ⁻ NK cells is a discrete phenomenon.