HIV is associated with an increased risk of age-related clonal hematopoiesis among older adults

People with human immunodeficiency virus (HIV) have higher rates of certain comorbidities, particularly cardiovascular disease and cancer, than people without HIV1–5. In view of observations that somatic mutations associated with age-related clonal hematopoiesis (CH) are linked to similar comorbidities in the general population6–10, we hypothesized that CH may be more prevalent in people with HIV. To address this issue, we established a prospective cohort study, the ARCHIVE study (NCT04641013), in which 220 HIV-positive and 226 HIV-negative participants aged 55 years or older were recruited in Australia. Demographic characteristics, clinical data and peripheral blood were collected to assess the presence of CH mutations and to identify potential risk factors for and clinical sequelae of CH. In total, 135 CH mutations were identified in 100 (22.4%) of 446 participants. CH was more prevalent in HIV-positive participants than in HIV-negative participants (28.2% versus 16.8%, P = 0.004), overall and across all age groups; the adjusted odds ratio for having CH in those with HIV was 2.16 (95% confidence interval 1.34–3.48, P = 0.002). The most common genes mutated overall were DNMT3A (47.4%), TET2 (20.0%) and ASXL1 (13.3%). CH and HIV infection were independently associated with increases in blood parameters and biomarkers associated with inflammation. These data suggest a selective advantage for the emergence of CH in the context of chronic infection and inflammation related to HIV infection. In a prospective cohort study, HIV-positive participants have a higher incidence of clonal hematopoiesis than HIV-negative ones, implicating clonal hematopoiesis in the increased risks of individuals with HIV for malignancy and cardiovascular disease.

People with human immunodeficiency virus (HIV) have higher rates of certain comorbidities, particularly cardiovascular disease and cancer, than people without HIV [1][2][3][4][5] . In view of observations that somatic mutations associated with age-related clonal hematopoiesis (CH) are linked to similar comorbidities in the general population 6-10 , we hypothesized that CH may be more prevalent in people with HIV. To address this issue, we established a prospective cohort study, the ARCHIVE study (NCT04641013), in which 220 HIV-positive and 226 HIV-negative participants aged 55 years or older were recruited in Australia. Demographic characteristics, clinical data and peripheral blood were collected to assess the presence of CH mutations and to identify potential risk factors for and clinical sequelae of CH. In total, 135 CH mutations were identified in 100 (22.4%) of 446 participants. CH was more prevalent in HIV-positive participants than in HIV-negative participants (28.2% versus 16.8%, P = 0.004), overall and across all age groups; the adjusted odds ratio for having CH in those with HIV was 2.16 (95% confidence interval 1.34-3.48, P = 0.002). The most common genes mutated overall were DNMT3A (47.4%), TET2 (20.0%) and ASXL1 (13.3%).

CH and HIV infection were independently associated with increases in blood parameters and biomarkers associated with inflammation. These data suggest a selective advantage for the emergence of CH in the context of chronic infection and inflammation related to HIV infection.
As effective and well-tolerated modern antiretroviral therapy (ART) regimens are becoming widely available in many settings, people with HIV who receive optimal ART are living longer 11 with life expectancies approaching those of people without HIV 1,12 . However, HIV infection remains associated with an increased prevalence of comorbidities 1,2 such as cardiovascular disease [2][3][4] and non-AIDS-defining malignancies 3,5 . While the reasons for this are not clear, increased comorbidities are found even in people with HIV who are virologically suppressed on ART 13,14 , suggesting factors such as inflammation and immune activation [14][15][16] , adverse effects of immunodeficiency 5,17,18 , ART 19,20 and altered coagulation 21,22 may be playing a role.
CH is characterized by the overproduction of blood cells derived from a single hematopoietic stem cell (HSC) harboring somatic mutation(s) that confer a selective clonal advantage. In recent years, large studies in the general population have shown that CH increases with age and is associated with the development of hematologic malignancies, cardiovascular disease, stroke and all-cause mortality [6][7][8][9][10] . While the pathophysiology of clonal dominance in CH is still largely unexplained, a number of studies have suggested an important interplay between CH and inflammation [23][24][25] . Despite the established causal relationship between chronic viral infections and chronic immunoinflammation, the consequence of a sustained virally mediated underlying inflammatory state on the prevalence of CH remains unknown. Given the increased incidence of cardiovascular disease and malignancies in people with HIV and the established link between CH and cardiovascular and malignant outcomes in the general population, we sought to evaluate the prevalence of CH among persons with HIV.
The Age-Related Clonal haematopoiesis in an HIV Evaluation cohort (ARCHIVE) study (NCT04641013) was established to examine genomic factors associated with aging and the development HIV is associated with an increased risk of age-related clonal hematopoiesis among older adults Letters Nature MediciNe of comorbidities among men and women with and without HIV over the age of 55 years living in the community. Our objectives in the ARCHIVE cohort were to: (1) compare the prevalence and type of somatic CH mutations in people with and without HIV; (2) explore socio-demographic and clinical characteristics among all participants, and characteristics of HIV infection among HIV-positive participants, to identify risk factors for the development of CH mutations; and (3) explore clinical outcomes including inflammatory biomarker profiles and medical comorbidities among people with and without somatic mutations, stratified by HIV status. A total of 461 patients were prospectively consented and 446 participants (220 HIV positive and 226 HIV negative) were included in the final analysis (Extended Data Fig. 1a). The demographic and clinical characteristics of HIV-positive and HIV-negative participants are shown in Table 1 3 . Seventeen percent had a previous AIDS-defining condition; the most common were Pneumocystis jiroveci pneumonia (6.8%), oral candidiasis (5.9%) and Kaposi's sarcoma (4.1%) (Supplementary Table 1). Most HIV-positive participants (70.2%) had a current CD4 count of ≥500 cells per mm 3 , 95.5% had a current HIV viral load of less than 40 copies per ml and 99.6% were currently taking ART (Supplementary Table 2).
To compare the prevalence of CH in HIV-positive and HIV-negative cohorts, we performed targeted amplicon sequencing of genomic DNA extracted from whole blood on all participants (Extended Data Fig. 1b and Methods). We identified a higher prevalence of CH among HIV-positive participants compared with HIV-negative participants. Overall, CH-associated mutations were identified in 100 (22.4%) of all 446 study participants: 62 (28.2%) of the 220 HIV-positive participants and 38 (16.8%) of the 226 HIV-negative participants (P = 0.004). Of note, the prevalence of CH in the HIV-negative control cohort was comparable to other studies using similar sequencing methodology 26 . The increased prevalence of CH in the HIV-positive cohort was seen across all age groups but was greatest in those over 75 years of age (Fig. 1a).
A total of 135 CH mutations were identified in the 100 participants; 83 (61.5%) in HIV-positive participants compared with 52 (38.5%) in HIV-negative participants (P = 0.002). The majority occurred in three of the main genes identified in other studies of CH in the general population 6-8 : DNMT3A (47.4%), TET2 (20.0%) and ASXL1 (13.3%). HIV-positive participants had a higher proportion of mutations in all of these genes, particularly ASXL1, compared with HIV-negative participants (Fig. 1b). Overall, missense mutations were the most frequent mutation type (39.3%), although a higher proportion of truncating mutations were observed than previously reported 6,7 , which included frameshift (36.3%), stop-gain (15.6%) and splice site mutations (6.7%; Fig. 1b and Supplementary Tables 3 and 4).
The variant allele fractions (VAFs) of the identified mutations increased with age at a greater rate among HIV-positive participants, compared with HIV-negative participants (P value for interaction = 0.01, Fig. 1c); however, there was no difference in the median (IQR) VAF by HIV status (4.7 (2.6-8.6) in HIV-positive participants versus 4.0 (2.4-8.4) in HIV-negative participants; P = 0.59). More HIV-positive participants had two or more pathogenic CH mutations detected compared to HIV-negative participants (16 (7.3%) versus 11 (4.9%), respectively, P = 0.29), but the difference was not statistically significant (Fig. 1d).
Of the 226 HIV-negative participants, 60 (26.5%) had taken pre-exposure prophylaxis (PrEP) to prevent HIV acquisition and of those, 50% were taking PrEP at the time of enrollment. All participants had used tenofovir/emtricitabine (reported for 96.7%). Among HIV-negative participants, there was no difference in exposure to PrEP among those with and without CH: 49 (26.1%) participants without CH had previously used PrEP compared to 11 (28.9%) of the participants who had at least one CH mutation (P = 0.71).
When adjusted for age, gender, sexual orientation and history of smoking, the odds ratio (OR) associated with having at least one CH mutation among HIV-positive participants compared with HIV-negative participants was 2.16 (confidence interval (CI) 1.34-3.48), P = 0.002; Fig. 2). As the relationship between age and CH is nonlinear, in addition to fitting linear age in the model, we also examined a quadratic term for age but found no statistical significance (P = 0.61). Furthermore, we performed sensitivity analyses of the association between HIV infection and CH by conducting sub-analyses for mutation VAF, specific genes mutated, gender and sexual orientation and found no significant differences in our results (Supplementary Table 5a-e).
We then explored other factors that might increase the risk of having at least one CH mutation, including ancestry, BMI, extent of smoking exposure and alcohol use, recreational and injecting drug use, annual household income and type of health insurance cover, but we did not identify any statistically significant covariates (Supplementary Table 6). Among HIV-positive participants, we did not identify any HIV-specific characteristics that were associated with an increased risk of having at least one CH mutation in the univariate analysis (Supplementary Table 7). We also explored whether duration of HIV infection correlated with the VAF of the CH mutation (averaged within individuals) or the presence of two or more CH mutations but did not find a significant association for either relationship (Extended Data Fig. 2).
As more HIV-positive participants were enrolled from hospital clinics than HIV-negative participants (64 (29.1%) versus 11 (4.9%), respectively; P < 0.001), we explored whether this led to a selection bias for more unwell and possibly older, HIV-positive participants, which may have influenced our results toward finding a difference in CH by HIV status. We found no difference in the proportion of participants with and without CH that were enrolled at tertiary-care sites (55 (15.9%) versus 20 (20.0%), respectively; P = 0.33) and no difference in the median age of participants enrolled at general practitioner versus tertiary-care sites (median IQR, 63 (59-69) versus 66 (60-69) years; P = 0.15). In addition, we found no significant difference in the OR associated with having at least one CH mutation if HIV positive, adjusting for age, gender, sexual orientation, history of smoking and enrollment site type (general practitioner versus tertiary care: OR (CI) = 2.26 (1.35-3.76), P = 0.002).

Nature MediciNe
We then conducted a multivariable regression analysis of the association between the presence of CH, HIV status and our prespecified clinical outcomes including full blood examination (FBE) characteristics, inflammatory markers and medical comorbidities. Importantly, here we adjusted for age, gender and history of smoking, which are known confounders of these clinical outcomes ( Table  2; for full table see Supplementary Table 8). The presence of at least one CH mutation and HIV infection was independently associated with a statistically significant increase (P < 0.05) in several blood parameters associated with chronic inflammation including neutrophil count, interleukin (IL)-6 and C-reactive protein (CRP) levels ( Table 2 and Supplementary Table 8). However, there were little differences found in other blood count parameters including hemoglobin, platelets or red cell distribution width.
Recent data have suggested that the common hypomorphic variant of the IL-6 receptor (IL6R p.Asp358Ala) potentially negates the increased risk of cardiovascular disease observed in patients with CH 27 . Interestingly, within our cohort, we found no significant difference in the prevalence of the IL-6 receptor polymorphism by HIV status (62.7% among HIV-positive versus 70.8% among HIV-negative participants, P = 0.07) or according to the presence of CH (62.0% among participants with CH versus 68.2% among participants without CH, P = 0.25). We also found that the increase in serum IL-6, which was associated with both the presence of CH and HIV infection (Table 2), could not be explained by expression of the IL6R p.Asp358Ala variant (Extended Data Fig. 3).
In this cohort of HIV-positive and HIV-negative Australian adults over the age of 55, HIV infection was associated with more than double the odds of having CH, and this association was independent of previously reported risk factors for CH such as smoking, gender and age. Both HIV infection and the presence of CH were independently associated with increases in several parameters associated with inflammation even when adjusting for known risk factors.
We found that the increased prevalence of CH in HIV-positive individuals was most marked in older age groups. Although these individuals had higher numbers of DNMT3A and TET2 mutations, the greatest difference across HIV status occurred in pathogenic mutations in ASXL1 (Fig. 1b). Notably, ASXL1 and DNMT3A mutations have been identified to be more frequent in patients with myelodysplastic syndrome who have HIV and are associated with a poorer

Letters
Nature MediciNe prognosis 28 . The emergence of well-tolerated disease-modifying therapies for myelodysplastic syndrome 29-31 may argue for increased surveillance in older HIV-positive patients with CH. Raised inflammatory markers, which are well documented in HIV infection, have recently been implicated in the pathogenesis of CH [23][24][25]32 . Increasing experimental evidence has also highlighted that repeated or sustained exposure to inflammatory stimuli are toxic to normal HSCs ultimately leading to their depletion and bone marrow failure [33][34][35] . Remarkably, in the context of sustained chronic inflammation, HSCs with mutations in CH genes such as TET2 have a major competitive advantage 36,37 . Unlike normal HSCs, TET2 mutant HSCs show reduced apoptosis and enhanced activity in response to inflammatory cytokines, particularly IL-6 (refs. 36,37 ). Consistent with our findings, IL-6 is known to be increased in HIV-positive individuals even in those who are virologically suppressed on highly effective ART regimens 22,38 . Taken together, these findings raise the prospect that the underlying chronic inflammatory state seen in HIV infection provides a supportive milieu that stimulates clonal dominance of HSC with CH mutations and may account for the increased rate of CH seen in HIV-positive individuals. It is also possible that the immunodeficiency associated with acute HIV infection or the chronic inflammation and immune activation associated with long-term HIV infection may lead to impaired immune surveillance and reduced clearance of clonal populations of HSC in people with HIV, further contributing to the higher prevalence of CH.
While these prospective data are from a carefully conducted clinical study designed specifically to investigate the prevalence of CH in people with and without HIV, our cohort only enrolled people over 55 years of age who were predominantly male and whose sexual orientation was defined as men who have sex with men (MSM), in a health system that provides universal access to standard-of-care antiretrovirals. Further clinical studies in larger and more diverse populations, including younger individuals, women with HIV and those who have not had sufficient access to ART, are needed to evaluate HIV-specific risk factors for CH. These insights may inform future interventions to guide surveillance for and management of CH in people with HIV.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41591-021-01357-y.  represents an OR of 1 (no effect). Logistic regression was used to calculate the unadjusted and adjusted ORs and 95% CIs for the risk of CH among HIV-positive participants, compared with HIV-negative participants. We adjusted a priori for age, gender, whether the participant had ever smoked and for MSM status; therefore, no adjustments were made for multiple comparisons. Two-sided Wald test statistics are reported for testing each covariate in adjusted and unadjusted analyses.

Table 2 | Multivariable linear regression analysis of associations between the presence of at least one CH mutation and HIV status and clinical outcomes (blood cell counts and inflammatory markers)
Covariate ln IL-6 (pg ml −1 ) ln CRP (mg l −1 ) ln neutrophil count (× 10 9 l -1 ) ln hemoglobin (g l −1 ) a Estimate (95% CI) P value Estimate (95% CI) P value Estimate (95% CI) P value Estimate (95% CI) P value Analyses were adjusted a priori for age, gender and history of ever smoking. Continuous outcomes underwent natural log (ln) transformation. Two-sided Wald test statistics are reported for testing each covariate in these analyses. a Women were excluded from the hemoglobin model due to varying reference ranges for men compared with women.

Nature MediciNe
Methods Clinical study procedures. The detailed ARCHIVE study protocol is available in the Supplementary Note. In brief, the study enrolled participants over the age of 55 years from five primary-care practices with high HIV caseloads and one tertiary hospital (a designated HIV care center) in Sydney, Australia, and one primary-care practice with high HIV caseloads and two tertiary-care hospitals (both designated HIV care centers) in Melbourne, Australia. HIV-negative participants were required to have a negative HIV test within 12 months of enrollment. After providing written informed consent, participants provided a whole-blood sample for FBE, inflammatory biomarker testing (Supplementary Table 1) and CH testing. Relevant data were obtained from participants' medical records including: age, gender, sexual orientation, most recent vital signs, results of routine standard-of-care blood tests (Supplementary Table 1), medical diagnoses and use of PrEP among HIV-negative participants. For participants with HIV, date of HIV diagnosis, date of initiation of ART, lowest CD4 count on record, most recent CD4 count and HIV viral load, prior exposure to stavudine and zidovudine, which are known to cause myelosuppression 39 , and current ART regimen were also recorded. All participants were asked to complete a questionnaire on their ancestry, physical health/frailty, use of tobacco, alcohol and recreational drugs, and household income. BMI was calculated using height and weight reported from the enrolling clinical site, and was categorized using ranges published by the Centers for Disease Control and Prevention 40 . Data on gender and sexual orientation were obtained by asking the enrolling clinical site to report the participant's gender and whether the participant was sexually active with men, women or both. Sexual orientation was categorized as MSM, defined as male participants who were reported by the clinical site as having sex with either men only or both men and women.
Whole-blood sample processing. Whole blood from enrolled participants was collected in EDTA tubes. DNA from whole blood was extracted using the QIAsymphony DNA Midi Kit (Qiagen, 931255) according to the manufacturer's protocols.
Targeted deep sequencing. Participant screening for CH was performed on whole-blood DNA at the Peter MacCallum Cancer Centre using a bespoke targeted deep sequencing (TS) amplicon panel designed across 55 genes recurrently mutated in hematological malignancies (Supplementary Table 9) 41 . Investigators completing the testing for CH were blinded to participants' HIV status.
Target specific amplification was performed using the Fluidigm Access Array system. Following amplification, products were collected, tagged with sample-specific barcodes, pooled together and purified using AMPure XP beads. All samples were analyzed in duplicate to control for PCR artifacts. The purified libraries were then sequenced using a 150-bp read length on the Illumina MiSeq or NextSeq platform. Sequenced reads were mapped to the human reference genome (version hg19) using BWA-MEM (version 0.7.12) with default parameters. All samples were tested in at least two technical replicates. Variant calling was performed using an in-house pipeline 42 . Mutations with at least 300× coverage with a minimum of 30 reads supporting the variant and a mutant allele fraction greater than 1% were retained for further analysis. The median targeted sequencing coverage over variant reads was 3,210-fold. Variants that were recurrently observed in more than 1% of the samples (representing likely sequencing and/or PCR artifacts) and those with a high global allele frequency (>1%) in population databases (gnomAD) were flagged and removed from this curated list. Variants in the curated list were then annotated based on their prognostic or functional relevance 43 . Our analysis did not include paired germline samples to definitively exclude germline variants. To account for this, variants with greater than 30% allele frequency were not included in the final curated list unless they had been described in the COSMIC database.
Fragment-length analysis for ASXL1 c.1934dupG;p.Gly646Trpfs*12. The common ASXL1 variant c.1934dupG;p.Gly646Trpfs*12 is highly susceptible to false-positive results due to PCR artifacts within this region. To assess for the presence of this variant, we performed fragment-length analysis in addition to TS. PCR amplification of the ASXL1 region of interest was performed using ASXL1-specific primers 44 . PCR products were then denatured in the presence of Hi-Di formamide and sized using capillary electrophoresis using the Applied Biosystems 3730 Genetic Analyser. Data were analyzed using the GeneMarker software (v2.7.0) 44 . A positive AXSL1 fragment-length result was recorded as the presence of a +1-bp insertion. Positive ASXL1 fragment-length results were then integrated with TS data to confirm the presence of the ASXL1 c.1934dupG;p. Gly646Trpfs*12 variant. ASXL1 c.1934dupG;p.Gly646Trpfs*12 was recorded as detected if positive by fragment-length analysis and detected in TS at greater than 5% VAF.
Droplet digital PCR. Digital PCR was used to assess for the presence of the IL6R p.Asp358Ala polymorphism. In addition, it was used to validate hotspot mutations in DNMT3A Arg882His, Arg882Cys and JAK2 Val617Phe detected by targeted sequencing. This was performed using the Bio-Rad QX200 Droplet digital PCR system following the manufacturer's protocols. Allele-specific PCR assays to specifically detect and quantify the fractional abundance of point mutations and corresponding WT alleles were commercially obtained (PrimePCR PCR Primers and Assays, Bio-Rad). Each sample was analyzed by at least two technical replicates. Data analysis was carried out using the QuantaSoft Software (Bio-Rad).
Statistical considerations. Sample size calculation. We assumed that 10% of our HIV-negative participants would have somatic mutations associated with CH, which is consistent with published studies in the literature that report a range of 6-10% in older adults [6][7][8] . Testing 220 HIV-positive and 220 HIV-negative participants gave 80% power to detect an increase to 20% prevalence of mutations among HIV-positive participants compared to the background mutation rate of 10% among HIV-negative participants.
Statistical analysis. Conventional descriptive statistics were used to compare characteristics across HIV-positive and HIV-negative participants; P values for comparisons were generated using the chi-square test. The primary outcome of interest was the presence of any somatic mutation associated with CH. We used logistic regression to calculate the unadjusted and adjusted ORs and 95% CIs for the risk of CH among HIV-positive participants, compared with HIV-negative participants. We adjusted a priori for age, gender and whether the participant had ever smoked, which are factors that have been reported to be associated with CH (age, gender and smoking) in the general population [6][7][8] and with HIV infection (smoking and gender). We also adjusted for MSM status, which was found to differ slightly across HIV status in our study population.
In exploratory secondary analyses, we looked for associations between non-HIV-specific and HIV-specific risk factors and CH (by participant) using logistic regression. We also explored associations between our explanatory variables of interest (HIV infection and presence of CH) and clinical outcomes including elevations of: (1) inflammatory markers, (2) blood cell counts and characteristics and (3) select clinical comorbidities, including cardiovascular disease, malignancies and specifically, hematologic malignancies. To evaluate the relationship between HIV, CH and these clinical outcomes, separate multivariable linear regression models (for continuous measures) and multivariable logistic regression models (for dichotomous measures) were built to evaluate the relationship between HIV and CH and each of these clinical outcomes (inflammatory markers, FBE characteristics and comorbidities), adjusting a priori for age, gender and history of ever smoking. All continuous outcomes underwent natural log transformation. The primary comparison of interest in our analyses was HIV status, in which only minimal adjustments for factors a priori were made. Therefore, for this primary comparison and all other exploratory analyses we did not adjust significance levels or CIs for multiple comparisons. The data analysis for this paper was generated using SAS software (v9.4) of the SAS System for Windows (SAS Institute).
Human participants. The research protocol, consent form and associated documents were approved by the St Vincent's Hospital Human Research Ethics Committee in New South Wales, Australia. All individuals gave written informed consent to participate in the study. The study was registered with ClinicalTrials.gov (NCT04641013).