This study is part of the HFGP (www.humanfunctionalgenomics.org) (18). Between 14 December 2015 and 6 February 2017, individuals living with HIV were recruited from the HIV clinic of Radboud university medical center. Inclusion criteria were Caucasian ethnicity, age ≥ 18 years, receiving cART > 6 months, and latest HIV-RNA levels ≤200 copies/ml. Exclusion criteria were: signs of acute or opportunistic infections, antibiotic use <1 month prior to study visit, and active hepatitis B/C. The control population consisted of 56 healthy adult individuals (56P cohort), who did not suffer from any acute or chronic conditions and who were longitudinally sampled four times in three-month intervals. Inclusion, sampling and sample processing of both cohorts were conducted simultaneously and by the same personnel. The 56P participants were selected as a subset of a larger cohort of 534 healthy individuals (500FG) which was phenotypically assessed two years earlier according to the same methods (19, 21). Differences in cell-cell associations were compared between 200 HIV and 500FG cohort. The reason is that the larger sample size of this control cohort (n=534 vs n=56) improved statistical power, whilst batch effects between PLHIV and 500FG were deemed of less significance when comparing within-group correlations between cohorts. General information from all participants was recorded in the Castor Electronic Data Capture program (Castor EDC, CIWIT B.V., Amsterdam, The Netherlands), using questionnaires on socio-demographic information, lifestyle and life-events. Clinical data were extracted from medical files and the ‘Stichting HIV Monitoring’ registry (Amsterdam, The Netherlands).

The study protocols were approved by the Medical Ethical Review Committee region Arnhem-Nijmegen (ref. 42561.091.122) and conducted in accordance with the principles of the Declaration of Helsinki. All study participants provided written informed consent.

Venous blood was collected in sterile 10 ml EDTA BD Vacutainer^® tubes between 8 and 11 am and processed within 1-4 hours. Cell counts were determined using a Sysmex XN-450 automated hematology analyzer (Sysmex Corporation, Kobe, Japan). Erythrocyte-lysed whole blood samples were obtained after 10 minutes incubation of 1.5 ml EDTA-anticoagulated blood with lysis buffer containing 3.0 M NH₄Cl, 0.2 M KHCO₃ and 2 mM Na₄EDTA. The remaining WBC were washed twice, by adding 25 ml phosphate-buffered saline 1x (PBS, Braun, Melsungen, Germany) and centrifuging at 452 x g for 5 min at room temperature. Before staining, cells were resuspended in 300 μl of PBS enriched with 0.2% bovine serum albumin (BSA, Sigma-Aldrich, Zwijndrecht, Netherlands). Isolation of PBMCs was performed by density centrifugation of EDTA-anticoagulated blood diluted 1:1 in pyrogen-free PBS over Ficoll-Paque (VWR, Amsterdam, The Netherlands) as described previously (22). Cells were adjusted to 5.0 x 10⁶ cells/ml.

Supplementary Table 1 summarizes the antibody clones and the fluorochrome conjugates used for the different panel fluorescent staining mixes. Staining was performed on 100 μl/well erythrocyte-lysed blood (panel 1-3,5) or 0.5 x 10⁶ cells/well freshly isolated PBMCs (panel 4). Cells were stained according to previously described procedures (see also Supplementary Methods ) (19).

Samples were measured on a 10-color Navios flow cytometer (Beckman Coulter, Fullerton, CA, USA), equipped with three solid-state lasers (488 nm, 638 nm, and 405 nm) (19). Gating was conducted manually and verified by two independent specialists to prevent gating errors. Samples were analyzed within 4-5 hr after blood collection, using five distinct and complementary 10-color antibody panels: 1. general; 2. T cell; 3. B cell; 4. intracellular T cell/Treg; 5. chemokine receptors (CCR). Staining and gating strategies can be found in Supplementary Figure 1 . Flow cytometry data were analyzed using Kaluza software version 1.3. In our analyses, we focused on a set of 108 manually annotated WBC subsets based on the original 500FG study (panel 1-4) (19), with the addition of a fifth panel in which we classified monocytes, CD4+ memory and regulatory T cells, and CD8+ cells according to their expression of the CXCR3, CCR4, and CCR6 (panel 5).

Freshly isolated PBMCs were incubated with different stimuli ( Supplementary Table 2 ) including bacterial (Staphylococcus aureus, M. tuberculosis, Streptococcus pneumoniae [S. pneumoniae]), fungal (Cryptococcus gattii, Candida albicans [C. albicans] hyphae and yeast) and other relevant antigens (Imiquimod, TLR7 ligand), in round-bottom 96-well plates (Greiner Bio-One, Frickenhausen, Germany) with 0.5 x 10⁶ cells/well at 37°C and 5% CO₂ in the presence of 10% human pooled serum for seven days. Supernatants were stored at -20°C. Levels of the lymphocyte-derived cytokines IL-17, IL-22, and interferon (IFN)-γ were measured in the supernatants (PeliKine Compact or Duoset ELISA, R&D Systems).

Serum levels of immunoglobulin (Ig)M and IgG were measured by immunonephelometry using a Beckman Coulter Imager and Beckman Coulter reagents. Measurements were standardized using certified European reference material 470 (ERM^®-DA470). CMV IgG antibodies were measured in serum using ELISA (Genway Biotech Inc., San Diego, CA, USA) according to the manufacturer’s instructions. Markers of systemic inflammation, high-sensitive C-reactive protein (hsCRP) and soluble CD14 (sCD14), and microbial translocation, intestinal fatty acid binding protein (IFABP), were measured by ELISA (Quantikine, R&D Systems) according to the manufacturer’s instructions. IL-6 was measured using a SimplePlex Cartridge (Protein Simple, San Jose, CA, USA). Circulating plasma CCL20, IL-17A, and IFN-γ were measured using the commercially available Olink Proteomics AB Inflammation Panel as described previously (23, 24).

The HIV reservoir was assessed by analyzing CD4+ cell-associated HIV-1 DNA (CA-DNA) and CA-RNA. In virally suppressed patients, the CA-DNA roughly equals the integrated HIV-1 DNA, being replication competent or not (25), while CA-RNA is associated with recent HIV-1 transcriptional activity and serves as a proxy for the active proviral reservoir (26). CA-DNA and CA-RNA were measured in triplicate by droplet digital PCR (ddPCR – Bio-Rad) in CD4+ T cells isolated using EasySep Human CD4+ T Cell Isolation Kit (Stemcell technologies, Vancouver, Canada) as described previously (27). CA-DNA was extracted by the DNeasy Blood & Tissue kit (Qiagen, Hilden, Germany) according to the manufacturer’s protocol with the addition of 75µl elution buffer on the column heated at 56°C for 10 minutes. CA-RNA was extracted using the Innuprep RNA kit (Westburg, Leusden, The Netherlands) with 30µl elution buffer. Total RNA was reversely transcribed to cDNA by qScript cDNA SuperMix according to manufacturer’s protocol (Quantabio, Beverly, MA, USA). CA-DNA was normalized by measuring the reference gene RPP30 ( Supplementary Table 3 and Supplementary Methods ) in duplicate by ddPCR and expressed per million CD4+ T cells. CA-RNA was normalized using three reference genes per patient, (B2M, ACTB and GADPH), which were measured with LightCycler 480 SYBR Green I Master mix. HIV-1 RNA copies were divided by the geometric mean of the reference genes and expressed per million PBMCs. ddPCR data analysis was performed using the ddpcRquant tool with standard settings for thresholding and absolute quantification (28).

A detailed description of the statistical methods can be found in the Supplementary Methods . Given the impact of cohort differences in the absolute cell numbers of main WBC types (e.g. CD4+) on the numbers of their subsets (e.g. CD4+ regulatory T cells [Tregs]), results were primarily reported as WBC percentages, unless stated otherwise. WBC percentages were calculated by dividing the cell count of each subpopulation by its respective parent (one level up) or, where relevant, grandparent (two levels up; Supplementary Table 4 ). Data curation of cytokine, soluble marker, and immunoglobulin data was done according to previous methods (21). For comparisons between cohorts, we used all four longitudinally collected data points from the 56P cohort as independent measurements. Because of the known seasonality effects on WBC, this approach was considered preferable over selecting one of the data collection points or summarizing the data points. Data were normalized using an inverse rank transformation algorithm.

Comparisons in baseline characteristics between groups were made using Student’s t-test (or Mann-Whitney U test) for continuous variables, and Pearson’s Chi-square test (or Fisher’s exact) for categorical variables. Linear regression was used to compare WBC between cohorts, and to calculate associations between WBC and clinical and virological factors. All analyses were corrected for age, sex, time since January 2015 and seasonal effects. Cytokine analyses between cohorts were also corrected for CD4+ and CD8+ T cell percentages. Spearman correlations were used as the distance metric for unsupervised hierarchical WBC count clustering. Two-sided FDR-corrected p-values < 0.05 were considered statistically significant (29). Effect sizes are reported as Spearman’s Rho or standardized beta coefficients (β). Data were analyzed using the statistical programming language R (version 3.4.3, R Core Team, 2012).

Data from 211 PLHIV and 56 controls were analyzed in this study ( Table 1 ). PLHIV were older (median [IQR] 52.5 [46.2 – 59.4] vs 30.0 [25.1 – 52.2] years, p<0.0001) and more often male (193/211 [91.5%] vs 34/56 [60.7%], p<0.0001) than controls. PLHIV had received cART for a median of 6.6 (4.2–11.9) years and 205/210 (98%) had plasma HIV-1 RNA <50 copies/mL at time of study visit. Analyzing the HIV reservoir, we found detectable levels of total CA-DNA in 207/208 (99.5%) and CA-RNA in 210/210 (100%); CA-DNA and CA-RNA levels were highly correlated ( Supplementary Figure 2A ; R=0.68, p<2.2·10^-16). In general, PLHIV with higher CA-DNA and CA-RNA levels were older, had been living with and treated for HIV for a longer period, and had lower nadir CD4+ T cell counts ( Supplementary Figure 2A ). We observed no significant differences in CA-RNA and CA-DNA between PLHIV with plasma HIV-1 RNA <50 copies/mL (205/210 [98%]) and those with plasma HIV-1 RNA 50-200 copies/mL (5/210 [2%]; Supplementary Figure 2B ). PLHIV with at least once a viral load of >50 copies/mL during year prior to study visit (23/210 [11.0%]) had higher levels of CA-RNA (p=0.0077), but no differences in CA-DNA (p=0.076; Supplementary Figure 2B ).

We analyzed 108 WBC populations, including 93/107 (87%) B and T cell subsets and 14/107 (13%) innate cell subsets (neutrophils, monocytes, natural killer [NK] and natural killer T cells [NKT]). Figure 1 and Supplementary Figure 3 show the differences in WBC composition between PLHIV and controls. An overview of median (IQR) WBC percentages and numbers in PLHIV and controls can be found in Supplementary Table 5 .

Principal component analysis (PCA) of all WBC populations only explained a limited amount of the total variance. Still, the PCA plot showed clear differences in clustering between PLHIV and controls ( Figure 1A ). As expected, PLHIV exhibited an expansion of CD8+ and a reduction of CD4+ T cell numbers and percentages ( Figure 1B and Supplementary Table 5 ). Functionally, CD4+ T cells comprise a diverse population of cells. CD4+ T helper (Th) cells fulfill essential roles for viral (Th1, CCR6-CXCR3+CCR4-), parasitic (Th2, CCR6-CXCR3-CCR4+), and mucosal (Th17, CCR6⁺CXCR3-CCR4+) immunity (30). In addition, CD4+ Tregs are essential for controlling inflammation (31). Despite reduced CD4+ T cell counts, numbers of all Th cell subsets (CD4+ CD45RA- CD25-) were markedly increased in PLHIV compared to controls ( Figure 1B and Supplementary Table 5 ). Within the Th pool, Th2 percentages were reduced in PLHIV compared to controls, whereas Th1 percentages did not differ. Remarkably, Th17 percentages and numbers were increased in PLHIV ( Figures 1B, C and Supplementary Table 5 ). While Treg percentages (of CD4+ cells), including highly suppressive Tregs co-expressing the transcription factors FoxP3 and Helios, were also increased in PLHIV, absolute Treg numbers were reduced ( Figures 1B, C and Supplementary Table 5 ). This relative increase of Tregs may result from a loss of other CD4+ subsets (32). Among Tregs, we found no differences in the percentage of activated (HLA-DR+) and effector Tregs (CD45RA-). The relationship between pro-inflammatory Th17 cells and Treg must remain balanced to preserve functional immunity. Altered ratios have been described in untreated HIV (lower Th17/Treg), autoimmune disease and cancer (higher Th17/Treg) (9, 33, 34). Here, we found increased Th17/Treg ratios among virally suppressed PLHIV, irrespective of the Treg identification marker used ( Figure 1C and Supplementary Figure 4 ). Furthermore, out of the Treg population, the percentage of Th17-like CCR6+ Tregs was increased in PLHIV, further contributing to a pro-inflammatory state. Developmentally, T cells evolve from naive T cells to antigen experienced central memory (CM), effector memory (EM), and effector cells (35, 36). HIV not only differentially affects functional subpopulations, but also disrupts these developmental stages. While naïve CD4⁺ and CD8⁺ T cells were reduced, memory and effector cells were expanded in PLHIV ( Figure 1B and Supplementary Table 5 ). Likewise, we found higher percentages of proliferating (Ki67+) CD4+ and CD8+ T cells in PLHIV, indicating increased cell turnover. These results suggest a shift from naive cells towards terminally differentiated cells in HIV, even if viral replication is under control (37), which cannot be explained by age, sex, or season as we corrected our models for these factors.

Changes in the B cell compartment have also been described in PLHIV, including loss of CD27+ memory B cells (38, 39). Furthermore, viremia and low CD4+ T cell counts have been associated with the expansion of terminally differentiated B cells and immature B cells respectively (38, 39). In our study, we observed clear differences in B cell development in PLHIV illustrated by increased percentages of naïve B cells (IgD+IgM+CD27-) and reduced percentages of memory B cells (IgD+IgM+CD27+), transitional B cells (CD24++CD38++) and natural effector B cells (CD24+CD38+IgD+IgM+). In addition, the number of plasmablasts (IgD- IgM- CD38++) was increased in PLHIV, ( Figures 1B, C and Supplementary Table 5 ). Adequate B cell maturation further requires optimal communication between B and T cells. We therefore sought to identify differences in WBC co-regulation between PLHIV and HIV-uninfected individuals by comparing cell-cell associations between PLHIV and participants from the 500FG cohort. Details of this 500FG cohort have been reported previously (500FG, n=534) (19). Using the same immunophenotyping techniques, we observed weaker correlations between naïve, CM and EM CD4+ T cells and several B cell subpopulations (including class-switch memory B cells) within PLHIV than in participants of the 500FG cohort, suggesting altered B/T cell interactions ( Figure 2 ). These differences were not attributable to the influence of age, sex, or season as these factors were regressed out of the analysis.

Apart from changes in the adaptive cell compartment, we observed clear changes in the innate WBC compartment in PLHIV. Monocyte and NKT cell numbers and percentages were increased, whereas NK cell numbers and percentages were reduced, specifically the cytokine-producing NK bright cells ( Figure 1B and Supplementary Table 5 ) Together, these data show a widespread functional and developmental dysregulation of the immune system in virally suppressed PLHIV. This dysregulation encompasses both the innate and the adaptive compartment and results in a pro-inflammatory immune environment with an expansion of monocytes, pro-inflammatory Th and effector T cells, and dysregulated B cell memory responses.

Apart from the effects of HIV, demographic and lifestyle-related factors may influence the composition of the circulating WBC populations. Indeed, we found that older age was associated with an increase of innate immune cells and differential effects on B and T cell populations, for example with lower percentages of naïve T cells (CD4+ T cells β=-0.29, p=0.00020; CD8+ T cells β=-0.29, p=0.00020) and B memory cells (β=-0.39, p=1.5·10^-7) and higher percentages of memory T cells (e.g. CD8+ CM β=0.22, p=0.00064) and mature naïve B cells in PLHIV (β=0.25, p=0.00087, Figure 3 and Supplementary Table 6 ). Sex-dependent influences of the WBC composition, reflected by more effector and EM cells and fewer naïve (B and T) cells in males in HIV infection, resembled those observed previously observed in healthy individuals (21, 40).

Lifestyle risk behaviors such as smoking [63/211 (29.9%)], heavy drinking [28/211 (13.3%)], and regular drug use [61/211 (28.9%)] were highly prevalent in PLHIV ( Table 1 ). We found that packyears (reflecting total tobacco exposure) were associated with higher frequencies of neutrophils (β=0.22, p=0.033), Treg (β=0.22, p=0.033), CD8+ subsets (e.g. CCR6-CXCR3-CCR4+ β=0.24, p=0.025), and class switched memory B cells (β=0.20, p=0.042, Supplementary Table 6 ). Active smoking correlated with higher percentages of Th17 (β=0.21, p=0.031) and CCR6+ Tregs (β=0.26, p=0.0028, Figure 3 and Supplementary Table 6 ). Neither heavy drinking, nor regular drug use affected the WBC composition.

We further assessed the effects of relevant clinical factors on the WBC composition in PLHIV, by exploring associations with the history of immune suppression and treatment-related factors. We found that nadir CD4+ cell counts were closely associated with both B and T cell percentages in PLHIV ( Figure 3 and Supplementary Table 6 ). For example, higher counts were associated with higher percentages of naïve CD4+ T cells (β=0.45, p=1.2·10^-9) and memory B cells (β=0.18, p=0.034). In contrast, we observed no effects of the duration of HIV infection or cART (regimen) on the WBC composition ( Supplementary Tables 6 ).

First, we explored the relationship with markers of the HIV reservoir. CA-DNA was negatively associated with CD4+ T cell percentages (β=-0.33, p=9.8·10^-5) and positively with CD8+ T cell percentages (β=0.34, p=7.0·10^-5). Within the CD4+ T cell pool, CA-DNA correlated with higher percentages of CM and Th17 cells (β=0.21, p=0.026 and β=0.24, p=0.0081 respectively; Figure 4 and Supplementary Table 7 ) Higher CA-DNA was also associated with more CD4+ T cell proliferation (Ki67+, β=0.35, p=3.3·10^-5) and Treg activation (HLA-DR+, β=0.23, p=0.0081). We found similar associations between T cells and CA-RNA, whereas we found no relation between T cells and the relative HIV transcription level (CA-RNA/CA-DNA ratios; Supplementary Table 7 ). Lastly, we observed several associations between CA-DNA and B cells: higher CA-DNA levels correlated with reduced percentages of IgD+ only memory B cells (β=-0.27, p=0.00068) and, albeit non-significantly, with reduced percentages of natural effector B cells (CD24+CD38+IgD+IgM+, β=-0.18, p=0.056), and class unswitched memory B cells (IgD+IgM+CD27+, β=-0.18, p=0.053). Exclusion of PLHIV with HIV-RNA 50-200 copies/mL at time of study visit (5/210 [2%]) did not change the main conclusions of our paper ( Supplementary Tables 8, 9 ).

Second, CMV co-infection may contribute to immune dysregulation in both treated and untreated PLHIV. As CMV is known to affect WBC populations in healthy controls (41), we explored the association of CMV serostatus with WBC composition. 198 of the PLHIV (93.8%) were seropositive for CMV. In line with findings in healthy individuals (41), CMV seropositivity correlated with higher percentages of effector, EM T cells (e.g. with CD8+ effector cells β=0.43, p=1.7·10^-9 and CD4+ EM cells β=0.30, p=0.00011; Figure 4 and Supplementary Table 7 ) and NKT cells (β=0.29, p=0.00029), yet with lower percentages of Th17 cells (β=-0.26, p=0.00065). We observed no associations between B cell subsets and CMV.

Our main findings include a loss of naïve T cells and an expansion of Th17 in PLHIV compared to healthy controls, which related with lower nadir CD4+ T cells counts and higher levels of CA-DNA. To further assess the potential underlying mechanisms, we explored the relationship between these WBC subsets, markers of chronic inflammation, and markers of microbial translocation. As discussed above, naïve T cells are able to differentiate into different Th subsets depending on the cytokine environment (42). We previously showed that PLHIV in our cohort exhibited a pro-inflammatory profile with increased levels of hsCRP (p=0.00022) and sCD14 (a marker of monocyte activation, p=0.0025; Figure 5A ) as well as markedly elevated monocyte-derived cytokine responses, particularly IL-1β (20). Such a pro-inflammatory cytokine environment may push the differentiation of naive CD4+ cells into Th17 cells (42). Indeed, we observed positive associations between Th17 percentages and circulating IL-6 (β=0.21, p=0.014) and sCD14 (β =0.17, p=0.035; Figures 5B, C ). Notably, Th17 cells fulfill an essential role in mucosal defense and gut Th17 cells are known to be severely depleted during acute HIV (9). To test whether gut integrity might have been compromised in our cohort, we measured levels of the microbial translocation marker IFABP and found increased levels in PLHIV compared to healthy controls (p=7.6·10^-5; Figure 5D ), suggesting ongoing microbial translocation in chronic treated HIV, especially in those with higher CA-RNA and CA-DNA (CA-RNA β =0.18, p=0.0091 and CA-DNA β=0.21, p=0.0025, Figure 5E ). Higher levels of IFABP were associated with an expansion of peripheral blood Th17 (β=0.17, p0.043; Figure 5B ). Homing of Th17 to the gut (and other tissues, such as the skin) is directed by the chemokine CCL20, which is produced by tissue and immune cells (neutrophils and monocytes) and binds uniquely to the CCR6+ receptor (43). Correspondingly, we found strong associations between CCL20 and Th17 percentages (β=0.22, p=0.0037), CCL20 and Th17/Treg (β=0.26, p=0.00081 for Th17/FoxP3+ Treg, β=0.27, p=0.00081 for Th17/CD127low Treg, and β=0.19, p=0.013 for Th17/CD25++ Treg), CCL20 and circulating IL-17A levels (β=0.47, p=1.2·10^-11) and CCL20 and IFABP (β=0.24, p=0.0011; Figure 5F ). Together, these results suggest that the pro-inflammatory environment in chronic HIV may promote the differentiation of circulating naïve CD4+ T cells into Th17 cells and that these changes may be associated with changes in gut permeability and gut homing.

As our results indicated significant changes in circulating immune cell populations in PLHIV, we next analyzed the possible functional consequences. First, we measured ex vivo cytokine responses of PBMCs after stimulation with different stimuli. We found strong correlations between NK and T-cell percentages and ex vivo IFN-γ, IL-17, and IL-22 responses ( Figure 6A and Supplementary Table 10 ): IFN-γ responses correlated with percentages of NK dim, CD4+ and CD8+ EM cells, whereas IL-22 responses correlated with CCR6+ CM CD4+ cell percentages. As expected, Th17 percentages were associated with increased circulating IL-17A (β=0.16, p=0.029) and ex vivo IL-17 responses to C. albicans (β=0.19, p=0.047) and with reduced IFN-γ responses ( Figures 6A, E and Supplementary Table 10 ). Despite the expansion of Th17 cells among PLHIV compared to controls, ex vivo responses of IL-17 or IL-22 did not differ, suggesting that the functional capacity of these cells may be compromised. In contrast, IFN-γ production upon stimulation with C. albicans hyphae (p=0.012) and M. tuberculosis (p=0.0025) was reduced in PLHIV ( Figure 6B ). Among PLHIV, lower ex vivo IFN-γ production to stimulation with Imiquimod (β =-0.21, p=0.0020) and S. pneumoniae (β=-0.22, p=0.0011) and lower circulating IFN-γ concentrations (β=-0.18, p=0.014) were associated with higher CA-DNA levels ( Figures 6C, D ). No associations were found between ex vivo cytokine production and CMV seropositivity.

Second, we measured serum immunoglobulins and found significant correlations between serum IgM and IgM+ B cell populations ( Figure 6F ), but no cohort differences in IgM levels [median (IQR) 0.76 (0.56 – 1.05) g/L in PLHIV versus 0.82 (0.65 – 1.21) g/L in controls, p= 0.22] or IgG levels [10.03 (8.34 - 11.59) g/L in PLHIV versus 9.10 (7.59 – 11.38) g/L in controls, p= 0.12]. These results indicate that while some of the alterations in adaptive immune function may be reversed by long-term cART (44, 45), others, such as impaired IFN-γ responses, remain.