Targeted plasma proteomics reveals upregulation of distinct inflammatory pathways in people living with HIV

Summary Despite antiretroviral therapy (ART), people living with HIV (PLHIV) display persistent inflammation leading to non-AIDS-related co-morbidities. To better understand underlying mechanisms, we compared targeted plasma inflammatory protein concentration (n = 92) between a cohort of 192 virally suppressed PLHIV, who were followed-up for five years, and 416 healthy controls (HC). Findings were validated in an independent cohort of 649 virally suppressed PLHIV and 98 HC. Compared to HC, PLHIV exhibited distinctively upregulated inflammatory proteins, including mucosal defense chemokines, CCR5 and CXCR3 ligands, and growth factors. Unsupervised clustering of inflammatory proteins clearly differentiated PLHIV with low (n = 123) and high inflammation (n = 65), the latter having a 3.4 relative risk (95% confidence interval 1.2–9.8) to develop malignancies and trend for cardiovascular events during a 5-year follow-up. The best protein predictors discriminating the two inflammatory endotypes were PD-L1, VEGFA, LAP TGF β-1, and TNFRSF9. Our data provide insights into co-morbidities associated inflammatory changes in PLHIV on long-term ART.

Assessment of the profile of plasma inflammatory proteins allows the study of complex biological pathways that may lead to the identification of novel therapeutic targets and disease biomarkers. Previous studies that assessed plasma protein profiles were limited by small sample size and/or lack of a proper validation cohort (Babu et al., 2019a(Babu et al., , 2019bdeFilippi et al., 2020;Lemma et al., 2020;Sperk et al., 2018;Vos et al., 2021). In addition, other studies included only patients with central obesity and insulin resistance using statins (deFilippi et al., 2020) or children with HIV infection (Lemma et al., 2020).
In the present study, we identified signatures of 92 inflammation-related plasma proteins in virally suppressed PLHIV (n = 841) and compared them to healthy controls (HC) (n = 514). Using a discovery cohort and an independent validation cohort, we found mostly upregulation of plasma inflammatory protein concentrations in PLHIV compared to HC. Furthermore, stratification of PLHIV based on the inflammatory proteome revealed two distinct clusters, one with a high-and the other with a low-inflammation profile.

Age, sex, and BMI influence plasma inflammatory protein concentrations in PLHIV and HC
We first explore the relationship among plasma inflammatory proteins in PLHIV and HC from the discovery cohort and found general strong positive correlations among plasma inflammatory proteins (Figure S3). The strongest associations were found among intracellular proteins (4E-BP1, STAMBP, AXIN1, ST1A1, SIRT2, and CASP8), chemokines associated with neutrophil (CXCL1, CXCL5, and CXCL6) and monocytes chemotaxis (MCP-2 and MCP4), natural killer cell surface receptor (CD244), and immune mediators related with T and B cells development and activation (IL7, CXCL11, CD40/TNRSF5 and TNFSF14).
Next, we examined the influence of host factors on plasma inflammatory protein concentrations in PLHIV and HC from the discovery cohort. Advancing age was associated with an overall higher concentration of inflammatory proteins in both groups ( Figure S4). Female sex was significantly associated with reduced inflammatory proteins in HC, but not in PLHIV. The latter may have resulted from insufficient statistical power because of the small number of females in PLHIV (9%). BMI was positively associated with increased plasma inflammatory proteins concentrations, but to a lesser extent than age ( Figure S4).

Increased plasma inflammatory protein concentrations in PLHIV
To compare the inflammatory profile between virally suppressed PLHIV and HC, we first performed differential expression (DE) analysis using 78 circulating inflammatory proteins measurements from the discovery cohort of PLHIV (n = 192) and HC (n = 416). Subsequently, we validated the significant results from the discovery cohort using a second independent cohort of PLHIV (n = 649) and HC (n = 98). The analytical process of DE plasma inflammatory protein analysis is depicted in Figure 1A.
First, we performed an unsupervised hierarchical clustering analysis using 78 circulating inflammatory proteins measurements from the discovery cohort ( Figure 1B). We observed a distinct separation between the majority of PLHIV and HC individuals, suggesting an overall difference in their inflammatory profiles. This observation was confirmed through PCA showing separate clusters between PLHIV and HC ( Figure 1C). We next performed DE analysis to assess differences in individual plasma inflammatory protein concentrations between PLHIV and HC from the discovery cohort. Given the effect of age and sex on the inflammatory protein concentrations ( Figure S4), DE analysis was performed using a linear model with age and sex as covariates. The results of the DE analysis are presented in a volcano plot (Figure 1D). In total, 64 out of 78 proteins concentrations (82%) were differentially expressed (FDR<0.05) between PLHIV and HC, and most of the statistically significant proteins were upregulated in PLHIV. We found similar results when DE analysis between PLHIV and HC was performed with age, sex, BMI, and smoking status as covariates ( Figure S5).
We confirmed our findings in the independent validation cohort of 649 PLHIV and 98 HC ( Figure S1). In the validation cohort, PCA using relative concentrations of 62 proteins showed a different inflammatory profile between PLHIV and HC ( Figure 2A). DE analysis in the validation cohort identified 40/62 (64.5%) DEP between PLHIV and HC with pvalue<0.05 ( Figure S6), of which 29/40 (72.5%) proteins were upregulated in both the discovery and validation cohort ( Figure 2B and Table S1). iScience Article Finally, we analyzed whether the relative concentration of the validated differentially abundant plasma inflammatory proteins (n = 29) measured by Olink was associated with the absolute concentration of plasma markers measured by ELISA. In general, DEP were positively associated with acute-phase proteins (TNF-a and hsCRP), adipokines (resistin), cytokines (IL-6, IL-1Ra, IL-18BP, IL-18, and IL-10), and monocyte activation markers (sCD14 and sCD163) (FDR<0.05) ( Figure 2C). As expected, relative concentrations of IL-18 and TNF-a from the DE analysis significantly correlated with the absolute concentrations of IL-18 and TNF-a (FDR<0.005). In addition, hepatocyte growth factor (HGF) showed a significant positive correlation with other adipokines, such as resistin and leptin, and a negative correlation with adiponectin ( Figure 2C).

Network analysis reveals upregulation of specific inflammatory pathways in PLHIV
We further investigated the inter-relationship among the 29 significantly upregulated proteins in PLHIV compared to HC identified in the discovery and validation cohort. For this, we performed network analysis using relative concentrations of the 29 proteins from the PLHIV of the discovery cohort. Moderate to strong correlations (Spearman's Rho>0.3) are displayed in Figure 3A. Overall, significant positive correlations were observed among DEP (FDR<0.05). Protein-protein interactions among DEP were further visualized by dendrogram based on hierarchical clustering analysis ( Figure 3B). Four different clusters of proteins that shared similar functions were identified through the network and hierarchical clustering analysis ( Figures 3A and 3B).
The first cluster consisted of the growth regulator oncostatin M (OSM) and several growth factors, including HGF, vascular endothelial growth factor A (VEGFA), transforming growth factor a (TGF-a). VEGFA and TGF-a were the most central proteins in the network analysis, indicating that these proteins showed the most pairwise correlations with other proteins ( Figure 3A). OSM has been shown to stimulate the accumulation of immature and mature T-cells in lymph nodes, restoring immune responsiveness in immune-deficient mice (Clegg et al., 1996). Furthermore, OSM is known to play a role in the initiation and progression of Kaposi sarcoma (Miles et al., 1992;Nair et al., 1992), a common Herpes virus 8 related opportunistic cancer in PLHIV.
To identify the cellular origin of our differentially expressed proteins (n = 29), we used single-cell transcriptomic publicly available data from the Human Proteomic Atlas (HPA) project (Karlsson et al., 2021). HPA used consensus transcriptomics data in 76 single cell types to classify genes according to their singlecell type-specific category. We found innate (dendritic cells, macrophages, langerhans cells, natural killer (NK) cells) and adaptive immune cells (T cells) among cells that produce most of our differentially expressed proteins ( Figure S7).
Lastly, given that many of the identified proteins are primarily released by immune cells modulating subsequently their proliferation and migration, we investigated whether the DEP correlated with the proportion of circulating immune cells. The strongest association was found for OSM, which was positively and negatively associated with neutrophils and lymphocytes percentages respectively ( Figure 3D). These findings are consistent with a previous study showing that OSM is primarily expressed in neutrophils and stored in neutrophils granules in the circulation (Uriarte et al., 2008). In addition, increased proportions of Th2, and Th17 cells were associated with higher HGF, MCP-2, MCP-4, OSM, TNFSF14, and VEGFA. A significant and positive correlation was also found between NK bright cells proportion with CXCL11, which is a chemotactic factor for activated T-cells.
Unsupervised clustering of plasma inflammatory protein concentrations revealed two distinct clusters of PLHIV with high and low inflammation profile.
To assess the heterogeneity of plasma inflammatory proteins concentrations (n = 74) among PLHIV, we performed unsupervised hierarchical clustering analysis using the protein measurements (n = 74) of PLHIV from the discovery cohort. We identified two distinct clusters of PLHIV, one with low and one with a high inflammation profile ( Figure 4A). Seventy-one out of 74 plasma inflammatory proteins were significantly upregulated in the high inflammation group (FDR<0.05) compared to the low inflammation group. In addition, PCA using the first two principal components showed a limited overlap between the two clusters of PLHIV and HC, with individuals from the low inflammation group showing an inflammatory profile between HC and the high inflammation group ( Figure 4B). Furthermore, the absolute concentrations of plasma inflammatory markers were significantly higher in the high inflammation group compared to the low inflammation group, including acute phase proteins (TNF-a, hsCRP and AAT), adipokines (leptin and resistin), cytokines (IL-6, IL-1Ra, IL-18BP, IL-18, and IL-10), and monocyte activation markers (sCD14 and sCD163) ( Figure 4C).
PLHIV of the high inflammation group were significantly older (median 55.4 years) compared to the low inflammation group (median 50.6 years) (pvalue<0.05). Of note, the distinct inflammatory clusters found within PLHIV was not explained by sex and BMI ( Figure 4A), nor by HIV-related parameters, such as CD4 nadir and latest, CD4/CD8 ratio, HIV RNA zenith and latest value, viral blips, HIV and ART duration (Figure S8). Of importance, participants in the high inflammation group had more hypertension (pvalue<0.05) and a trend for previously diagnosed malignancies than those in the low inflammation group ( Figure 4D). Compared to the low inflammation group, PLHIV in the high inflammation group used more anti-hypertension drugs (pvalue = 0.06) and less vitamin D at inclusion ( Figure 4D). After inclusion, PLHIV from the discovery cohort were followed-up and relevant medical events were noted. After a 5-year follow-up period, we observed that PLHIV in high inflammation group developed more malignancies (13.8 versus 4%; relative iScience Article risk (RR) 3.4; 95% confidence interval (CI) 1.2 to 9.8) and had a trend for more CVD events (12.3 versus 7.3%; RR 1.7; 95% CI 0.6 to 4.2) compared to PLHIV in low inflammation group ( Figure 4D). Being in the high inflammation group was associated with an increased risk of malignancies even after controlling for age and sex (pvalue<0.05; binomial logistic regression model using age and sex as covariates) ( Figure 4D). Details for the type of malignancy and CVD events seen during 5-year follow-up in high and low inflammation group of PLHIV were described in Table S2.
Clear separation of the PLHIV into a low and high inflammation group was also observed in the validation cohort, mirroring the heterogeneity of inflammatory protein concentrations in well-treated PLHIV iScience Article ( Figure S9). Sixty-nine proteins were upregulated in the high inflammation group compared to the low inflammation group from the validation cohort, of which 67 proteins overlapped with those found in the discovery cohort (Table S3 and Figure 4E). The range of fold change of these 67 DEP identified in the high versus low inflammation group of PLHIV was comparable to the range of fold change of 29 DEP identified in PLHIV versus HC ( Figure 4E). Next, to identify the best protein predictors discriminating the high and low inflammation groups, we applied a random forest classification model in PLHIV from the discovery and validation cohort. The model was built using DEP and host demographic factors (age, sex, and BMI) as the input predictors. The predictors were ordered according to the mean decrease accuracy values, representing the importance of variables on distinguishing PLHIV of the high and low inflammation clusters. Out of the ten top-ranked predictors, three proteins were overlapped between the discovery and validation cohort ( Figure 4F). These proteins include PD-L1, VEGFA, and latency-associated peptide transforming growth factor beta-1 (Lap TGF b-1). The classification model performance was tested by calculating the area under the curve (AUC) of the receiver operating characteristic (ROC) curve, yielding an AUC value of about 99% in train and test sets of the discovery and validation cohorts, respectively ( Figure 4G).

Association of plasma inflammatory proteins with HIV-related parameters and smoking
Many factors may influence inflammation, such as HIV-related clinical parameters, comorbidities, co-medication, and smoking. We, therefore, correlated the plasma inflammatory proteins (n = 74) from PLHIV of the discovery cohort with HIV-related parameters (CD4 nadir and latest, CD4/CD8 ratio, HIV RNA zenith and latest, viral blips, HIV and ART duration, and HIV medication), HIV reservoirs, comorbidities, co-medication, and history of smoking in PLHIV from the discovery cohort.
First of all, we found strong positive associations between fibroblast growth factor 23 (FGF-23) and IL-6 concentrations with HIV duration (pvalue<0.005) ( Figure S10A). Moreover, apart from higher FGF-23, CCL19, and OSM (FDR<0.05), PLHIV with a history of CVD also had higher signaling lymphocytic activation molecule family member 1 (SLAMF1) and HGF (pvalue<0.05) compared to those that were not having a history of CVD. Circulating concentrations of CST5, an early biomarker for traumatic brain injury (Hill et al., 2017), were significantly increased in PLHIV with psychiatric conditions (pvalue<0.005) ( Figure S10B). Regarding co-medication, participants using antihypertensive and anticoagulant drugs had increased concentrations of FGF-23, CCL19, OSM, and HGF (pvalue<0.05) ( Figure S10C). Of importance, increased concentrations of several inflammatory proteins, including OSM and SLAMF1 (pvalue<0.005), as well as FGF-23, HGF, and CCL19 (pvalue<0.05), were significantly associated with a higher incidence of CVD during a 5-year follow-up period. In addition, tumor necrosis factor receptor superfamily member 9 (TNFRSF9) was significantly associated with the development of malignancies during a 5-year follow-up period (pvalue<0.005) ( Figure S10D). Furthermore, we observed comparable plasma inflammatory profiles between different ART groups (INSTI, NNRTI, and PI), as shown by PCA using the protein measurements of 74 inflammatory proteins ( Figure S10E). This observation is in agreement with a previous study in well-controlled PLHIV with dyslipidemia (deFilippi et al., 2020).
Finally, we investigated the influence of smoking on plasma inflammatory proteins in PLHIV using our discovery cohort ( Figures 5A and 5B). PLHIV were grouped based on smoking history (non-smokers, currently active smokers, passive smokers, and smoked in the past) and duration of smoking (those that do not smoke/with <1, 11-20, 6-10, and >20 years of active smoking). Non-smokers had a significantly higher stem cell factor (SCF) (FDR<0.005) and IL-12B (FDR<0.05) compared to a currently active smoker. CXCL6 and IL-12B concentrations were significantly higher in those who smoked in the past compared to the currently active smokers (FDR<0.05). In addition, matrix metallopeptidase 1 (MMP-1) and OSM concentrations were significantly increased in participants with 11-20 years of smoking or those who smoked for more than 20 years compared to those who never smoked or smoked for less than one year, respectively (FDR<0.05) ( Figure 5A). Consistent with previous results, SCF concentrations were lower in participants with 6-10 years and more than 20 years history of smoking compared to those who never smoked or smoked for less than one year (FDR<0.05) ( Figure 5A).  Figure S9.
(B) PCA of plasma inflammatory proteins (n = 74) measured in the discovery cohort using the first two principal components. Each dot represent participant from the HC (n = 415), and two clusters of PLHIV (low (n = 123) and high inflammation group (n = 65)). The ellipses were centered based on the median of the PC1 and PC2 for each group (HC, low, and high inflammation group). Protein distributions across PC1 and PC2 for each group are presented in marginal histogram plots. The median differences of the protein distribution across PC1 or PC2 between PLHIV and HC were calculated by Mann-Whitney-U test. ***p-value<0.0001. (C) Heatmap presenting FDR corrected pvalues of the comparison between the absolute concentration of plasma inflammatory markers in high inflammation group versus low inflammation group. The analysis was performed by linear regression model using age and sex as covariates.
(D) Barplot showing the frequency (yaxis) of comorbidities and co-medication at the baseline, and a 5-year follow-up clinical events in two clusters of PLHIV of the discovery cohort. The analysis was performed using the binomial logistic regression model using age and sex as covariates. *p-value<0.05. See also Figure S8 and iScience Article PD-L1 was positively associated with the number of cigarettes smoked per day (FDR<0.05) ( Figure 5B). Altogether, these findings corroborate the pro-inflammatory effect of cigarette smoking on the concentrations of a limited number of proteins (n = 6) in PLHIV. Given the limited influence of smoking on plasma inflammatory proteins, the DE analysis between PLHIV and HC showed comparable results before and after adjustment for smoking status in the discovery cohort ( Figure S5).

DISCUSSION
We profiled 92 circulating inflammatory proteins in the plasma of 192 virally suppressed PLHIV and 416 HC of European ancestry and further validated our findings in a second independent cohort. This is the first study that comprehensively assessed the inflammatory profile in PLHIV using a broad-scale proteomic approach. Previous studies identified only a limited number of inflammatory biomarkers in PLHIV. For example, nine inflammation markers were found to be differentially expressed in a study of 185 well-treated PLHIV and 104 HC of South African ancestry, with only three proteins (CCL25, PD-L1, and CXCL10) being in agreement with our findings (FDR<0.05, same direction) (Vos et al., 2021). Furthermore, two studies by Babu et al. in 53 PLHIV and 41 HC and 22 PLHIV and matched HC reported changes in 11 (pvalue<0.01) and 3 (FDR <0.1) proteins in well-treated PLHIV compared to HC respectively. Only five (ADA, CD8A, 4E-BP1, CCL23, and CST5) and one (CD8A) protein were similar to our findings respectively (Babu et al., 2019a(Babu et al., , 2019b. Another study using a cohort of PLHIV with dyslipidemia (n = 89) and HC (n = 46) also showed little overlap with our results, with only four inflammatory proteins (4E-BP1, ADA, TNFSF14, and CD40) overlapping with our results (FDR<0.05, same direction) (deFilippi et al., 2020). Of note, a study using 43 children with HIV infection and matched controls showed 15 proteins being downregulated in children with HIV infection compared to HC (FDR<0.01), with none of the proteins being identified in a similar direction compared to our study (Lemma et al., 2020). These differences between studies can be attributed to different cohort characteristics, especially the ethnicities, number of participants, proteins tested, and analysis strategies. Notably, one major disadvantage of the earlier studies is the small number of PLHIV and healthy controls studied, leading to a limited statistical power to identify true changes. In our study, we included a large number of subjects, confirmed the pro-inflammatory status of the PLHIV in the discovery cohort (van der Heijden et al., 2021), and found that the upregulated DEP in PLHIV strongly associated with the absolute concentration of plasma inflammatory markers, such as TNF-a, IL-6, and monocyte activation markers (sCD14 and sCD163), which have been previously described as predictors of CVD and mortality in PLHIV (McKibben et al., 2015;Sandler et al., 2011).
Most of our DEP were derived from innate immune cells ( Figure S7), corroborating the functional role of circulating innate immune cells on the persistent inflammation in PLHIV on ART (Altfeld et al., 2011;Altfeld and Gale, 2015;Rustagi and Gale, 2014). We previously reported increased monocyte-derived cytokine responsiveness in the same cohort of PLHIV (van der Heijden et al., 2021). Among DEP identified in this study are mucosal defense chemokines (CCL11, MCP-4, CCL20, CCL25, and CCL28) which are upregulated in PLHIV. Changes in mucosal defense chemokines, which are pivotal to maintain intestinal barrier function (Kulkarni et al., 2017;Sokol and Luster, 2015) may be a consequence or a result of the disruption of the intestinal immune system. The association of the intestinal immune system and HIV infection has great interest, and the assessment of the intestinal mucosal immune system has provided novel directions for therapeutic interventions that modify the consequences of acute HIV infection (Brenchley and Douek, 2008). Also, chronic microbial translocation in well-treated PLHIV because of intestinal damage has been linked with persistent immune activation (Brenchley et al., 2006;Dinh et al., 2015;van der Heijden et al., 2021). Given the importance of the intestinal mucosal immune system in HIV infection, we found a strong association between an inflammatory marker for the intestinal lymphocytes recruitment, CCL25, and a marker for intestinal barrier function, that is, IFABP, indicating a role of the intestinal barrier in the inflammatory status of PLHIV. iScience Article Furthermore, CCL3 and CCL4 chemokines that were upregulated in PLHIV are known to be ligands for CCR5, the main HIV co-receptor. We observed that CCL4 was negatively correlated with CCR5 expression in several subsets of CD4 + and CD8 + cells, suggesting protective effects against cell-to-cell HIV transmission ( Figure 3C). It has been shown that the dose-dependent administration of the recombinant CCL3 and CCL4 inhibit different strains of HIV-1, and HIV-2 as well as simian immunodeficiency virus (SIV) in vitro (Cocchi et al., 1995). Given that CCR5 is predominantly found in CXCR3 expressing cells, we assumed that the upregulation of CXCR3 ligands in PLHIV (CXCL9, CXCL10, and CXCL11) might facilitate the recruitment of T cells and enhance virus propagation or development on CVDs (Li and Ley, 2015). Of interest, CXCL9, CXCL10, and CXCL11 have been reported to predict the progression of HIV disease in the primary HIV infection (Yin et al., 2019). Moreover, we identified growth factors and regulator proteins (HGF, VEGFA, TGF-a, and OSM) to be upregulated in PLHIV compared to HC. Higher concentrations of these proteins seem to reflect alteration in the regulation of many important cellular processes in PLHIV.
To identify whether there are subgroups within the PLHIV based on their plasma inflammatory profile, we performed unsupervised clustering using the discovery (n PLHIV = 188) and validation cohort (n PLHIV = 649). PLHIV were clustered into two groups: those with a low and high inflammatory profile ( Figure 4A). The high inflammation group showed upregulation of 67 out of 74 proteins and increased absolute concentrations of circulating inflammatory markers compared to the low inflammation group. The upregulation of almost all inflammatory proteins may reflect a broad dysregulation of immune responses in the high inflammation group as these proteins are involved in various inflammatory pathways. Upregulation of intracellular proteins (4E-BP1, STAMBP, AXIN1, ST1A1, SIRT2, and CASP8) in the high inflammation group may suggest a higher intracellular protein leakage and/or cell death in this group, given strong interrelations between intracellular proteins with proteins involved in innate and adaptive immune response ( Figure S3). Furthermore, the range of fold change of DEP in the high versus low inflammation group of PLHIV ( Figure 4E) was similar to the fold change in PLHIV versus HC ( Figure 2B), suggesting a comparable level of dysregulation of inflammation between these groups.
Compared to the low inflammation group, PLHIV in the high inflammation group had more hypertension at baseline and a higher number of people developing various types of malignancies (Table S2) (RR 3.4; 95% CI 1.2 to 9.8) and a trend of CVD (RR 1.7; 95% CI 0.6 to 4.2) during a 5-year follow-up period ( Figure 4D). This association appears to be independent of age, HIV-clinical parameters (CD4 nadir and latest, CD4/CD8 ratio, HIV RNA zenith and latest, viral blips, HIV and ART duration, and HIV medication), and other iScience Article recognized pro-inflammatory risk factors (obesity and smoking status). Of note, although the high inflammation group were older than those in the low inflammation group, age has not been observed as one of the top ten predictors that separated the two inflammatory endotypes in both cohorts (discovery and validation), indicating that the differences in the inflammation levels may not be primarily attributed to the age differences. Three proteins (PD-L1, VEGFA, and Lap TGFb-1) appeared to be the best predictors discriminating the two inflammatory endotypes in the discovery and validation cohort ( Figure 4F). PD-1 and PD-L1 inhibitors are well-known checkpoint anti-cancer drugs (Chen et al., 2016). TGF-b and VEGFA can induce the expression of PD-L1 and its receptors (PD-1), respectively (Du et al., 2021;Song et al., 2014;Voron et al., 2015), promoting immunosuppression during malignant transformation. Anti-cancer immunotherapy targeting either one or a combination of PD-L1, TGFb or VEGF has demonstrated a synergistic anti-tumor effect and is subjected to ongoing research (Ciardiello et al., 2020;Courau et al., 2016;Gulley et al., 2021;Hack et al., 2020). The safety and efficacy of anti-PD1 or VEGF inhibitor therapy have been reported in studies of HIV patients with different types of malignancies (Bari et al., 2019;Bender Ignacio et al., 2016;Lavolé et al., 2018;Ostios-Garcia et al., 2018). In addition, TNFRSF9 showed the strongest association with the development of malignancy during a 5-year follow-up period ( Figure S10D) and was among the top predictors discriminating the two inflammatory endotypes in PLHIV ( Figure 4F). TNFRSF9 is an immune costimulatory receptor expressed on activated T-and natural killer (NK) cells and has been proposed as a new target for cancer immunotherapy. Prior studies have shown that utomilumab and urelumab, which are TNFRSF9 agonistic antibodies, can deliver costimulatory signals, enhancing T-cell-mediated anti-tumor activity in vitro and in vivo (reviewed in (Chester et al., 2018)).
In conclusion, our study underscores the importance of targeting specific inflammatory pathways that are upregulated in virally suppressed PLHIV compared to HC. Although upregulation of mucosal defense chemokines that represent disruption of intestinal immunity is well known in HIV infection, other inflammatory pathways, such as CCR5 ligands, CXCR3 ligands, and growth factors proteins, were less known and can pave the way toward new therapeutic options. Last but not least, clinicians should be aware that PLHIV with a high inflammation endotype are at increased risk to develop malignancies and CVDs. Therefore, patient stratification based on inflammatory profile warrants further research in order to develop better therapeutic and preventive strategies against malignancies and CVDs in virally suppressed PLHIV.

Limitations of the study
There were several limitations to our study. First, the findings are mainly correlative that hinders causation inference as correlation does not imply causation. Second, demographic differences between groups across cohorts may introduce bias in the results of our study. However, all analyses were performed with adjustment for age and sex to account for the demographic differences. Next, a low percentage of participants with comorbidities may limit the power to find statistically significant associations with plasma inflammatory markers in PLHIV after accounting for multiple testing burden. However, we identified significant associations between DEP and different comorbidities before multiple testing correction (pvalue< 0.05), which is in agreement with previous studies.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

DECLARATION OF INTERESTS
The authors have declared that no conflict of interests exists.

Lead contact
Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Nadira Vadaq (N.Nadira@radboudumc.nl).

Materials availability
This study did not generate new unique reagents.
Data and code availability d All data reported in this article will be shared by the lead contact on request.
d This study does not report original code.
d Any additional information required to reanalyze the data reported in this article is available from the lead contact on request.

Cohort of study participants
This study included discovery and an independent validation cohort of virally suppressed PLHIV and HC and is part of the Human Functional Genomics Project (HFGP) (www.humanfunctionalgenomics.org) (Netea et al., 2016). A schematic representation of the cohorts is shown in Figure S1.

Discovery cohort
The discovery cohort consisted of PLHIV recruited between December 2015 and February 2017 at Radboudumc, Nijmegen, the Netherlands. Participants were 18 years and older, received ART for more than six months, and had HIV-RNA levels <200 copies/ml. Detailed patient characteristics have been reported elsewhere (Van deWijer et al., 2021;van der Heijden et al., 2021). Relevant comorbidities and co-medication data with a prevalence of about 10% of total study participants were presented. PLHIV were followed for the development of clinical events during a period of five years (2016)(2017)(2018)(2019)(2020)(2021). The control group consisted of a historical cohort of healthy Dutch individuals (500 Functional Genomic [500FG] cohort) recruited between August 2013 and December 2014 at Radboudumc, the Netherlands. Details about the 500FG cohort have been previously described ter Horst et al., 2016). In total, protein measurements were available for 198/211 of the PLHIV and 423/534 of the healthy controls.

Validation cohort
The iScience Article criteria were applied for the discovery cohort. Healthy control samples were obtained from the 200 Functional Genomic (200FG) cohort, as described elsewhere . Samples were collected between 2019 and 2020. In total, protein measurements were available for 661 of the PLHIV and 100 sex-and agematched healthy controls.

Study approval
The studies involving human participants were reviewed and approved by the

Sample processing
Sample collection and processing in the cohorts of the HFGP are performed using similar study procedures (Netea et al., 2016). Blood samples of the discovery cohort were collected at Radboudumc and were processed immediately. For the validation cohort, blood samples from different centers were sent overnight to Radboudumc or left on the bench at room temperature to be processed the next day. The second and first freeze-thaw cycles of plasma fractions were used for proteomic analysis in the discovery and validation cohort, respectively. Samples were centrifuged using similar settings and stored at À80 C.

Proteomic profiling of circulating inflammatory proteins
For the discovery cohort, 92 unique protein biomarkers were measured using the Olink inflammation panel (Olink Proteomics, Uppsala, Sweden) (Assarsson et al., 2014). For the validation cohort, the Olink Explore panel (Filbin et al., 2021) (n = 92/1472 proteins) was used. PLHIV and healthy controls were measured simultaneously. Proximity extension assay (PEA) technology was applied to measure relative concentration of the proteins, which were presented as log 2 normalized protein expression level (NPX) values.

Immunophenotyping and gating strategies
Immunophenotyping data were available for PLHIV from the discovery cohort. The complete protocol of immunophenotyping and gating strategies has been described before (Van deWijer et al., 2021). Samples were measured on a 10-color Navios flow cytometer (Beckman Coulter, Fullerton, CA, USA) equipped with 488, 638, and 405 nm solid-state lasers. Stained blood samples were analyzed using five supplemental 10-color antibody panels. Flow cytometry data were analyzed using Kaluza software version 1.3 and version 2.1. Gating was conducted and verified by two independent specialists. The absolute number of white blood cells (WBC) per mL of blood (by Beckman Coulter AcT Diff Hematology Analyzer) were used to calculate absolute numbers of leukocyte (CD45 + ) cell subsets as measured by flow cytometry. A subset (n = 33) of the most relevant WBC population representing innate and adaptive cell compartments was selected. Data were presented as WBC percentages based on the cell count of each subpopulation by its respective population (one level up).

QUANTIFICATION AND STATISTICAL ANALYSIS
Quality control of proteomic data The proteomic data from the discovery cohort were normalized using inter-plate controls for batch variation correction and were reported in the log 2 scale. Data values below the limit of detection (<LOD) were handled using the actual measured values to increase the statistical power and give a complete data distribution. Outlier detection was done using principal component analysis (PCA), in which data points falling R3 standard deviations (SD) from the mean of principal component one (PC1) and two (PC2) were excluded. Proteins with (1) <LOD values >25% of the samples in both groups (PLHIV and HC) and (2) a difference in the <LOD values between PLHIV and HC <20% were excluded. For analysis within the PLHIV, proteins with <LOD values >25% of the samples were excluded for the follow-up data analysis. Details of pre-analytical steps for both discovery and validation cohort were described in Figure S1. Percentages of <LOD values per protein were visualized in Figure S2.

Differential expression protein analysis
For protein analyses in the discovery cohort, we compared the relative concentration of plasma inflammatory proteins between PLHIV and HC. During QC per sample, 12 outliers and one sample with no measurements were excluded from the analysis. In total, 78 proteins in 192 PLHIV and 416 HC participants from the discovery cohort were available for differential expression (DE) analysis ( Figure S1).
In the validation cohort, we selected the significantly differentially expressed proteins (DEP) (n = 64) identified in the discovery cohort. Similar pre-analytical approaches were implemented in the data analysis as in the discovery stage. Sixty-two proteins were used for DE analysis after quality control per protein, as described above. After excluding participants that were not using ART (elite controllers; n = 4) and the outliers (n = 10) as described above, we used 649 PLHIV and 98 HC for DE analysis ( Figure S1).
We used a linear regression model with age and sex as covariates in the discovery and validation cohort. Adjustment for multiple testing comparisons was done using false discovery rate (FDR) method. Proteins with FDR <0.05 were considered statistically significant in the discovery cohort, and pvalue<0.05 were considered statistically significant in the validation cohort.

Clustering analysis
To assess any similarities and dissimilarities within groups, we performed unsupervised hierarchical clustering using k-nearest neighbors with 100 repetitions. Before analysis, NPX values for each protein were scaled to have a mean of 0 and a SD of 1. Results were visualized as a heatmap by calculating the matrix of Euclidean distances from the scaled NPX value. Scaled data exceeding color break value (À2 or 2) were assigned with corresponding maximum or minimum colors.

Network analysis of DEP
Network analysis was performed using Spearman's rank correlation, and association with moderate to high correlation coefficient (rho>3) was visualized. Each protein was visualized as a node and colored based on node's importance according to node's closeness. For the dendrogram of DEP between PLHIV and HC, the distances between proteins measured in PLHIV of the discovery cohort were calculated using Ward's agglomeration method, and hierarchical clustering of the proteins was calculated based on the pairwise distances of proteins using Spearman's rank correlation.

Random forest
Random forest classification model was performed to predict the high and low inflammation groups of PLHIV in the discovery and validation cohort. In addition to demographic factors (age, sex, and BMI), concentration of DEP between the high and low inflammation groups of the discovery (n = 71) and validation cohort (n = 69) were used as input. Before the random forest-based modeling, PLHIV in the discovery and validation cohorts were randomly split into a training (70%) and a validation set (30%). The model was built with 1000 trees and 4 (the discovery cohort) or 2 (the validation cohort) random variables that were considered at each tree. This model was chosen based on the best prediction accuracy on the training set. The iScience Article importance of each variable in the prediction model was expressed as a mean decrease in accuracy values. The performance of the prediction model of the discovery and validation cohort was visualized using the receiver operating characteristic curve (ROC).

Data analysis and visualization
All statistical analyses were performed using R version 4.0.2. The following R packages were used for results visualization: 'ggbiplot' for PCA plot, 'ComplexHeatmap' for unsupervised hierarchical clustering analysis, 'randomForest' for random forest analysis, 'tidygraph' and 'ggraph' for network analysis, and 'ggplot2' for the rest of plots. 'Limma' was used for DE analysis which uses an empirical Bayes method to moderate the standard errors of the estimated log-fold changes (Ritchie et al., 2015). 'randomForest' and 'ROCR' were used for random forest and ROC analyses, respectively. Complementary approach to identify the cellular origin of DEP was performed using single-cell transcriptomic publicly available data from the Human Proteomic Atlas (HPA) project (proteinatlas.org) (Karlsson et al., 2021