The study population was enrolled between 2013 and 2014 at the Manhiça District Hospital (MDH) in the district of Manhiça, Southern Mozambique. The present analysis is a sub-study of a prospective cohort of primary HIV-infected adults enrolled and followed up for 12 months in the gastrointestinal biomarkers in acute-HIV infected Mozambican adults study (GAMA) (34).

This study was approved by local institutional review boards at Barcelona Clinic Hospital (2011/6264) and by the Ministry of Health of Mozambique (461/CNBS/12). All methods were carried out in accordance with the relevant guidelines and regulations. Written informed consent was obtained from patients prior to participation.

All study participants were over 18 years of age and residents of the established District Surveillance System study area. During the screening, subjects presenting to the outpatient clinic of MDH for non-specific febrile symptoms or voluntary HIV counseling and testing (VCT) were included in the PHI group if they were negative or indeterminate for rapid test serology and HIV-RNA positive for pooled-viral load (VL) testing (n = 85). A control population was established by random selection among HIV-uninfected and individuals were invited to attend a study visit 1 month after the screening date (n = 58). PHI individuals were followed up at seven consecutive visits 1, 2, 3, 4, 6, 9, and 12 months after the screening visit. Technical information and procedures regarding HIV diagnosis and monitoring, as well as the screening profile have been previously described (34). Additionally, adults with documented HIV diagnosis ≥12 months earlier attending routine scheduled outpatient visits for clinical management of HIV/AIDS at the MDH were enrolled as CHI patients. CHI patients were included in the CHI-naïve or the CHI-ART, depending whether they had previously initiated treatment according to the current national guidelines (patients with a CD4 T-cell count ≤ 350 cells/mm³ or presenting and AIDS-associated disease).

After screening, demographic and clinical data was collected, medical consultation and HIV counseling was provided, and blood and stool samples were collected at all the study visits. Determination of CD4 and CD8 T-cell counts was performed on fresh whole blood in a single platform system using Trucount tubes and a FACScalibur flow cytometer. Peripheral blood mononuclear cells (PBMCs) were isolated by Ficoll density gradients and immediately stored in liquid nitrogen. VL determination was performed in plasma samples as previously described (34). Microbiological evaluation was performed in plasma and stool samples, testing for the most prevalent infections in the area including malaria, hepatitis B virus, syphilis and gastrointestinal protozoa, bacteria, and parasites (File S1 in Supplementary Material).

HIV-specific serology was subsequently performed on frozen plasma samples by western blot. VL and WB results at screening visit were employed to categorize individuals into Fiebig stages I–III (VL positive, WB negative), IV (VL positive, WB indeterminate), V (VL positive, WB positive with p31 band negative), and VI (VL positive, WB positive with p31 band positive) as described in previous work (34, 35). In order to approximate similar time since infection for the PHI individuals at every study visit, visits from individuals categorized in Fiebig stage V and VI at screening were moved 1 and 2 months forward, respectively, according to estimated days post infection previously described (35–37). After adjustment by Fiebig, new visits were grouped into M1, M2, M3, M4, M5, M6, M7–8, M9–11, and M12–15 according to estimated months since infection (Figure S1 in Supplementary Material). PBMCs or clinical data were not collected at the screening visit, so all the individuals included in M1 will not have this information, as well those individuals categorized in Fiebig stage V and VI at M2 and M3, respectively.

Multiplex biomarker profiling was performed for a total of 61 immune response biomarkers in plasma samples. Determinations were performed by ELISA commercial assays or Luminex technology as previously described (34–38). From the resulting 49 quantifiable biomarkers, the difference in the median levels between PHI and non-HIV-infected individuals with reported fever was highly significant for 13 soluble biomarkers (P < 0.001) (38). In order to represent the dynamics along the first year of infection, these 13 soluble biomarkers were grouped into two main pathways depending on their main function: (1) lymphocyte and monocyte function and (2) inflammation, intestinal damage, cell death, and proliferation.

Cryopreserved PBMCs were thawed at 37°C, washed in RPMI/60% and RPMI/20% of fetal bovine serum (FBS), and incubated for 1 h at 37°C in RPMI/10% FBS. PBMCs were then stained with the Fixable Viability Stain-FVS780r (APC-H7 detect, BD Biosciences) for 15 min. After PBMCs washing in PBS/1% FBS, cells were plotted to a U-bottom 96-well plate at a density of 1.5 millions/well and stained with selected 14-color panel including CD3-BV605 (Clone SK7), CD4-FITC (Clone RPA-T4), CD8-V500 (Clone SK1), CD45RA-Alexa Fluor^®700 (Clone HI100), CD197-PE-CF594 (Clone 150503), CD57-APC (Clone NK-1), CD279-BV421 (Clone EH12.1), HLA-DR-BV650 (Clone G46-6), CD38-PerCp-Cy5.5 (Clone HIT2), CD25-PE (Clone M-A251), CD127-BV786 (Clone HIL-7R-M21), CD196-BV711 (Clone 11A9), and CD183-PE-Cy7 (Clone 1C6/CXCR3) (from BD Biosciences) for 15 min. After washing twice in PBS/1% FBS, cells were fixed in PBS/1% formaldehyde, acquired in a BD LSRFortessa cytometer using a plate HTS loader (BD Biosciences) and analyzed with FlowJo software (Tree Star). Gating strategy is described in Figure S2 in Supplementary Material. Lymphocyte gate was defined manually by morphological parameters excluding nonviable cells and singlets. Median of viability + lymphocytes was 2.5% [IQR 1.6–4.1], as an estimation of death cells per sample. Subsets were identified as CD3+ cells and gated as CD4+CD8− or CD8+CD4−, while double-positive cells and double-negative cells were excluded from the analysis. T-cell maturation stage was analyzed automatically using R software for CD45RA and CD197/CCR7 expression to define naive (TN, CD45RA+CCR7+), central memory (TCM, CD45RA−CCR7+), effector memory (TEM, CD45RA−CCR7−) and effector memory RA+ cells (TEMRA, CD45RA+CCR7−). T-cell subsets were also analyzed automatically for the expression of HLA-DR and CD38 to define activated cells (HLA-DR+ and CD38+), CD279/PD-1 to define exhausted cells (CD279+) and CD57 to define immunosenescent cells (CD57+). CD4+ T-cells were subsequently analyzed manually for the expression of CD25 and CD27 to define Treg subset (CD25+++CD27−) and automatically for the expression of CD183 and CD196 to define Th1/Th17 cells (CD183+CD196+).

Group comparisons were performed using the Fisher’s exact test for categorical variables and the non-parametric Kruskal–Wallis test for continue variables. Spearman’s correlation was used to assess the strength of relationship between continuous variables and multiple testing was further adjusted by false discovery rate. Individual comparisons between the different groups were performed using post hoc pairwise comparisons with the Tukey and Kramer (Nemenyi) test. Relative changes (Z-score) with respect to the HIV-uninfected group (in the case of VL the Z-score was calculated relative to the CHI-naïve group) have been represented by a transformation of the fitted longitudinal models by subtracting the mean and dividing by the standard deviation of HIV-non infected distribution, after a logarithmic transformation in the cases where it was required for normal distribution. Longitudinal behavior for analytes and immunological variables were modeled by fitting smoothing-splines mixed-effects models using the “sme” package of R. To infer if there was a significant association of selected biomarkers with the time variable, polynomial time effects approximation until third degree were fitted using linear mixed-effects regression models. Best model was selected based on likelihood ratio tests under maximum likelihood models estimations. A two-phase exponential decay regression model was employed in the case of VL modeling. Statistical analyses were performed using R-3.3.1 and Stata14 software.

In the context of this study, we recruited 57 PHI patients identified during the screening process, as described in Section “Materials and Methods.” From these, 40 individuals attended a follow-up visit 1 month later and 26, 21, 14, 14, 13, and 11 of these patients continued visits at 2, 3, 4, 6, 9, and 12 months after screening, respectively. There was no significant difference in age, gender balance, or HIV-RNA VL between PHI patients who returned for enrollment and those who were lost to follow-up (34). HIV-uninfected individuals were randomly selected from screened individuals and 58 subjects representing the control population attended a visit 1 month later. During the cross-sectional recruitment of CHI individuals, 26 patients were included in the ART-naïve group and 30 patients in the ART group. The demographic and clinical characteristics of the 40 PHI individuals who started follow-up, the HIV-uninfected and the CHI groups, are summarized in Table 1. Significant differences were found for age and body mass index (BMI, P < 0.0001) but no differences were found for the clinical variables between the study groups (P > 0.05).

Among the 57 PHI identified, 28, 5, 7, and 17 were categorized into Fiebig I-III, Fiebig IV, Fiebig V, and Fiebig VI stages, respectively, and were adjusted for time since infection as described in Section “Materials and Methods” (Figure S1 in Supplementary Material). After categorizing by Fiebig stage, and adjusting for time since infection, median VL in the PHI group was 6.9 RNA Log10 copies/mL (IQR 6.2–7.5) at month one (M1) and significantly decreased to 5.1 RNA Log10 copies/mL (IQR 4.7–5.6) at month 2 postinfection (M2), (P = 0.0001, Figure 1A). In the CHI-naïve patients, median VL was 4.5 RNA Log10 copies/mL (IQR 3.9–4.9).

In order to evaluate the dynamics of the different parameters along the first year of HIV infection, two approaches based on non-parametric modeling and linear regression modeling were performed as described in Section “Materials and Methods.” Longitudinal analysis confirmed the rapid decrease in VL with either a non-parametric models or a nadir VL set point. A biphasic exponential decay model revealed a second phase of VL decay with a lower but significant slope until 6 months after infection (M6, Figure 1B).

Regarding the dynamics of CD4 and CD8 T cells from M2, we observed that at M2 median CD4 T-cell count in PHI individuals was significantly lower than in the HIV-uninfected group, 565 (IQR 387–675) vs. 955 (IQR 773–1,149) cells/mm³, respectively (P = 0.0001, Figure 1C). CD4 T cells also showed an initial decrease that stabilized at months 5–6 postinfection (M5–6) in non-parametric longitudinal analysis, while significant linear decay was observed overtime in the regression model (P = 0.033 for the slope, Figure 1D). Median CD8 T-cell count was significantly higher in PHI at M2 than in HIV-uninfected controls, 1,175 (IQR 771–1,683) vs. 591 cells/mm³ (IQR 417–746), respectively (P = 0.0001, Figure 1E). Both longitudinal models show that the initial increase in CD8 T cells is followed by a significant decay that also stabilized at M5-6, remaining stable and high until CHI (P = 0.002, Figure 1F). In the CHI-naïve group, median CD4 T-cell and CD8 T-cell count were 595 (IQR 466–729) and 1,029 (IQR 685–1,562), respectively, while in the CHI-ART group, median CD4 T-cell and CD8 T-cell count were 474 (IQR 377–590) and 830 (IQR 617–1,061), respectively.

The frequency of functionally distinct CD4 T cells was analyzed by the cell surface expression of CD127 and CD25 (for Treg) and CD183 (CXCR3) and CD196 (CCR6) as described in Section “Materials and Methods.” This latter combination identifies Th1Th17 cells as CD183+CD196+, while CD183+CD196− cells are mostly Th1 and CD183− CD196+cells contain the Th17 population (Figure S2 in Supplementary Material) (19). No major changes were observed in CD4 CD183+CD196− or CD183−CD196+ cells during PHI (data not shown). However, the frequency of CD183+CD196+ (Th1Th17 cells) cells in PBMCs significantly decayed overtime until 7–8 months postinfection (M7–M8, P = 0.0127, Figures 2A,B). The percentage of activation in Th1Th17 cells at M2 (as measure by CD38 and HLA-DR co-expression) was significantly increased compared to HIV-uninfected (P < 0.0001) and continued to increase along the first year postinfection (data not shown). Looking at the maturation stage of the Th1Th17 cells, a significant increase in the frequency of naive (TN, P = 0.0002) and a significant decrease in the frequency of effector memory (TEM, P = 0.0007) was observed at M2 compared to HIV-uninfected. Since the definition of Th1Th17 cells involves cell surface expression of CXCR3, the receptor for IP-10, we assessed the relationship between CXCR3 expression and IP-10 levels. Although a significant negative correlation was observed between IP-10 plasma levels and circulating CD4 T cells at M2 (rho = −0.49, P = 0.0282); such association was positive and borderline significant between plasma IP-10 levels and CXCR3 + CD4 T cell frequencies (rho = 0.43, P = 0.0574), and no association was observed between plasma IP-10 and the frequency of Th1Th17 CD4 T cells. This fact could be explained because the kinetics of CXCR3+ and Th1Th17 CD4 T cells (CXCR3+CCR6+) are different, considering that expression of CCR6 has been reported to increase susceptibility to HIV infection (39).

No significant oscillations were observed during PHI for the CD4 Tregs (Figure 2C), but comparing to non-HIV infected individuals, CD4 Treg frequency was significantly higher in the CHI-ART group (P = 0.0267). Consistently, regression models showed no significant difference of Treg levels during the first year postinfection (Figure 2D).

Dynamics of the CD4 and CD8 T-cell maturation subsets showed a different profile. Although differences were not significant, the CD4 T-cell compartment showed a prompt increase in the resting phenotypes (TN and central memory T cells [TCM]) and a decrease in the effector phenotypes [TEM and effector RA+ T cells ([TEMRA)] at M2 (Figure 3A) compared to HIV-uninfected, that normalized several months after infection. These longitudinal changes observed in the CD4 T-cell compartment over the first year of infection, were not significant for any subset. When comparing CHI-naïve and CHI-ART groups, CHI-ART group showed a non-significant tendency toward higher levels of TEM and lower significant levels of TN (P = 0.0413).

Conversely, we observed a marked loss of the TN and TCM CD8 T-cell pool after HIV infection concomitant to an increase in the frequencies of the TEM subset (Figure 3B). Comparing to HIV-uninfected individuals, CD8 TN significantly decreased in PHI at M2 and in CHI-naïve groups (P < 0.0001) and CD8 TEM significantly increased (P < 0.0001, P = 0.0002, respectively). No significant changes were observed in TCM; however, TEMRA was significantly increased in CHI-ART compared to PHI at M2 (P = 0.0027). Longitudinal analysis confirmed these changes (data not shown).

The dynamics of activation (CD38+ HLA-DR+ cells) exhaustion (CD279+ cells) and immunosenescence (CD57+ cells) were also analyzed in CD4 and CD8 T cells. As for maturation markers, most relevant changes were noticed in CD8 T cells. Although both activated and exhausted CD4 T cells were significantly increased at M2 when compared to HIV-uninfected individuals (P < 0.0001, P = 0.0056, respectively), and slowly but significantly decayed over the first year of infection (Figures S3A,B in Supplementary Material). Immunosenescent CD4 T cells (CD57+) showed no significant changes in the PHI group as compared to HIV-uninfected individuals but they were significantly increased in the CHI-ART group (P = 0.0001, Figure S3C in Supplementary Material).

The analysis of activation in CD8-T cells showed a significant increase at M2 as compared to HIV-uninfected that remained in CHI-naïve subjects (P < 0.0001) and normalized in CHI-ART (Figure 4A). The CD8 T-cell activation observed in the first months of infection was significantly reduced over time with a slow but significant decrease until months 9–11 (M9–11) (P < 0.0001, Figure 4B). Exhausted CD8 T cells also showed a transient but less marked significant increase as compared to HIV-uninfected individuals (P < 0.0001, Figure 4C), with a significant and slow decrease over time (P = 0.0056, Figure 4D). Along the course of HIV infection, the frequency of senescent CD8 T-cells was significantly higher in the CHI-ART group as compared to the HIV-uninfected group (P < 0.0001, Figure 4E). However, no significant changes over time were observed in the frequency of senescent CD8 T-cells by linear regression models (Figure 4F). There were no significant differences in percentages of activated, exhausted or senescent CD8 T-cells associated with the presence of co-infection in any of the study groups.

As described in Section “Materials and Methods,” the kinetics of 13 soluble biomarkers with expression levels most significantly different between febrile PHI and HIV-uninfected individuals (38) were characterized in detail during the first year after infection (Figures 5A,B; Figure S4 in Supplementary Material). Levels of IP-10, MCP-1, BAFF, soluble (s)CD14, tumor necrosis factor receptor-2 (TNFR2), and TRAIL were highly overexpressed at the first month of infection (M1) and underwent a prompt decrease in the subsequent months. This decrease was more gradual in the case of IP-10, which remained overexpressed even 5 months postinfection. On the contrary, MIG, sCD27, and sCD23 levels started to increase 1 month after infection and remained overexpressed for almost 1 year postinfection, while GSCF had a later upregulation at 7 months of infection.

Correlations between the plasma biomarker levels at M1 and activated or exhausted CD4 and CD8 T-cell phenotypes at 2 months postinfection were assessed as described in Section “Materials and Methods” (Figure 5C). We observed a significant positive correlation between M1 TNFR2 and sCD27 levels and the frequency of exhausted CD4 T cells and CD8 T cells at M2 (rho = 0.77, P < 0.0001; rho = 0.54, P = 0.0157, respectively). Similarly, we saw a significant association between M1 levels of BAFF (rho = 0.50, P = 0.0241), IL10 (rho = 0.44, P = 0.05), and sCD14 (rho = 0.44, P = 0.05) and the frequency of activated CD8 T-cells at M2. However, after adjustment by multiple testing, only the significance of TNFR2 with exhausted CD4 T-cells levels was maintained (P = 0.0135). We did not observe significant differences in the expression level of these 13 selected soluble biomarkers by the presence of any co-infection in the PHI or the CHI groups; however, BAFF, MCP-1, MIG, and TRAIL expression levels were significantly lower among the individuals included in the HIV-uninfected control group with a co-infection detected (P = 0.0254, P = 0.0418, P = 0.0001, P = 0.0032, respectively).