NK cell dysregulation may potentiate cardiovascular disease in adolescents with perinatally acquired HIV on antiretroviral therapy

Abstract

Perinatally acquired HIV (PHIV) and antiretroviral therapy (ART) can alter innate immune cells, (monocytes and natural killer [NK] cells) which are important in the pathogenesis of cardiovascular disease (CVD). We compare cardiovascular biomarkers and immune signatures between adolescents with PHIV (APHIV) on suppressive ART and HIV-unexposed, adolescents without HIV in Uganda. Carotid intima-media thickness (IMT) is increased in APHIV, suggesting a higher CVD risk. Flow cytometry analysis reveals greater activation, memory, and migratory capabilities of NK cells, and increased pro-inflammatory intermediate monocytes in APHIV, and these observations are supported by transcriptomics. Many of these innate immune cell subsets are associated with carotid IMT. Plasma oxidized-LDL (Ox-LDL) is significantly lower among APHIV, and negatively correlates with pro-inflammatory, memory-like NK subsets. We demonstrate increased uptake of Ox-LDL by macrophages in the presence of activated, memory-like NK cells in vitro, suggesting a possible mechanism for greater CVD risk in APHIV. Collectively, our data demonstrate associations between dysregulated NK cell signatures and increased CVD risk among APHIV.

Perinatally acquired HIV and long-term antiretroviral therapy can affect innate immune cells and may contribute to cardiovascular disease (CVD) pathogenesis. Here, the authors compare immune cell profiles in perinatally HIV infected adolescents who are virally suppressed on antiretroviral therapy and adolescents without HIV in Uganda and show that the adolescents with HIV demonstrate early features of CVD risk, which may be linked to an altered NK cell phenotype.

Introduction

HIV remains a significant public health challenge, particularly in pediatric and adolescent populations. According to the Joint United Nations Program on HIV/AIDS (UNAIDS), there were an estimated 1.5 million children (aged 0–14 years) and 1.1 million adolescents (aged 15–19 years) living with HIV globally in 2022 (UNAIDS, 2023 estimates). The majority (almost 86%) of these children reside in Sub-Saharan Africa where the primary mode of transmission for those under 15 years of age is perinatal transmission (UNICEF, Global and regional trends 2023). With access to ART, however, these children are living well into adulthood¹. HIV and its treatment with ART are associated with persistent immune activation and immune dysfunction. This may be driven by low level HIV replication, co-pathogens, gut microbial translocation and ART toxicity². This persistent immune dysregulation may contribute to an increased incidence of non-AIDS related co-morbidities in this population^3–8. Cardiovascular disease (CVD) is one of the most common causes of morbidity and mortality in people living with HIV (PLWH)^4,9–12 and is significantly more common in these individuals compared to those without HIV¹³. The CVD risk and underlying metabolic and immunologic factors in pediatric and adolescent populations perinatally infected with HIV have not been sufficiently investigated.

Inflammation and innate immune cell activation likely drives atherosclerotic cardiovascular disease¹⁴ with several studies linking monocytes/macrophages with vascular inflammation and CVD^15,16. Other innate immune cells also contribute to CVD; natural killer (NK) cells are a group of innate lymphoid cells that are important in the surveillance and clearance of viral infections and play a key role in the immunopathogenesis of HIV^17,18. NK cells are critical mediators of the innate immune response and are especially important in the pediatric population, where they take precedence during pathogen challenge over a still-developing adaptive immune response in early childhood¹⁹. There is mounting evidence that human NK cells display features of antigen specificity and memory, as depicted by their vaccination-dependent antigen-specific recall responses²⁰. Such “recall responses” may be related to trained immunity, where innate immune cells are epigenetically reprogrammed to respond differently on subsequent exposure to a microbial product. Trained immunity can be induced by microbial products such as lipopolysaccharide and β-D-glucan, which frequently enter the blood stream because of increased gut permeability in PLWH²¹.

NK cells can be classified into different subtypes based on the relative expression of CD56 and CD16 receptors and by their different functional capabilities, with higher expression of CD56 signifying increased cytokine producing potential (CD56^briCD16^dim/-), and higher expression of CD16 (CD56^dim/-CD16^bri) reflecting more cytotoxic capabilities²². Among these subpopulations, varying degrees of expression of activating (e.g., CD69, NKp44, HLA-DR) and inhibitory (e.g., NKG2A) receptors, markers signifying NK cell memory (CD57) and cytokine/chemokine receptors (e.g., CCR5, CXCR3, CXCR5) determine their general functional and migratory characteristics²³. CD57⁺NK cells, originally thought to represent an NK subset with features of maturation and memory, are now considered to possess substantial cytolytic potential and an intrinsic capability to produce IFNγ^24,25. However, their requirements for activation may differ from other NK cell subsets. In the context of infections such as CMV and HIV, expansion of CD57⁺(NKG2C⁺) NK cells represents a subset presumably induced by persistent antigenic stimulation²⁶.

The contribution of NK cells to the pathogenesis and pathophysiology of inflammatory disease has received considerable attention. Accordingly, alterations in NK cell activity have been described in various types of inflammatory disorders, including cancer, type I diabetes, periodontitis, and atherosclerosis²⁷. In the case of atherosclerosis, animal studies have revealed that changes in NK cell number and/or activity may influence a proatherogenic state specifically in the setting of chronic viral infection²⁸. Indeed, among patients with chronic cytomegalovirus infection, expansion of NKG2C⁺ NK cells has been associated with plaque instability²⁹. These observations suggest a role for NK cells in the pathogenesis of coronary artery disease among those suffering from chronic infections³⁰.

Studies have demonstrated that HIV infection, both in viremic and virally suppressed states, differentially affects protein expression among several NK cell subsets^22,31. A study among a cohort of neonates in Botswana with PHIV revealed that CD57⁺CD56^dimCD16^dim NK cells, cells with elevated antiviral and cytotoxic properties, increased over time following birth³². Among PLWH, these activated cytotoxic cells can express chemokine receptors like CCR5 and CXCR4 through trogocytosis, thereby enhancing their pro-inflammatory potential³³.

While substantial evidence exists regarding the significance of NK cell subsets in HIV, a research gap exists in understanding their role in APHIV exposed to HIV and/or ART since birth. Emerging concerns suggest that HIV’s impact on adolescents may surpass that in adults due to lifelong viral persistence, chronic immune activation, and lifelong ART exposure. Our objectives were to investigate whether persistent NK cell activation/alteration in PHIV may contribute to increased CVD risk in APHIV. In this study, we show that APHIV display early features of increased CVD risk and that this is associated with dysregulated plasma cardiometabolic biomarkers and activated, pro-inflammatory and memory-like NK cell profiles.

Results

Study design and cohort demographics

We used clinical and radiological parameters, cryopreserved peripheral blood mononuclear cells (PBMC) and plasma samples obtained from two groups: APHIV (n = 18) and a group of age- and sex-matched adolescents without HIV and who were unexposed to the virus (controls, n = 20). These specimens were collected at the Joint Clinical Research Center (JCRC) in Kampala, Uganda, between 2017 and 2021. These 38 participants represent a subset of the larger prospective cohort and were included based on samples and cardiovascular measures availability. Overall, the mean age was 14 years (SD = 1.7), 45% were females, and 14/18 APHIVs had undetectable viral load (<50 copies/mL; 1/18 had a low viral load, and viral load data was not available for 3 participants). Amongst APHIV, mean ART duration was 9.38 years, and about 44% were on a dolutegravir-based regimen, the remaining were on nevirapine, efavirenz or the boosted protease inhibitor lopinavir/ritonavir. The demographics of participants in this sub-study are similar to the parent study previously described^34,35. There were no significant differences in CMV seropositivity between the groups (Supplementary Table 1 and Supplementary Table 2). No significant differences were observed in levels of cholesterol, high-density lipoprotein, low-density lipoprotein, very low-density lipoprotein, and triglycerides between APHIV and controls. (Supplementary Table 3, Supplementary Fig. 11a to h and, Supplementary Fig. 16). To assess potential selection bias, baseline characteristics of participants included in the substudy were compared with those of participants from the parent cohort who were not included. No meaningful differences were observed with respect to age, sex, body mass index, blood pressure, or lipid parameters (Supplementary Table 6 and Supplementary Table 7 panel A, B and, C).

APHIVs show a greater cardiovascular disease risk

To assess subclinical vascular disease and CVD risk, we measured carotid intima-media thickness (IMT) (Fig. 1a) and pulse wave velocity (PWV) (Fig. 1b) among participants. While we found no significant difference in the PWV between the groups, IMT was significantly increased among APHIV (p < 0.05); this is similar to findings previously published from the entire cohort³⁶. We also measured levels of plasma biomarkers of inflammation, gut barrier function, and CVD risk in participants (Supplementary Table 4,Fig. 1c, d and Supplementary Fig. 11). Many of these markers have been linked to morbidity and mortality in PLWH (CRP, IL-6, sCD14)^37,38. Furthermore, Ox-LDL levels are associated with CVD in adults without HIV ^39,40and changes in Ox-LDL during statin treatment in adults with HIV are link to carotid IMT⁴¹. Interestingly, levels of plasma oxidized low-density lipoprotein (Ox-LDL) and soluble CD163 (sCD163) were lower among APHIV compared with levels in controls. (Figs. 1c, d).

Fig. 1. Cardiovascular- and plasma- biomarker profiling reveal increased cardiovascular disease risk in APHIV.

Schematic diagrams of the vascular measurements and group comparisons between perinatally HIV aquired adolescents (APHIV) and HIV-unexposed adolescents (HIV − ) (a) Carotid intima-media thickness (IMT; maximum IMT, mm) and (b) pulse wave velocity (PWV, m/s) are shown with accompanying schematics. c Oxidized LDL (Ox-LDL, mIU/L) and (d) soluble CD163 (sCD163, ng/mL) are shown as violin plots. Each dot represents one participant. For IMT and PWV (a, b), HIV − , n = 19; APHIV, n = 18. For Ox-LDL and sCD163 (c, d), HIV − , n = 20; APHIV, n = 18. For box-and-whisker plots (a, b), the centre line indicates the median, the box bounds indicate the 25th and 75th percentiles, and the whiskers extend to the minimum and maximum values; individual participant values are overlaid as points. For violin plots (c, d), violin width represents the distribution of participant values, with individual participant values overlaid as points. P values shown above group comparisons were calculated using a two-tailed Mann–Whitney test. In the PWV schematic, d denotes the distance between femoral and carotid recording sites and T denotes the arterial pulse wave delay (PWV = D/T). Schematic elements in panels a and b were created in BioRender.com. De Silva, I. (2026) https://BioRender.com/3jrgyex.

Comprehensive transcriptomic profiling unveils differential regulation of innate immunity and pathways relating to vascular remodeling and lipid metabolism

Given differences in circulating immune biomarkers between APHIV and adolescents without HIV, we investigated the circulating immune cell landscape of APHIV to identify broad differences that may have resulted as a result of PHIV and/or ART. We conducted an in-depth transcriptomic analysis and identified 1638 differentially expressed genes (DEGs) in PBMCs obtained from APHIV and controls. Among these statistically significant DEGs (p < 0.05), 777 were upregulated, while 861 were downregulated (Fig. 2a and Supplementary Fig. 1). In terms of innate immune activation, noteworthy upregulated DEGs included Myd88, Tlr1, Tlr2, Clec4d, Clec12b, Irak3, Itgb3bp and Tnf, while expression of Gzmm and Il6r were downregulated in APHIV. Additionally, we observed significant variations in gene expression associated with immune cell migration to inflamed tissues, including upregulation of Cxcl10, Itgb3bp, and downregulation of Itgb7, Ccr4, Cxcr6, Cxcr5, and Ccr6. Our analysis also revealed altered transcriptional profiles related to vascular wall remodeling genes like S1pr1, Flt4, and Col18a1. We observed upregulated expression of Cd36, Clec10a, and Colec12; molecules important in the uptake of Ox-LDL^42,43. Furthermore, our analysis predicted several epigenetic modifications occurring in PBMCs of APHIV. This prediction was substantiated by the high number of DEGs associated with structural modification of chromatin and DNA methylation, exemplified by genes like Hist1h1c, Hist1h2ah, and Hdac1. We conducted functional and pathway enrichment analyses to ascertain the biological significance of these DEGs. Our gene ontology (GO) analysis revealed a profound impact on innate immune activation, and vascular and lipid metabolism-related biological processes particularly involving activated pro-inflammatory innate immune cells among APHIV, including NK cell activation, NK cell degranulation, leukocyte adhesion to arterial endothelial cells, myeloid cell migration, myeloid cell differentiation, regulation of lipid storage, regulation of lipid catabolism, and vascular wall remodeling (Fig. 2b and Supplementary Fig. 2). Overall, the protein-protein interaction network analysis highlighted the enrichment of pathways specifically related to innate immunity and epigenetic regulation (Fig. 2c). Observed clustering based on the top differentially expressed genes and transcriptional pathways (Supplementary Fig. 3), underscores the phenotypical and functional distinctions that exist between APHIV compared to HIV unexposed controls, offering insights into the intricate transcriptional profiles shaping these differences.

Fig. 2. Differentially expressed genes (DEGs) and enriched transcriptional pathways of cellular biological processes of PHIV adolescents vs HIV- controls.

a Volcano plot of DEGs between PHIVs and controls (screening thresholds: p value < 0.05). b Gene Ontology (GO) analysis was performed and revealed significantly enriched pathways in PBMC samples from PHIV adolescent vs HIV unexposed participants related to biological processes (screening threshold: adjusted p value < 0.05). c Protein-protein interaction network highlighting enriched pathways among APHIVs. The top 777 upregulated genes in APHIVs and the STRING interactome database were used as input. Upregulated genes were selected with a threshold of p value < 0.05 and putative potential protein-protein interactions were predicted using an interaction confidence threshold of 0.700.

Enrichment of Memory-like CD57⁺ NK Cells and Innate Immune activation in adolescents with perinatally acquired HIV (APHIV)

Based on transcriptomic analyses which revealed predominantly innate immune dysregulation, specifically in NK cells and monocytes, we performed targeted phenotypic and functional analyses of these innate cell subsets. In line with the transcriptional findings, monocyte subset distribution was significantly altered in APHIV. Specifically, we observed elevated frequencies of CD14⁺CD16⁺ intermediate monocytes in the APHIV population (p < 0.05; Supplementary Fig. 4b). We next compared NK cell phenotypes between APHIV and control groups using an unbiased clustering approach to identify dominant phenotypic clusters based on surface marker expression relevant to our study. To achieve this, we applied FlowSOM-based clustering on CD45⁺CD3⁻CD19⁻CD14⁻ leukocytes and visualized the results using UMAP, which revealed 24 unsupervised clusters representing distinct NK cell phenotypes. However, no significant differences were observed between APHIV and controls (Supplementary Fig. 6). Subsequently, we employed a conventional gating strategy, previously described in the literature⁴⁴, to defined six NK cell subsets based on the relative expression of CD56 and CD16 receptors (Fig. 3a). Although the overall distribution of NK subsets did not differ significantly between APHIV and controls (Supplementary Fig. 7), we observed notable differences in surface immune marker expression across subsets. In the CD56^dimCD16^dim subset, APHIV exhibited increased levels of CD57, CXCR3, and NKp44, while CD69 and HLA-DR were elevated in the CD56^dimCD16^- subset. Similarly, NKp44 and HLA-DR were significantly upregulated in the CD56^briCD16^- subset (p < 0.05 for all; Fig. 3b and Supplementary Fig. 8), collectively indicating an activated NK cell phenotype.

Fig. 3. Enrichment of memory-like CD57⁺ NK cells and selective activation in adolescents with perinatally acquired HIV (APHIV).

a A representative pseudocolor flow-cytometry plot illustrates the gating strategy used to define six NK cell subsets based on CD56 and CD16 expression.b Surface expression of CD69, HLA-DR, NKp44, CD57, and CXCR3 across the indicated NK subsets in HIV-negative controls (HIV − ) and APHIV adolescents, displayed as dot plots with individual participant values overlaid (one dot = one participant). Horizontal lines indicate the median for each group. c To further characterize these changes, t-SNE projections visualize the distribution of immune markers within the CD57⁺CD56^dimCD^16dim NK cell subset, highlighting differences between the two study populations. d A focused t-SNE analysis of CXCR3 expression within this memory-like (CD57⁺CD56^dimCD^16dim)NK subset demonstrates its marked upregulation in APHIV, suggesting enhanced tissue-homing potential. e Quantification of CXCR3 within the CD57⁺CD56dimCD16dim subset shown as box-and-whisker plots with individual participant values overlaid. The centre line indicates the median, the box bounds indicate the 25th and 75th percentiles (interquartile range), and whiskers extend to the minimum and maximum values. Sample sizes were HIV − , n = 20 and APHIV, n = 18. Between-group comparisons were performed using a two-tailed Mann–Whitney test; exact P values are shown.

Among these changes, the CD57⁺CD56^dimCD16^dim NK cell subset recognized as a memory-like NK population with functional relevance in chronic viral infections including HIV^45,46, was notably enriched in APHIV adolescents (highlighted in Fig. 3d). This subset also showed marked upregulation of CXCR3 (p < 0.05; Fig. 3c–e), a chemokine receptor involved in tissue homing⁴⁷. The elevated expression of CXCR3 suggests that these memory-like NK cells are primed for migration to inflamed tissues, potentially contributing to chronic immune activation and tissue remodeling in APHIV. Notably, we found no significant differences in the expression of other chemokine receptors (CXCR5, CCR5, CCR7) or activation/inhibition markers (HLA-DR, CD69, CD27, PD-1, NKG2A) within the CD57⁺CD56^dimCD16^dim subset (Fig. 3c and Supplementary Fig. 9), indicating that the observed phenotypic changes are selective and may reflect a specialized tissue-homing program rather than broad activation. Despite these phenotypic alterations, no significant differences were observed in the expression of cytokines (IFN-γ, TNF) or cytotoxic (Granzyme B) molecules in total NK cells between the two groups (Supplementary Fig. 10).

Immune-vascular interactions reveal early atherosclerotic risk in adolescents with perinatal HIV

Next, we explored the relationships among innate immune cell subsets, plasma biomarkers, and cardiovascular disease (CVD) indicators, specifically carotid intima-media thickness (IMT) and pulse wave velocity (PWV), by applying eXtreme Gradient Boosting (XGBoost) models. Using gain as a feature importance metric we quantified the contribution of a given feature to the model’s predictive accuracy. A higher gain value indicates that the feature is more influential in shaping the model’s predictions, such as IMT max or PWV in this study. Accordingly, the CXCR5⁺ CD56^bri CD16^lo subset, shows the highest gain among innate immune markers after CRP. Interestingly, CXCR3⁺CD56^dimCD16^dim NK cells, which were found to be increased in APHIV (compared to controls) had high gain and was therefore considered influential on IMT (Fig. 4a). As CXCR3 mediates cell migration to inflamed tissue⁴⁸, including atherogenic vascular tissue sites, this finding suggests that these NK subsets may contribute to the early pathogenesis of atherosclerotic CVD in A PHIV. We also noted an association between β-D-glucan and IMT. Intermediate monocytes (IM) were significantly associated with elevated PWV in APHIV (Fig. 4b); these cells are known to be correlated with atherosclerotic CVD⁴⁹. Several NK subsets were also found to have higher gain and therefore be important to PWV, including the CD69⁺CD56^dimCD16^- NK subset which was enriched in the APHIV cohort. Given that CD69 is a marker of NK cell activation, this may reflect the potential role of activated NK cells in modifying vascular architecture and promoting atherosclerosis⁵⁰.

Fig. 4. Machine learning-based statistical algorithms were used to explore relationships between markers of cardiovascular risk and innate immune signatures in APHIV.

XGBoost algorithm model-based associations between immune signatures/cardiometabolic biomarkers and (a) carotid IMT and (b) PWV in the APHIV population. Glasso network plots visualizing the dependencies among the cardiometabolic and immune marker networks for (c) IMT max and (d) PWV. The size of each node in the network plot is proportional to the magnitude of feature importance for the dependent cardiovascular risk, as determined by XGBoost. The colors of the edges reflect the sign of dependencies: pink for positive and blue for negative. The black borders of the nodes indicate biomarkers that are important for distinguishing HIV status based on OPLS-DA analysis.

Figures 4c, d were generated using the Glasso (Graphical Lasso) method with modifications. In addition to visualizing significant independencies among biomarkers and covariates, we included nodes for biomarkers with weak or no dependencies (threshold <0.4). This approach allows us to illustrate the relationships between biomarkers identified as important contributors to cardiovascular outcomes by XGBoost, while adjusting for the covariates. Furthermore, we found several positive associations between monocytes and activated, pro-inflammatory NK cell subsets for both IMT and PWV which may suggest physiological interactions between these cell subsets leading to the promotion of atherosclerotic CVD in the APHIV population.

Distinct relationships are observed between Ox-LDL and NK cell subsets

Given its associations with atherosclerotic cardiovascular disease and its unexpected reduction in plasma Ox-LDL among APHIV, we performed a comprehensive correlational analysis between plasma Ox-LDL and NK subsets. Pearson correlation coefficient analysis showed significant associations between NK cell subsets and Ox-LDL in APHIV (Fig. 5a). A strong negative correlation was observed between CCR5⁺ NK cell subsets (Fig. 5b–e) and Ox-LDL levels, specifically within the APHIV cohort. Of these, NK subsets including CD65^dimCD16^- and CD65^dimCD16^dim as demonstrated previously are of an activated phenotype among APHIV. These activated NK cells express the chemokine receptor CCR5 and can potentially migrate towards vascular tissue that express the corresponding chemokine CCL5⁵¹. Murine studies have demonstrated an abundance of CCL5 and a high proportion of CCR5-expressing immune cells in the atherogenic plaque microenvironments^52,53,supporting the relevance of this axis. Furthermore, the memory-like NK cell subset CD57⁺ CD56^dimCD16^dim, which was increased in APHIV exhibited robust negative correlations with Ox-LDL (Fig. 5f), providing deeper insights into a likely relationship between activated, memory-like NK cells and Ox-LDL dynamics.

Fig. 5. Associations between activated, pro-inflammatory NK cell subsets and plasma Ox-LDL levels among PHIV adolescents.

a Heatmap showing Pearson correlation coefficient (r) between NK cell subsets and plasma Ox-LDL in HIV- and APHIV. Colors ranging from blue to red depict the strength of the correlation, with asterisks indicating significant correlations (two-sided p < 0.05). b–e Scatter plots depicting Pearson correlation analysis between plasma Ox-LDL and CCR5⁺ NK cell subsets in APHIV (one dot per participant): (b)CD56^dimCD16^- NK cell subset, (c) CD56^-CD16^dim NK cell, (d) CD56^dimCD16^dim NK cell and (e) CD56^briCD16^lo NK cell subsets. f Scatter plot depicting Pearson correlation analysis between plasma Ox-LDL and CD57⁺CD56^dimCD16^dim NK cell subset in APHIV. Pearson r and exact two-sided P values are shown in each panel. g Schematic representation of an in-vitro experiment designed to investigate the role of activated NK cells in Ox-LDL uptake by monocyte-derived macrophages (MDMs). h Histogram depicting the impact of activated NK cells with and without NKG2D receptor blocking on Ox-LDL uptake by macrophages, assessed using flow cytometry. i Quantification of Dil-Ox-LDL uptake by MDMs under the indicated conditions (presence of activated NK cells, with and without NKG2D antibodies), plotted as paired donor-matched measurements (each line connects values from the same donor; n = 6 independent donors, D1–D6). Two-sided Wilcoxon signed-rank tests were used for paired comparisons; exact P values are shown (P < 0.0001 and P = 0.001, as indicated). j Transwell co-culture experiments physically separating NK cells from MDMs while allowing diffusion of soluble factors (n = 3 independent donors, as labelled), showing that separation abrogates the increase in Dil–Ox-LDL uptake, consistent with a requirement for direct NK–MDM contact. For ex vivo analyses (a–f), HIV− (n = 20) and APHIV (n = 18), where n denotes independent participants. For in vitro assays (g–j), n denotes independent donors. Schematic elements in panels g was created in BioRender.com. De Silva, I. (2026) https://BioRender.com/plw7ylq.

To assess whether these relationships were specific to APHIV, we performed Pearson Correlation Coefficient Analysis between Ox-LDL and the above-mentioned NK cells subsets in controls (Supplementary Fig. 12). While the direction of the relationship was similar, no statistically significant associations were observed, indicating that these immune-Ox-LDL interactions may be unique to the APHIV context.

Based on these findings, we hypothesize that activated, memory-like NK cells in the vascular intima may promote Ox-LDL uptake by macrophages, leading to foam cell generation and initiating atherogenesis. To test this as a possible mechanism (Fig. 5g), we isolated NK cells from PBMCs of HIV negative donors (n = 6) and stimulated these cells using IL-12, IL-15, and IL-18, inducing their differentiation into activated, memory-like NK cells with enhanced functionality and effector responses^54–56. Concurrently, we differentiated autologous monocyte-derived macrophages (MDMs) from peripheral blood as described previously⁵⁷. After six days, we incubated these MDMs in a medium containing commercially prepared Dil-labeled Ox-LDL for four hours, in the presence and absence of activated autologous NK cells. In all donor samples tested, we observed significantly higher Dil-Ox-LDL uptake in the presence of activated, memory-like NK cells (Wilcoxon signed-rank test; p < 0.05 for all) (Fig. 5g–I, Supplementary Fig. 14, 15). Mechanistically, the enhanced effector responses of memory-like NK cells are heavily influenced by the activating receptor NKG2D and blocking this receptor has been found to significantly dampen memory-like NK cell function⁵⁸. To further investigate its role in activated, memory-like NK cell mediated Ox-LDL uptake by macrophages, we introduced blocking antibodies against NKG2D while co-culturing NK cells with MDMs. Blocking NKG2D reduced Ox-LDL uptake in all tested samples (p < 0.05). To assess the importance of direct cell-to-cell contact, we repeated the co-culture experiments using trans well plates, which physically separate NK cells from MDMs while allowing soluble factors to diffuse (Supplementary Fig. 14, Fig. 5j). Notably, Ox-LDL uptake by MDMs was significantly reduced under these conditions, confirming that direct interaction between NK cells and macrophages plays a pivotal role in facilitating Ox-LDL uptake and potentially initiating foam cell formation.

These findings suggest a role for activated, memory-like NK cells in enhancing Ox-LDL uptake by MDMs and also suggest that the NK cell receptor NKG2D may play an essential role in this process. This adds to previous reports of HIV-associated increased uptake of Ox-LDL by MDMs that may be mediated by high TNF levels; TNF also upregulates major histocompatibility complex class I chain-related protein A (MICA), an NKG2D ligand that is expressed in macrophages in the intimal tissue environment^59,60.

Discussion

In this study we found an elevated risk of cardiovascular disease (CVD) among our APHIV population in Uganda, compared to age and sex-matched without HIV controls, as evidenced by a significant increase in carotid intima-media thickness (IMT). Furthermore, our comprehensive analysis of the major innate immune cells (i.e., NK cells and monocytes) in both groups revealed activation of multiple NK cell subsets, characterized by increased expression of CD69, NKp44 and HLA DR, along with enhanced memory-like phenotypes (CD57) in the APHIV cohort (Fig. 3). In addition, correlational analyses (Figs. 4, 5) revealed a likely association between activated, pro-inflammatory and memory-like NK cell subsets and increased CVD risk, potentially mediated through their influence on tissue Ox-LDL uptake. The above findings were supported by bulk RNA sequencing data of PBMCs which revealed the enrichment of several DEGs and transcriptional pathways related to NK cell activation, innate immune activation, lipid metabolism and vessel wall remodeling (Fig. 2).

Endogenous metabolites that are abundant in the plaque microenvironment, including Ox-LDL, train the generation of pro-atherogenic monocytes and macrophages²¹. Additionally, dominant NK cell subsets in our APHIV cohort expressing high levels of memory markers like CD57⁴⁷ may suggest training of these circulating NK cells possibly through epigenetic modification, and this is supported by results from our bulk RNA-seq which hint at epigenetic modifications occurring in the leukocytes of APHIV. In the case of APHIV, these potentially trained memory-like NK cells and monocytes/macrophages (either by persistent viral replication, gut microbial translocation, co-pathogens, ART and/or Ox-LDL) may be key innate immune cells involved in increasing atherosclerotic CVD risk⁶¹. NK cells have traditionally been considered as important mediators of anti-tumor⁶² and anti-viral⁶³ immunity. More recently, NK cells have been also linked to various pathophysiological disease processes including cardiovascular diseases⁶⁴ and atherosclerosis⁵⁰. Bonaccorsi et al. studied phenotypic and functional characteristics of NK cells within the atherosclerotic plaque environment and identified increased frequencies of CD56^briperforin^lo NK cells, expressing tissue-resident markers such as CD103, CD69 and CD49a, and producing increased amounts of IFNγ⁶⁵. Chronic HIV infection results in a persistent pro-inflammatory state in people living with HIV, even during viral suppression by ART. Consistent with prior studies^66,67, we observed increased activation across multiple NK subsets in the APHIV cohort. These cells had increased expression of activation markers such as CD69 (also a marker of tissue residency), NKp44 and HLA-DR in many NK subpopulations and markers suggesting NK memory (CD57). Our RNA-seq data further confirmed upregulated transcription of signaling adaptors like Myd88 which is critical for NK and myeloid cell activation⁶⁸, along with enriched GO pathways pertaining to biological processes involving NK cell activation. We hypothesize that these activated, memory-like NK subsets, through expression of chemokine receptors such as CXCR3 (CD57⁺CXCR3⁺ NK cells, Fig. 3 c–e) may play a role in the pathogenesis of cardiometabolic disease among the APHIV population. We are able to compare our findings to data from Rebuffet et al., which highlighted the presence of three major NK cell subpopulations with unique pathophysiological functions⁶⁹. The NK3 subset, characterized by high CD57 expression, exhibited transcriptional and protein profiles resembling adaptive NK cells, similar to the CD57⁺CD56^dimCD16^dim NK subset enriched in our APHIV cohort. These cells were also associated with low plasma Ox-LDL levels suggesting the importance of adaptive NK phenotypes in modulating lipid metabolism and CVD risk in APHIV. The NK2 subset, which is comparable to our CD56^bri/+CD16^- NK cells, demonstrated enhanced activation (higher CD69, NKp44 and HLADR expression) and tissue infiltration potential, consistent with our finding in APHIV. The strong associations we found between several of these subsets with enhanced chemotactic potential (high CCR5 expression) and plasma Ox-LDL (Fig. 5) in APHIVs suggest that these activated NK subsets may migrate to inflamed vascular tissue, where they interact with monocytes/macrophages to influence lipid uptake and foam cell generation.

In adult populations, PLWH have elevated arterial wall inflammation compared to levels in people without HIV¹⁶. Activated myeloid cells within vascular tissue secrete various chemokines such as RANTES/CCL5, CCL4, CX3CL1, CXCL10⁷⁰. Murine studies have demonstrated an abundance of CCL5 and a high proportion of the chemokine receptor CCR5-expressing immune cells in the atherogenic plaque microenvironment^52,71. Several activated and memory-like NK subsets that were observed among APHIV are pro-inflammatory in nature and through expression of the chemokine receptor CCR5, these cells can potentially migrate towards vascular tissue that express the corresponding chemokine CCL5⁵¹. It is likely that the enriched subsets of vascular tissue-homing pro-inflammatory memory-like NK cells observed among APHIV in our study likely express high levels of additional activating receptors like NKG2D⁷². As suggested by our in vitro experiments, these cells may engage in NKG2D-mediated crosstalk with vascular tissue-resident macrophages to promote their increased uptake of Ox-LDL. The negative correlations we observed between several of these activated NK subsets among APHIV expressing CCR5 and plasma Ox-LDL suggest a potential mechanism that may contribute to the low plasma Ox-LDL levels observed among our APHIV cohort. This may result in the increased likelihood of atherogenesis and subsequent increase in IMT among these individuals. Thickening of the arterial wall, which is a structural change, is one of the earliest detectable stages in atherogenesis, and, in our APHIV cohort, may have preceded the development of significantly increased arterial stiffness measured by PWV^73,74. This supports our observation of significantly elevated IMT without a significant change in PWV in our APHIV cohort.

As LDL circulates in the plasma, a portion traverses the subendothelial space, where it undergoes oxidation. At the arterial wall end, Ox-LDL is consumed by macrophages, leading to foam cell generation and triggering atherogenesis (Fig. 6). Additionally, some Ox-LDL may spill over into the circulation⁷⁵. Lower circulating Ox-LDL levels could result from either decreased production or increased consumption by arterial macrophages. CD163, a receptor of monocytes / macrophages, is shed as soluble CD163 (sCD163), indicating monocyte / macrophage activation. In adults without HIV, elevated sCD163 levels are associated with increased CVD risk. However, this has not been investigated in the pediatric and adolescent population⁷⁶. It is unclear whether sCD163 is differentially associated with CVD risk in APHIV, warranting further investigation. Notably, the lower levels of sCD163, β-D-glucan, and oxidized LDL observed in APWH in this study are consistent with findings reported in a prior large analysis of the samecohort³⁴.

Fig. 6. Conceptual model of lipid interactions driving atherogenesis in HIV-infected individuals.

Schematic illustrating the proposed mechanisms linking altered lipid metabolism and innate immune activation to accelerated atherogenesis in individuals. In healthy conditions, low-density lipoprotein (LDL) undergoes limited oxidation, and resident macrophages (Mφ) and natural killer (NK) cells maintain vascular homeostasis. In PHIV, increased LDL oxidation leads to accumulation of oxidized LDL (OxLDL), promoting enhanced uptake by macrophages and formation of lipid-laden foam cells. Concurrently, NK cells exhibit an activated, pro-inflammatory phenotype (e.g., CD69⁺, CXCR3⁺), contributing to endothelial inflammation and amplifying vascular immune responses. With aging into adulthood, comorbid factors including diabetes, hypertension, smoking, sedentary lifestyle, and dyslipidemia further exacerbate lipid dysregulation, increasing OxLDL levels and altering the lipid landscape. These changes promote macrophage lipid uptake, impaired egress, and sustained inflammation, ultimately accelerating atherogenesis compared to healthy individuals. This panel was created in BioRender.com. I. De Silva, I. (2026) https://BioRender.com/ xvm4e93.

We also found a significant increase in the levels of circulating intermediate monocytes among our APHIVs, along with the enriched GO terms involving mononuclear cell proliferation, differentiation and migration, and regulation of lipid storage. Previous work in adults with HIV link increased proportions of intermediate monocytes with monocyte derived macrophage profiles that are proinflammatory, pro-coagulant and have increased capacity for lipid uptake⁵⁷. Interestingly, we also report upregulated transcription of CD36 in our APHIV cohort, which encodes for the receptor mediating uptake of Ox-LDL into macrophages. Results from our in vitro studies suggest that activated, memory-like NK cells can potentially influence uptake of Ox-LDL by macrophages in HIV.

The hypothesis that macrophages actively interact with NK cells in the atherosclerotic lesion has been under investigation and is supported by the fact that ligands like major histocompatibility complex class I chain-related protein (MIC)⁷² to the NK cell-activating receptor, NKG2D, have been found to be expressed on foam cells derived from macrophages exposed to Ox-LDL⁷⁷. Murine study has shown that preventing NKG2D/ligand interaction suppresses plaque formation in ApoE2/2 mice⁷⁸. However, since NKG2D has also been found on subsets of T lymphocytes⁷², T cells may also contribute to atherosclerotic plaque formation in this manner (as mentioned above), in our analysis of PBMC, no significant differences were observed in CD57, CD69, or PD-1 expression in either CD4⁺ or CD8⁺ T cells between APHIV and HIV-negative adolescents (Supplementary Fig. 13). Our in vitro assay supports the potential of NKG2D-mediated interactions between activated, memory-like NK cells and macrophages in increasing macrophage uptake of Ox-LDL. This further supports the strong associations we observed between plasma Ox-LDL levels and activated NK cell subsets with migratory potential in our APHIV.

The Mean ART duration in our cohort was ~9.5 years, with 90% of the participants initiating ART between the ages of 2 and 10 years. We have previously found that ART duration is not associated with any of the measured inflammatory biomarkers or immune cell types³⁴. A recent study in Kenya compared early vs late ART initiation in children with PHIV, showing that early initiators had higher naïve-to-effector memory T cell ratios, lower stress marker expression, and improved antiviral responses, while late initiators had reduced CD4⁺ T cell percentages and elevated non-classical monocytes⁷⁹. These findings suggest that our cohort, which initiated ART relatively late, may have persistent immune dysregulation despite long-term viral suppression.

The interplay between lipids, inflammation and cardiovascular disease is likely complex and age-dependent. Adolescence and adulthood represent two different physiological states⁸⁰, and these differences become more pronounced due to common pathologies in adulthood such as diabetes, hypertension and dyslipidemia, and modifications to lifestyle such as smoking and a sedentary lifestyle. While type 2 diabetes can influence lipidome alterations and oxidation of LDL⁸¹, hypertension influences LDL transport in the arterial wall⁸², and smoking enhances oxidation of LDL⁸³. Indeed, high levels of circulating Ox-LDL have been associated with blood pressure, fasting glucose and lipid levels⁸⁴. Hence, the collective influence of adult physiology and these common pathological states may render their lipid landscape vastly different to that seen among adolescents. We have previously shown that Ox-LDL levels are increased in adults with HIV and linked levels of this pro-inflammatory lipid molecule to monocyte activation. Further, decreases in Ox-LDL were directly related to improvements in monocyte activation and IMT in adults with HIV following statin therapy⁴¹. We have also reported associations among immune activation, carotid IMT, and elevated levels of pro-inflammatory lipid classes and species among the full cohort of APHIV⁸⁵. This suggests that numerous mechanisms collectively contribute towards a proatherogenic state among APHIVs and may warrant further investigation. In parallel with these findings, we also observed a role for NK cells and monocytes in promoting atherogenesis in patients with severe COVID-19, further supporting the relevance of innate immune dysregulation in virus-associated cardiovascular pathology⁸⁶.

We aim to frame the findings of this study within an exploratory and hypothesis-generating context. While we use immunophenotyping, transcriptomics, and in-vitro assays to demonstrate associations between innate immune activation and cardiovascular risk in APHIV, the study design is predominantly observational and correlative. The strong correlations we observed between activated, memory-like NK cell subsets and plasma Ox-LDL do not establish causality but rather signify a potential mechanism for atherogenesis occurring in HIV. Bulk RNA-seq data builds on this hypothesis by revealing differences in NK cell activation, lipid metabolism, and epigenetic regulation among APHIVs. Furthermore, through our in vitro experiments, we provide a mechanistic model that suggests a role for NK cells in promoting Ox-LDL uptake by macrophages, although this requires further validation in subsequent studies. Hence, this work should be considered as a foundational framework that can be used to explore in depth innate immune dysregulation in APHIV and its impact on atherosclerotic cardiovascular disease.

Key questions arise from our findings, including: 1) whether the generation of trained NK cells in APHIV is driven by persistent antigenic exposure (i.e., viral, or products of gut microbial translocation) and persist due to epigenetic reprogramming; 2) what are the specific dynamics of the proposed NK cell-macrophage crosstalk within the arterial wall microenvironment in APHIV, which could be explored using advanced models or targeted imaging studies; and 3) can these expanded NK cell subsets, alterations in plasma proteins, and enriched transcriptional signatures serve as predictive biomarkers for the progression of subclinical atherosclerotic CVD in APHIV? Addressing these questions will be critical for determining the therapeutic potential of targeting these innate immune pathways through immunomodulatory strategies to mitigate the CVD risk in this vulnerable and aging population.

Our study has several limitations. First, the study was conducted using a small sample number which may not accurately reflect the population of APHIV. Second, due to the lack of access to vascular tissue samples, all studies were conducted on PBMCs, and this may not be truly representative of the immunopathology in the vessel wall. Third, while supported by in-vitro experiments, many of our conclusions of the study are correlative in nature. In addition, although no baseline differences in lipid levels, blood pressure, or measured socioeconomic indicators were detected between groups, residual confounding by unmeasured or imperfectly captured socioeconomic factors cannot be fully excluded. Despite these limitations, we show that APHIV display early signs of atherosclerotic cardiovascular risk, and that activated pro-inflammatory NK cell phenotypes with memory-like features may be involved in its pathogenesis. This is concerning for these individuals who are likely to develop early onset CVDs. Furthermore, by identifying a possible mechanism for atherosclerotic CVD development through the involvement of activated NK cells, we hope that our findings would stimulate further investigations that would lead to development of therapeutic or prophylactic strategies to mitigate CVDs risk in APHIV.

Methods

Study participants and selection

This study represents a cross-sectional analysis nested within a prospective observational cohort, using data collected at a single study visit. Participants were enrolled at the Joint Clinical Research Center (JCRC) in Kampala, Uganda between 2017-2021 as previously described^34,35. Participants (APHIV and controls) were prospectively enrolled in the parent cohort, and for this analysis only those with available PBMCs, plasma and cardiovascular measures were included. All participants were 10–18 years of age and as per Ugandan research guidelines, provided written informed assent. By design, APHIV participants (n = 18; median age =14.4 years) were on ART for at least 2 years with a stable regimen for at least the last 6 months with the majority (83.3%) having low or undetectable (<50 copies/mL) HIV-1 RNA. Self-reported co-infections and malnutrition were exclusionary. Adolescents with pregnancy or intent to become pregnant were excluded. The control group consisted of age/sex matched adolescents from the same geographical location (n = 20; median age=14.2 years) who had not been exposed to HIV in utero. None of the participants had active tuberculosis or diabetes mellitus or were on anti-inflammatory or cholesterol lowering medication at the time of sample collection. The study was approved by the Research Ethics Committee in Uganda and IRB of University Hospitals Cleveland Medical Center, The Ohio State University and Ann and Robert Lurie Children’s Hospital. All participants provided written informed assent per the local Ugandan IRB guidelines.

Cell collection and processing

Participants were seen for blood draw obtained after an 8 h fast. Plasma, serum, and PBMCs were cryopreserved and shipped to University Hospitals Cleveland Medical Center, Cleveland, Ohio. All assays were performed in batches and without prior thaw. For this analysis, we included participants with plasma, serum and PBMCs available at the same time point.

Carotid intima-media thickness and pulse wave velocity assessment

An experienced ultrasonographer performed all IMT and PWV at the JCRC following published guidelines and as previously described^35,36. Briefly, B mode ultrasound scan of the carotid arteries was performed using a Philips iU22 ultrasound system with 12-3 MHz broadband linear array probe (Philips, Andover, MA). Measurements obtained from the right and left sides were then averaged and reported as a single mean-mean IMT (subsequently referred to as mean IMT) and mean-max IMT (referred to as max IMT). PWV was measured by carotid to femoral applanation tonometry (Vicorder, SMT Medical, Wurzburg, Germany). In brief, the carotid to femoral path length was measured, the tonometer was applied to the carotid artery and femoral artery in sequence to obtain waveforms in relationship to the ECG tracing. An average of three measurements was used for analyses. Higher velocity measurements correspond to greater arterial stiffness.

Flow cytometry cell staining and analysis

We assessed innate and adaptive immune cell surface markers in APHIV and adolescents without HIV (controls). Cryopreserved PBMCs were resuspended at 1–2 million cells per ml in R10 media (RPMI-1640 supplemented with 10% FBS [fetal bovine serum],2mM l-glutamine, 100 U/mL penicillin, and 100 mg/mL streptomycin). Cells were then pelleted by centrifugation at 700 × g for 5 min, the supernatant was decanted, and the cell pellet was resuspended in 100 μL of PBS. To discriminate dead cells, samples were stained with Fixability Viability Dye (Zombie NIR Cat No. 423105) and incubated for 15 min. Cells were then washed, followed by the addition of a surface antibody cocktail, the antibodies being previously titrated to determine the optimal concentration (antibodies used in this study are listed in Supplementary Table 5). After a 30 min incubation period cells were washed once again (All washing steps were carried out at 700 g for 5 min at 4 °C). Following the filtering of cells through strainer capped FACS tubes, samples were acquired on a Cytek Aurora flow cytometer. Spectral flow cytometry data were analyzed using FlowJo software (Tree Star). We obtained “fluorescence minus 1” controls (cells stained with all fluorochromes used in the experiment except 1) for each marker prior to spectral unmixing (Supplementary Fig. 5 b).

Measurement of plasma biomarkers

The plasma collected from blood samples were stored at −80 °C and batched until processing without a prior thaw. Then, using an enzyme-linked immunosorbent assay (ELISA), we measured the following inflammatory biomarkers in the plasma of APHIV adolescents and controls: soluble CD14 (sCD14), C reactive protein (CRP), oxidized low-density lipoprotein (Ox-LDL), soluble CD163 (sCD163), 1,3-β-D-glucan (BDG), intestinal fatty acid binding protein (IFAB) and high-sensitivity interleukin-6 (hsIL6)⁸⁷.

Data analysis using machine learning

For a robust analysis, eXtreme Gradient Boosting (XGBoost) algorithm models were used to assess the association between immune biomarkers and PWV/IMT-Max. The models were adjusted for the covariates of interest. The set of XGBoost parameters was tuned simultaneously and optimized through cross-validation. The PWV/IMT-Max attribution to each biomarker was evaluated by the mean of each feature importance based on the XGBoost results. We used Graphic Lasso (Glasso) to learn the structure and visualize the dependencies among the biomarker networks. We focused on those biomarkers with feature importance scale greater than 0.01 based on XGBoost results and regularization parameter rho above 0.1 based on the Glasso results. Additionally, we conducted orthogonal partial least squares discriminant analysis (OPLS-DA) to identify biomarkers that may distinguish between controls and APHIV. These biomarkers were subsequently displayed in the Glasso plot.

In vitro assay for Ox-LDL uptake by monocyte-derived macrophages (MDMs)

We conducted an in vitro experiment to investigate the impact of activated NK cells on the uptake of Ox-LDL by MDMs to substantiate our in vivo observations. Venous blood samples were collected in heparinized tubes in order to isolate PBMCs via Ficoll gradient centrifugation. NK cells were then isolated from these PBMCs through negative selection using the Miltenyi Biotec NK cell isolation kit (130-092-657). Isolated NK cells were incubated in vitro with interleukin 12 (IL-12), IL-15 and IL-18 (each added at 10 ng/mL) overnight⁵⁴ to induce activation. Given that we found predominantly activated phenotypes of NK cells and their negative association with lowered Ox-LDL in APHIVs, the main objective of this assay was to demonstrate differential influence of activated NK cells in the uptake of Ox-LDL by macrophages. Additional PBMCs isolated from the same donor were rested and incubated in Teflon-coated wells for up to six days, to induce differentiation of Monocytes into macrophages (MDMs)⁵⁷. After six days, we introduced these MDMs to a medium containing commercially manufactured Ox-LDL that had been tagged with Dil dye (10 ug/mL) to measure its uptake into MDMs during a four-hour incubation period both in the presence and absence of activated memory-like NK cells. Cells were washed twice with PBS, and the intensity of DiI staining was measured by flow cytometry. To identify MDMs, CD3, CD19, and CD14 surface immune markers were used. The above experiment was also carried out in the presence of NKG2D blocking antibodies (added during the four-hour incubation step with along with activated NK cells) to investigate the mechanism of NK cell-influence on MDMs to impact their uptake of Ox-LDL.

RNA Extraction and QC

Cell viability was measured prior to RNA extraction for PBMC samples using the Countess II FL Automated Cell Counter (ThermoFisher), and RNA extraction was performed using the RNeasy mini kit (Qiagen). The quality of all RNA was determined using the RNA integrity number as calculated from the Agilent Fragment Analyzer with the SS RNA assay kit (Agilent) while the concentration was determined by NanoDrop (ThermoFisher).

Bulk RNA sequencing

Total RNA prepared above was normalized to 100 ng input for library preparation with the TruSeq Stranded Total RNA with Ribo-Zero Globin kit (Illumina). The resulting libraries were assessed on the Agilent Fragment Analyzer with the HS NGS assay (Agilent) and quantified using the NEBNext Library Quant Kit for Illumina (New England Biolabs, Inc.) on an Applied Biosystems QuantStudio 7 Flex Real-Time PCR (ThermoFisher). Sequencing (30 million reads per sample) was performed with Novaseq (Illumina) on one SP flow cell as a 100-cycle paired-end run.

RNA-seq analysis

Raw demultiplexed fastq paired end read files were trimmed of adapters and filtered using the program skewer⁸⁸ to throw out any with an average phred quality score of less than 30 or a length of less than 10. Trimmed reads were then aligned using the STAR⁸⁹ aligner to the Homo sapiens NCBI reference genome assembly version GRCh38 and sorted using SAMtools⁹⁰. Aligned reads were counted and assigned to gene meta-features using the program featureCounts⁹¹ as part of the Subread package. These count files were imported into the R programming language and were assessed for quality control, normalized and analyzed using an in-house pipeline utilizing the limma voom method with quantile normalization⁹² for differential gene expression analysis. Gene set variation analysis was performed using the GSVA Bioconductor library and the Molecular Signatures Database v5.0^93,94. Protein-protein interaction visualizations were generated using Stringdb⁹⁵. Gene Ontology (GO) enrichment analysis was performed to gain insights into the biological processes associated with a set of genes of interest. The analysis was carried out using the clusterProfiler package in R (version 4.2.1). The significance threshold for enriched GO terms was set at a p-value cutoff of 0.05, and the Benjamini-Hochberg method was applied for multiple testing correction using GO terms with a q-value (adjusted p-value) below 0.05 were considered statistically significant. Cytoscape software (version 3.10.0) was used to visualize significant GO terms⁹⁶.

Statistical analyses

Graphs and heatmaps were prepared using Graph Pad Prism (version 9.3.1), while dot plots, contour plots and t-SNE dimensionality reduction plots were generated using FlowJo (version 10.8.2). Data were statistically analyzed using their bundled software. Statistical comparisons involving more than two experimental groups were performed using ANOVA followed by Tukey’s multiple comparison test for pairwise comparisons. Comparisons between two groups were performed using the Wilcoxon rank-sum test. To analyze the relationship between immune signatures and plasma biomarkers, the Pearson correlation coefficient was calculated and plotted on a correlogram using R 4.2.1. Scatter plots showing linear relationships between data were created for significant correlations.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

M.A. and M.G. performed experiments, analyzed data, and wrote the manuscript. A.K. and K.A. performed experiments. V.M. and C.K. oversaw and performed data and sample collection. B.R., B.T., and M.R., performed experiments and conducted transcriptome data analysis. W.M., I.S. and, D.S. performed in vitro assays and data analysis. G.A.M. helped design the parent study and provided oversight of the cardiovascular measures. D.K. supervised data analysis. T.D. provided intellectual input and contributed to study design. S.S. performed the machine learning analyses and data interpretation. C.C. and M.C. supervised transcriptome data analysis and edited the manuscript. N.F. supervised lipid biomarker experiments and edited the manuscript. S.D.F. led the clinical cohort and edited the manuscript. N.L. led the overall study, oversaw the study design and execution, supervised data analysis, and led manuscript preparation.

Peer review

Peer review information

Nature Communications thanks Dasja Pajkrt and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by National Institutes of Health (NIH/NIAID) grant U01 AI168630-01 awarded to N.L., N.F., S.D.F.

Data availability

The RNA-seq / transcriptomic / gene expression] data generated in this study have been deposited in the NCBI Gene Expression Omnibus (GEO) under accession code GSE270500 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE270500). The processed data supporting the findings of this study (including summary statistics used to generate the figures) are provided in the Supplementary Information/Source Data file. Individual-level raw data derived from human participants (for example, raw flow cytometry files and/or linked clinical metadata) are available under restricted access due to participant privacy and ethical restrictions. Access can be obtained by submitting a request to the corresponding author and completing a data use agreement, subject to approval by the relevant institutional review/ethics committee(s). All other data are available in the article and its Supplementary files or from the corresponding author upon request.  Source data are provided with this paper.

Code availability

Custom code used for data processing and statistical analyses is available at GitHub: https://github.com/gunasena1/PHIV_nature.git. The code has also been archived at Zenodo with the 10.5281/zenodo.20043319. The version used in this study is release v1.1 (commit: d6148b3).

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-74602-y.