2.1 Study Design and Participants

The study investigated two independent cohorts of PLHIV, the 200HIV and 2000HIV cohorts, and one cohort of people without HIV the 200FG cohort, which are all part of the Human Functional Genomics Project (HFGP, http://www.humanfunctionalgenomics.org). The 200FG cohort comprises adults of Western European ancestry, older than 18 years old, enrolled in 2019. Table 1 provides the demographic characteristics of the volunteers from the three cohorts. For the proteomic profiling of the 2000HIV, we used measurements available in 641 participants out of 1895 virally suppressed PLHIV of the 2000HIV cohort, who had not been vaccinated against COVID-19. The 2000HIV cohort comprises PLHIV of multiple ethnicities (European, African, Asian, Hispanic, Native American, and mixed ancestry) who were recruited in four HIV treatment centers (Radboudumc Nijmegen, Erasmus MC Rotterdam, and OLVG Amsterdam and Elisabeth-TweeSteden Ziekenhuis Tilburg) between 2019 and 2022. Inclusion criteria were as follows: age > 18 years, HIV-seropositive, receiving ART ≥ 6 months, and latest HIV-RNA levels < 200 copies/mL. Exclusion criteria were absence of informed consent, detectable viral hepatitis B or C DNA by polymerase chain reaction (PCR), any current acute infection, and pregnancy. PLHIV from the 200HIV (n = 210) cohort were enrolled between December 2015 and February 2017 at Radboudumc and consisted of PLHIV > 18 years old who received ART for ≥ 6 months, had HIV-RNA levels < 200 copies/mL, and were of European ancestry. Demographic and clinical characteristics and sample handling procedures were previously described elsewhere for the 200FG (Jaeger et al. 2019), 200 HIV (van der Heijden et al. 2021), and 2000HIV studies (Vos et al. 2022). The research was conducted following the principles of the Declaration of Helsinki and study protocols were approved by the Medical Ethical Review Committee Oost Nederland, Nijmegen, the Netherlands under the registration NL68056.091.81 [2000HIV], NL42561.091.12 [200HIV] and 2011-399 [200FG]. All participants provided written informed consent. For the current study, we used samples collected from 100 volunteers of the 200FG cohort, 641 volunteers from the 2000HIV, and 210 volunteers from 200HIV cohort.

2.2 Proteomic Profiling of Circulating Plasma Proteins

Collection, processing, and storage of the cohorts' blood samples were performed according to study protocols from the HFGP (Netea et al. 2016). Venous whole-blood samples were collected using EDTA tubes and centrifuged into plasma before being stored at −80°C. For the 2000HIV study participants, blood was collected during the baseline study visit and shipped to Laboratory of Experimental Internal Medicine, Radboudumc, Nijmegen, overnight at room temperature.

Proteomic profiling was conducted in plasma samples using a proximity extension assay coupled with next-generation sequencing as a readout method by Olink Proteomics AB (Uppsala Sweden) (Assarsson et al. 2014). The relative abundance of plasma proteins was measured using the library Olink Explore 1536 platform consisting of 1472 proteins divided into four 384-well multiplex panels focused on inflammation, oncology, cardiometabolic, and neurology protein. Protein measurements are presented as Normalized Protein Expression (NPX) values, which is Olink's relative protein quantification unit on log2 scale. Standard quality control (QC) per protein and sample was performed prior to statistical data analysis. In each of the four panels from the Olink Explore 1536 platform, IL-6, TNF, and CXCL8 were measured as technical duplicates for quality control purposes. Strong correlations (Spearman rho correlation r > 0.9) were observed between the technical duplicates among panels, and therefore, we selected the measurements from the inflammatory panel. Next, we excluded proteins with LOD ≥ 25 of the samples, resulting in 1306 proteins in the 2000HIV and 200FG cohort, and 1293 in the 200HIV cohort for follow-up analysis. During QC per sample, we performed principal component analysis (PCA) using the NPX values. Samples were defined as outliers when falling above or below four standard deviations (SD) from the mean of principal component one (PC1) and/or two (PC2). After removing outliers, 634, 98, and 205 samples remain in the 2000HIV, 200FG, and 200HIV cohorts, respectively. Moreover, to avoid confounding effect of COVID-19 infection, 13 COVID-19 positive individuals were removed from the 2000HIV. The overview of QC process is depicted in Figure S1.

2.3 Cytokine Production Capacity Upon Ex Vivo Stimulation of Peripheral Blood Mononuclear Cells (PBMCs)

PBMCs were seeded in round-bottom 96-well plates at 0.5 × 10⁶ cells/well in 0.2 mL of RPMI Dutch modified (Life Technologies) supplemented with Gentamycin 5 mg/mL (Centrafarm), Pyruvate 1 mM, and GlutaMAX 2 mM (Life Technologies). Cells were stimulated for either 24 h or 7 days at 37°C and 5% CO₂ with different bacterial, fungal, and viral stimuli (Table S1). In 7 days culture, the medium was supplemented with 10% human pool serum. Supernatants were collected and stored at −80°C until used for ELISA measurements. Concentrations of IL-1β, IL-6, IL-8, MCP1, and MIP1-a were determined in the supernatants of the 24-h PBMCs and IFNγ, IL-5, IL-10, IL-17, and IL-22 in the supernatants of the 7-day PBMCs, using commercial ELISA kits (Duoset ELISA, R&D Systems, catalog numbers: Table S1). QC was performed as follows: for 24-h cytokines, samples from 1814 participants were measured, of which 42 samples were excluded for being contaminated due to cytokine production in the RPMI well (negative control). RPMI-contaminated well was defined as having concentrations of above 2× lower limit of detection (LLOD) in two out of the three cytokines: TNF, IL-1β, or IL-6. Moreover, based on PCA analysis, samples falling above or below four standard deviations (SD) from the mean of principal component one (PC1) and/or two (PC2) were removed as outliers. After QC, measurements from a total of 1760 participants remained with 35 cytokine-stimulus measurements. For 7-day cytokines, samples from 1816 participants were measured, and after exclusion of samples due to RPMI contamination and PCA outliers, 1754 remained with 30 cytokine-stimulus measurements. For this study, we used 551 samples (24 h) and 550 samples (7 days), for which both ex vivo cytokine production and plasma proteomic data were available.

2.4 Association of Protein Relative Abundance With Chronological Age

The association of each NPX protein value with chronological age was determined using a linear regression model adjusted by sex in the 200FG (n = 98 samples) and 200HIV (n = 205 samples) cohorts. In the 2000HIV cohort, the linear model was adjusted by sex, ethnicity, inclusion center, and the time elapsed between patient's phlebotomy and blood sample processing in the laboratory. The top five principal components (PCs) extracted from the genotype of the individuals for which we had both genotype and plasma proteomic data (n = 588 samples) were included in the model for correction of ethnicity. To account for multiple testing, correction for the false discovery rate (FDR) was applied using the Benjamini-Hochberg method and proteins with adjusted p-value (FDR) < 0.05 were considered significantly associated with chronological age. A meta-analysis using the summary statistics of the proteins associated with age in the three cohorts was performed using a random effect model to calculate the summary fold change over the three cohorts using a summary p value that represents the probability that the summary fold-change is not different than zero as implemented in the MetavolcanoR package (Prada 2020). Overall, 77 proteins consistently associated with age in all the three cohorts in the same direction were included in the plasma proteome-based prediction model (p_{meta-analysis} < 0.05).

2.5 Plasma Proteomic-Based Age Prediction Model

To construct a plasma proteomic-based age prediction model, we initially examined three penalized regression models, Ridge, Lasso, and Elastic Net, using the NPX values of the 77 proteins associated with age from the 200FG cohort. The 200FG data were randomly split into a training set (80% for building a predictive model) and a test set (20% for evaluating the model). Hyperparameter tuning was performed for each model and RMSE (root mean square error) was calculated using the train and test dataset. Each hyperparameter combination (alpha, lambda) was evaluated using a 10-fold cross-validation to calculate the average performance of each model. Lasso regression was selected as the best model as it minimized the median RMSE compared to ridge and elastic net in the 10-fold cross-validation (median RMSE = 5.99 in Lasso versus 7.65 and 6.11 in Ridge and Elastic Net, respectively). When using Lasso, the following hyperparameters were set up on the training data set: alpha = 1; “lamda.min” as the shrinkage variable estimated after 10-fold cross-validation = 0.6086. We computed variable importance using varImp() to understand which proteins are the best predictors in the model. The varImp function tracks the changes in model statistics for each predictor and accumulates the reduction in the statistic when each predictor's feature is added to the model. This total reduction is used as the variable importance measure. Relative variable importance values range from 0% to 100%. The most important variable always has a relative importance of 100%. Finally, a predicted proteomic age was calculated for 205 and 588 PLHIV from the 200HIV and 2000HIV cohort, respectively, as a linear combination of the regression coefficients of 77 proteins from the Lasso regression model. For computing penalized linear regression models, glmnet package was used in R. Finally, model training and evaluation were performed using caret package in R (4.1.2).

2.6 Statistical Analysis

Age advancement, as a proxy of accelerated aging, was calculated as the difference between proteomic-based predicted age and chronological age, as previously described (De Francesco et al. 2019). Age advancement in each cohort is reported as a median and evaluated for statistical significance using Wilcoxon signed-rank tests. We calculated age advancement as a proxy of accelerated aging, with values above zero indicating a higher proteomic age compared to chronological age.

2.7 Role of the Funding Source

The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.

3.1 Baseline Clinical Characteristics of the Cohorts

In this cross-sectional study, we assessed the plasma proteome of 588 participants from the 2000HIV cohort (median age: 53 years, [IQR 45–60]; 91.5% male [n = 538] 8.5% female [n = 50]), 205 participants of the 200HIV cohort (median age: 52.5 years, [IQR 46.2–59.4]; 92.2% male [n = 189] 7.8% female [n = 16]), and 98 participants without HIV from the 200FG cohort (median age: 49.5 years, [IQR 36.9–57.7]; 75.5% male [n = 74] 24.5% female [n = 24]). Participants from the 2000HIV cohort had a slightly higher BMI (median: 25, IQR [IQR 22.7–27.4]), HIV duration (median: 11.8 [IQR 7–19]) and ART duration (median: 9.64, [IQR 5.73–15.9]), CD4 nadir (median: 0.26, [IQR 0.14–0.38]), and CD4 latest (median: 0.70 [IQR 0.52–0.92]), compared to participants from the 200HIV cohort (p < 0.05). No significant differences were found for the presence of comorbidities between the two PLHIV cohorts. General participant characteristics are described in Table 1.

3.2 Identification of Plasma Proteins That Predicts Biological Age in People Without HIV and PLHIV

To investigate which plasma proteins were associated with chronological age and whether PLHIV exhibit a premature aging phenotype, we first assessed the association between the plasma protein concentrations with chronological age in the 200FG, 200HIV, and 2000HIV cohorts. Of the total proteins evaluated, chronological age was significantly associated with 101, 322, and 467 proteins (FDR < 0.05) in the 200FG, 200HIV, and 2000HIV cohorts, respectively (Figure 1A, Tables S3–S5). To identify proteins consistently linked to age, we next performed a meta-analysis of the proteins significantly associated with age (FDR < 0.05) across all three cohorts. There were 77 proteins that reached statistical significance (p < 0.05) in the meta-analysis (Figure 1B and Table S6) demonstrating their association with age. Out of the 77 proteins identified, nine proteins (TGFB1, SCARB2, NEFL, LTBP2, IGFBP4, GDF15, EFEMP1, DCN, and ColA1) were described as components of the Senescence Associated Secretory Phenotype (SASP) factors according to the database of the secretomes of senescence cells: SASP Atlas (Basisty et al. 2020).

To evaluate whether the plasma proteome could predict chronological age, a prediction model using lasso regression was developed using the common 77 protein-signature associated with age. The predicted and chronological age showed a strong positive correlation (r = 0.93, p < 2.2e−16) in the 200FG cohort, indicating that the identified plasma proteins are a strong predictor of age.  This positive correlation was validated independently in the other two cohorts, with correlation coefficients between predicted and chronological age of 0.81 and 0.80 in the 200HIV and 2000HIV cohort, respectively (p < 2.2e−16 Figure 1C). In addition, lasso regression allowed to select the best predictor proteins based on their importance score (Figure S7 and Table S7). The following proteins were the top ten predictors with an importance score ranging between 40% and 100%: PODXL2, TNXB, COL1A1, GFAP, WNT9A, RET, SMOC1, BC, CTSV, and WISP2 (see Table 2). Among our top ten predictors for aging, PODXL2 was the most important predictor identified by the model.

Next, we calculated the age advancement, and we found that PLHIV from the 2000HIV cohort showed a significant increase in age advancement (median = 1.37, IQR = 8.91) compared to 200FG people without HIV (median = 0.045, IQR = 5.95, p = 0.0041). While the difference between the 200HIV and 200FG cohorts did not reach statistical significance (Figure 1D), this may be explained by the smaller sample size of the 200HIV compared to the 2000HIV cohort. In addition, differences in age advancement were investigated across the three cohorts following chronological age strata: young (18–35 years), middle age (36–60 years), and older (60+ years) individuals. Young PLHIV showed increased age advancement compared to people without HIV (2000HIV: median = 8.85, IQR = 7.94, p = 0.0041; 200HIV: median: 6.68, IQR = 6.02, p = 0.033). In PLHIV from 2000HIV cohort, a similar behavior was observed in the middle age group (2000HIV: median = 1.36, IQR = 8.27, p = 0.0017; 200HIV: median = 0.19, IQR = 7.05, p = 0.13). However, no significant differences were observed within the older age group (2000HIV: median = −1.02, IQR = 11.09, p = 0.36; 200HIV: median = −0.71, IQR = 7.38, p = 0.99) (Figure 1E).

3.3 Association of Age Advancement With Demographics, HIV- and ART-Related Parameters and Comorbidities in PLHIV

Next, to evaluate whether the calculated inflammatory age advancement is reflecting biological aging in PLHIV, we examined its association with various demographic, healthy status, and HIV-specific factors, as well as age-related comorbidities in the 2000HIV cohort. Age advancement was not correlated with sex, current smoking, BMI, co-infection with syphilis, CD4 nadir, or last CD4 counts (Figure 2). Prior viral co-infections, such as hepatitis C (HepC) and CMV (IgG titers), were positively correlated with age advancement, as well as HIV duration, ART duration, and the latest measurement of CD8 T-cell count. Importantly, age advancement was positively correlated with cardiovascular and metabolic comorbidities including hypertension, myocardial infarction, presence of plaques, angina pectoris, hypertriglyceridemia, type 2 diabetes mellitus (T2DM), metabolic syndrome, and liver stiffness measurement (used as liver fibrosis proxy) (Figure 2). An accelerated age advancement was, on the other hand, negatively correlated with HDL cholesterol concentrations (FDR < 0.05) (Figure 2 and Table S8).

To test whether the associations between inflammatory age advancement and the above-mentioned factors were confounded by traditional cardiovascular risk factors, we used a linear regression model using chronological age, sex, T2DM, hypercholesterolemia, hypertriglyceridemia, and hypertension as confounders. Age advancement remained significantly correlated with prior viral co-infections (HepC and CMV), higher CD8-T cell count, and lower HDL levels (FDR < 0.05) (Figure S8 and Table S9).

Since myocardial infarction and T2DM were the comorbidities that exhibited the strongest association with age advancement, we compared the relative concentrations of GFAP, SMOC1, WNT9A, and WISP2 (identified between the top ten predictors as the ones upregulated with chronological age) within participants from the 2000HIV cohort stratified based on the presence or absence of these comorbidities. The concentration of WNT9A and SMOC1 was significantly higher only in PLHIV with faster age advancement having at least one of the evaluated comorbidities. However, no significant differences were observed for these two proteins among participants with lower age advancement. Notably, this pattern was not observed for GFAP and WISP2 proteins (Figure S9).

Besides comorbidities and co-infections, ART usage can contribute to accelerated aging in PLHIV and, therefore, we investigated the association of age advancement with cumulative exposure to antiretroviral drugs. We found that cumulative exposure to the protease inhibitor ritonavir (RTV_cum) was positively correlated with an increased age advancement (FDR < 0.05). (Figure 3 and Table S10). Although not statistically significant after multiple testing correction, we observed a significant negative association (p-value < 0.05) between age advancement and cumulative exposure to Dolutegravir and Rilpivirine. Additionally, a positive association was found with Zidovudine and total cumulative exposure to protease inhibitors.

3.4 Association of Age Advancement With Ex Vivo Cytokine Production

To investigate the influence of age advancement on the functionality of immune cells in PLHIV, we assessed the secreted SASP-cytokines in the supernatant of PBMC after exposure to various microbial and (non)-microbial stimuli. Overall, a positive correlation was found between age advancement and a higher production of IL-1β in response to HIV-envelope peptide pool (HIV-ENV), Imiquimod (IMQ), and Streptococcus pneumoniae, as well as IL-8 in response to IMQ, and MCP1 in response to S. pneumoniae. A negative correlation was found between age advancement and the following cytokines produced by T cells: IL-5 in response to Candida albicans conidia, IFNγ in response to E. coli, and IL-22 in response to Mycobacterium tuberculosis (Figure 4, Figure S10, Tables S11 and S12).