Abstract
Background
There is an incompletely understood increased risk for cardiovascular disease (CVD) among people with HIV (PWH). We investigated if a collection of biomarkers were associated with CVD among PWH. Mendelian randomization (MR) was used to identify potentially causal associations.
Methods
Data from follow-up in 4 large trials among PWH were used to identify 131 incident CVD cases and they were matched to 259 participants without incident CVD (controls). Tests of associations between 460 baseline protein levels and case status were conducted.
Results
Univariate analysis found CLEC6A, HGF, IL-6, IL-10RB, and IGFBP7 as being associated with case status and a multivariate model identified 3 of these: CLEC6A (odds ratio [OR] = 1.48, P = .037), HGF (OR = 1.83, P = .012), and IL-6 (OR = 1.45, P = .016). MR methods identified 5 significantly associated proteins: AXL, CHI3L1, GAS6, IL-6RA, and SCGB3A2.
Conclusions
These results implicate inflammatory and fibrotic processes as contributing to CVD. While some of these biomarkers are well established in the general population and in PWH (IL-6 and its receptor), some are novel to PWH (HGF, AXL, and GAS6) and some are novel overall (CLEC6A). Further investigation into the uniqueness of these biomarkers in PWH and the role of these biomarkers as targets among PWH is warranted.
Keywords: mendelian randomization, fibrosis, inflammation
Using a collection of protein panels, associations between baseline protein levels and incident CVD were examined. Associations were detected between several proteins and incident CVD and a causal role for some of these associations was found using mendelian randomization.
People with human immunodeficiency virus (PWH) are at increased risk of cardiovascular disease (CVD) when compared to the general population [1–3]. The reasons for this increase in risk remain unclear. In addition to a higher prevalence of traditional risk factors [4], patients with HIV infection have evidence of persistent abnormalities in inflammation and coagulation that might be related to duration of infection [5–8], incomplete immune recovery [9, 10], ongoing viral replication [11], transcription of defective proviruses [12], and/or long-term antiretroviral therapy (ART) toxicity [13]. Recent meta-analyses have estimated this increased risk to be a factor of about 2 after adjustment for potential confounders [14, 15]. Recent reviews have summarized these and other mechanisms [16, 17]. Identifying mechanisms of this increased risk has the potential to suggest molecular treatment targets.
Recent advances in proteomics have enabled more precise quantification of protein levels in human specimens. While many of these advances have utilized mass spectrometry-based approaches, assays that rely on other molecular techniques have also seen widespread adoption. In particular, the proximity extension assay, made commercially available by Olink, has been used for protein quantification in multiple large cohort studies focused on identification of risk factors for CVD in the general population [18–20].
While investigations of associations between potential protein biomarkers and the development of CVD have the potential to elucidate mechanisms, these associations may be attributable to a variety of unmeasured and unknown confounding factors. One established method for overcoming this shortcoming is the use of mendelian randomization (MR) [21, 22]. Given several key assumptions, MR is a statistical technique that can be used to identify causal associations between risk factors and outcomes when one has genotypic data. In MR analysis, one tries to overcome the impact of confounders by identifying a genetic variant that is related to the risk factor but is unlikely to be related to confounders, and then one uses a technique called instrumental variable analysis to test for an association between the risk factor and the outcome using the identified genetic variant. MR analysis has been used to investigate the causal role of a wide variety of risk factors for CVD, including lower levels of high-density lipoprotein cholesterol [23] and higher levels of the interleukin 6 (IL-6) receptor [24].
Here we investigate the predictive utility and apply MR to these novel protein assays in a group of geographically and racially diverse PWH who were recruited into clinical trials conducted by the International Network for Strategic Initiatives in Global HIV Trials (INSIGHT). A total of 15 665 individuals participated in these trials, many contributing specimens, and were followed for up to 10 years with clinical evaluations. Detailed clinical outcomes were documented and assessed by endpoint review committees. These proteomic assays, specimens, statistical techniques, and clinical outcomes were used to investigate the roles of a range of candidate proteins on the development of CVD in PWH.
METHODS
Participants
Data from participants in 4 clinical trials (FIRST [25], ESPRIT [26], SMART [27], and START [28]) who consented to studies of genomics (8428 of 15 665) were combined and a case definition comprising a composite of death, AIDS, or serious non-AIDS conditions was used to identify 500 cases. (Some of these individuals could have experienced a previous event that met the criteria for case status used here.) Cases who had experienced at least 1 of these outcomes were individually matched on a 1 to 2 basis with controls who had not experienced any of these outcomes over follow-up. Matching was performed within treatment arms of each study and used randomization date and age. This resulted in a nested case-control dataset involving 1493 individuals.
Here, for analysis of CVD we focus on the 131 cases who experienced a composite CVD outcome, defined as stroke, myocardial infarction, or coronary revascularization, and 259 matched controls. For analyses not explicitly related to CVD we used the entire cohort of 1493 individuals; the latter analyses were focused on identification of genetic-based instruments and by using the larger cohort these analyses had greater power.
Assays
Plasma, mostly obtained during a fasting state, from the baseline study visit for these study participants was used to generate data for 460 proteins from 5 Olink protein panels (cardiometabolic, CVD2, CVD3, immune response, and inflammation). Some proteins were represented on more than 1 panel. Proteins were excluded from further consideration if more than 10% of the data were below the limit of detection. Observations that were below the limit of detection for retained proteins were imputed at one-half the limit of detection on the linear scale.
Genotypes for all 1493 individuals with protein level data were obtained using an Axion Affymetrix array with 770 558 probesets.
Statistical Methods
Tests for association between protein levels and the CVD composite outcome controlled for sex, age, self-declared black race, diabetes, hypertension, and treatment assignment in each trial. Generalized estimating equations (GEE) were used to account for matching among cases and controls: both linear and logistic regression results are reported due to the widespread use of linear regression in MR analysis with binary outcomes. These analyses were restricted to CVD cases and their matched controls.
Construction of predictive models for the CVD composite outcome using collections of potential biomarkers started with all proteins meeting marginal statistical significance based on Bonferroni criteria given the number of proteins and all covariates listed for the CVD composite outcome analysis. The protein with the smallest test statistic in absolute value was successively dropped according to a backwards model selection strategy. Once all remaining proteins were significant, each protein meeting the Bonferroni threshold that was dropped was re-entered into the model and was retained if the effect was statistically significant. These analyses were restricted to CVD cases and their matched controls and also used GEE versions of linear regression to be consistent with MR analyses but were supplemented with logistic regression.
One difficulty with using MR analysis in the current context is control of the type I error due to the large number of proteins and genetic variants. To best make use of the large number of proteins and genetic variants while controlling the number of false positives, we devised a sequential testing method to control the family-wise error rate of our collection of MR tests. The basic idea is that there is little power for finding a causal association unless there is an association between a genetic variant and a protein on the one hand and an association between the CVD outcome and a protein on the other. This suggests that we screen these pairs of associations and only conduct the MR test when both members of a pair of these tests reject the null hypothesis. If we have a collection of p proteins and we let αprotein, CVD represent the significance level we require for rejecting the null hypothesis of no association between protein levels and the CVD composite outcome, αprotein, SNP represent the significance level we require for rejecting the null hypothesis of no association between protein levels and single-nucleotide polymorphism (SNPs) and αMR represent the significance level we require for rejecting the null hypothesis of no association in our MR tests, then it transpires that the probability we reject 1 of more MR tests under the null hypothesis of no association between any of the factors involved is approximately p αMRαprotein, SNPαprotein, CVD. This is true under the set of assumptions employed in justifying the Bonferroni correction: details are provided in the Supplementary Materials.
Genetic variants were selected by finding SNPs near the coding region of each protein that were significantly associated with the protein's level, then haplotypes were estimated for each participant using this set of SNPs. The most common haplotype was determined, and individuals were coded as having 0–2 copies of this haplotype. This haplotype was used as a genetic variant in MR tests. Additional methodological details can be found in the Supplementary Materials.
RESULTS
Table 1 provides a summary of baseline participant characteristics for CVD cases and controls. A similar summary for all cases and controls is presented in Supplementary Table 1. Participants in 2 of the trials (FIRST and START) were ART naive at baseline and there is substantial variation among participants across the trials.
Table 1.
Summary of Participant Characteristics Among Cardiovascular Disease Cases and Controls
| Characteristic | ESPRIT | SMART | FIRST | START | Total | |||||
|---|---|---|---|---|---|---|---|---|---|---|
| Cases | Controls | Cases | Controls | Cases | Controls | Cases | Controls | Cases | Controls | |
| No. of participants | 77 | 151 | 16 | 32 | 8 | 16 | 30 | 60 | 131 | 259 |
| Age, y, median (interquartile range) | 48 (40–54) | 46 (40–52) | 52 (45–57) | 50 (46–58) | 44 (35–53) | 41 (35–50) | 50 (44–56) | 48 (43–55) | 49 (41–55) | 47 (41–53) |
| Female | 4 (5.2) | 13 (8.6) | 1 (6.3) | 8 (25.0) | 2 (25.0) | 3 (18.8) | 4 (13.3) | 14 (23.3) | 11 (8.4) | 38 (14.7) |
| Black race | 6 (7.8) | 14 (9.3) | 3 (18.8) | 7 (21.9) | 4 (50.0) | 11 (68.8) | 9 (30.0) | 12 (20.0) | 22 (16.8) | 44 (17.0) |
| Geographic region | ||||||||||
| ȃAfrica | 0 (0.0) | 0 (0.0) | 0 (0.0) | 0 (0.0) | 0 (0.0) | 0 (0.0) | 2 (6.7) | 4 (6.7) | 2 (1.5) | 4 (1.5) |
| ȃAsia | 2 (2.6) | 10 (6.6) | 0 (0.0) | 0 (0.0) | 0 (0.0) | 0 (0.0) | 0 (0.0) | 0 (0.0) | 2 (1.5) | 10 (3.9) |
| ȃEurope + Israel | 39 (50.6) | 73 (48.3) | 2 (12.5) | 3 (9.4) | 0 (0.0) | 0 (0.0) | 14 (46.7) | 33 (55.0) | 55 (42.0) | 109 (42.1) |
| ȃLatin America | 6 (7.8) | 13 (8.6) | 3 (18.8) | 2 (6.3) | 0 (0.0) | 0 (0.0) | 0 (0.0) | 11 (18.3) | 9 (6.9) | 26 (10.0) |
| ȃOceania | 6 (7.8) | 14 (9.3) | 3 (18.8) | 5 (15.6) | 0 (0.0) | 0 (0.0) | 4 (13.3) | 1 (1.7) | 13 (9.9) | 20 (7.7) |
| HIV RNA, median (interquartile range), copies/mL | 50 (50–400) | 50 (50–400) | 50 (50–50) | 50 (50–400) | 86833 (42056–310714) | 68043 (38073–152010) | 13791 (3900–47000) | 9895 (2698–30450) | 400 (50–7348) | 262 (50–6424) |
| Hepatitis B | 5 (7.1) | 7 (5.0) | 0 (0.0) | 0 (0.0) | 0 (0.0) | 0 (0.0) | 1 (3.7) | 1 (1.7) | 6 (5.0) | 8 (3.3) |
| Regimen at entry a | ||||||||||
| ȃNRTI, any | 76 (98.7) | 148 (98.0) | 16 (100.0) | 27 (100.0) | … | … | … | … | 92 (98.9) | 175 (98.3) |
| ȃȃAbacavir | 15 (19.5) | 36 (23.8) | 4 (25.0) | 6 (22.2) | … | … | … | … | 19 (20.4) | 42 (23.6) |
| ȃȃStavudine | 32 (41.6) | 62 (41.1) | 6 (37.5) | 5 (18.5) | … | … | … | … | 38 (40.9) | 67 (37.6) |
| ȃȃDidanosine | 19 (24.7) | 36 (23.8) | 1 (6.3) | 3 (11.1) | … | … | … | … | 20 (21.5) | 39 (21.9) |
| ȃȃLamivudine | 58 (75.3) | 108 (71.5) | 13 (81.3) | 21 (77.8) | … | … | … | … | 71 (76.3) | 129 (72.5) |
| ȃȃTenofovir | 3 (3.9) | 7 (4.6) | 3 (18.8) | 8 (29.6) | … | … | … | … | 6 (6.5) | 15 (8.4) |
| ȃȃZidovudine | 34 (44.2) | 59 (39.1) | 4 (25.0) | 12 (44.4) | … | … | … | … | 38 (40.9) | 71 (39.9) |
| ȃȃOther NRTI | 0 (0.0) | 0 (0.0) | 0 (0.0) | 0 (0.0) | … | … | … | … | 0 (0.0) | 0 (0.0) |
| ȃNNRTI, any | 36 (46.8) | 72 (47.7) | 7 (43.8) | 12 (44.4) | … | … | … | … | 43 (46.2) | 84 (47.2) |
| ȃȃEfavirenz | 17 (22.1) | 48 (31.8) | 2 (12.5) | 5 (18.5) | … | … | … | … | 19 (20.4) | 53 (29.8) |
| ȃȃNevirapine | 19 (24.7) | 28 (18.5) | 5 (31.3) | 6 (22.2) | … | … | … | 24 (25.8) | 34 (19.1) | |
| ȃȃOther NNRTI | 0 (0.0) | 1 (0.7) | 0 (0.0) | 1 (3.7) | … | … | … | … | 0 (0.0) | 2 (1.1) |
| ȃPI, any | 46 (59.7) | 76 (50.3) | 8 (50.0) | 12 (44.4) | … | … | … | … | 54 (58.1) | 88 (49.4) |
| ȃȃIndinavir | 12 (15.6) | 24 (15.9) | 0 (0.0) | 1 (3.7) | … | … | … | … | 12 (12.9) | 25 (14.0) |
| ȃȃLopinavir | 12 (15.6) | 16 (10.6) | 3 (18.8) | 6 (22.2) | … | … | … | … | 15 (16.1) | 22 (12.4) |
| ȃȃNelfinavir | 11 (14.3) | 21 (13.9) | 3 (18.8) | 1 (3.7) | … | … | … | … | 14 (15.1) | 22 (12.4) |
| ȃȃRitonavir | 22 (28.6) | 36 (23.8) | 3 (18.8) | 10 (37.0) | … | … | … | … | 25 (26.9) | 46 (25.8) |
| ȃȃSaquinavir | 12 (15.6) | 17 (11.3) | 0 (0.0) | 2 (7.4) | … | … | … | … | 12 (12.9) | 19 (10.7) |
| ȃȃOther PI | 1 (1.3) | 1 (0.7) | 2 (12.5) | 0 (0.0) | … | … | … | … | 3 (3.2) | 1 (0.6) |
Data are presented as No. (%) except where indicated.
Abbreviations: NNRTI, nonnucleoside reverse transcriptase inhibitor; NRTI, nucleoside reverse transcriptase inhibitor; PI, protease inhibitor.
Of those on antiretroviral treatment at entry.
Biomarkers Associated With Clinical Outcomes
Summaries of the protein data (along with the Uniprot ID for each protein symbol used here) can be found in Supplementary Tables 2–11. Overall, 86 of the 385 distinct proteins meeting quality control criteria were significantly associated with the CVD composite outcome using a significance level of .05. The number of significant proteins found using the individual panels (ignoring overlap) was as follows: cardiometabolic (8), cardiovascular II (14), cardiovascular III (32), immune response (15), and inflammation (24). Table 2 and Table 3 present results for all significant proteins using linear and logistic regression models while Supplementary Tables 12–16 present similar results for the proteins that failed to reach significance for each panel. Using a Bonferroni adjustment across all panels simultaneously, 5 proteins differed significantly between cases and controls while controlling for relevant covariates. These were C-type lectin domain containing 6A (CLEC6A), hepatocyte growth factor (HGF), IL-6, IL-10 receptor beta (IL-10RB), and insulin like growth factor binding protein 7 (IGFBP7).
Table 2.
Summaries of Significant Associations Between Protein Levels From the Cardiovascular Disease-Related Panels and Cardiovascular Disease Outcomes for Linear and Logistic Regression Models
| Protein | Linear Model | Logistic Model | ||||||
|---|---|---|---|---|---|---|---|---|
| Slope | SE | P Value | (95% CI) | OR | SE | P Value | (95% CI) | |
| Cardiometabolic panel | ||||||||
| ȃCCL18 | 0.08 | 0.023 | .001 | (.034–0.126) | 1.445 | 0.163 | .001 | (1.158–1.803) |
| ȃCD59 | 0.064 | 0.028 | .025 | (.008–.119) | 1.326 | 0.176 | .034 | (1.022–1.721) |
| ȃCDH1 | 0.126 | 0.052 | .017 | (.023–.229) | 1.773 | 0.436 | .02 | (1.095–2.873) |
| ȃCST3 | 0.108 | 0.049 | .027 | (.012–.203) | 1.661 | 0.395 | .033 | (1.042–2.648) |
| ȃGAS6 | 0.12 | 0.055 | .029 | (.012–.228) | 1.746 | 0.459 | .034 | (1.044–2.922) |
| ȃIGLC2 | 0.125 | 0.04 | .002 | (.046–.204) | 1.818 | 0.373 | .004 | (1.216–2.719) |
| ȃPLA2G7 | 0.142 | 0.056 | .011 | (.032–.251) | 1.943 | 0.514 | .012 | (1.157–3.264) |
| ȃPRSS2 | 0.097 | 0.041 | .017 | (.017–.178) | 1.572 | 0.314 | .024 | (1.063–2.326) |
| Cardiovascular panel II | ||||||||
| ȃACE2 | 0.05 | 0.023 | .03 | (.005–.095) | 1.261 | 0.134 | .029 | (1.024–1.554) |
| ȃADAMTS13 | −0.116 | 0.059 | .049 | (−.231 to 0) | 0.304 | 0.185 | .05 | (.092–1.002) |
| ȃADM | 0.127 | 0.044 | .004 | (.04–.214) | 1.867 | 0.445 | .009 | (1.17–2.979) |
| ȃCTSL1 | 0.142 | 0.045 | .002 | (.053–.23) | 1.935 | 0.412 | .002 | (1.275–2.937) |
| ȃFGF21 | 0.031 | 0.013 | .023 | (.004–.057) | 1.154 | 0.073 | .023 | (1.02–1.307) |
| ȃGAL9 | 0.193 | 0.062 | .002 | (.071–.316) | 2.569 | 0.806 | .003 | (1.39–4.751) |
| ȃIL-1RL2 | −0.07 | 0.031 | .022 | (−.13 to −.01) | 0.676 | 0.127 | .036 | (.468–.975) |
| ȃIL-6 | 0.099 | 0.027 | <.001 | (.047–.152) | 1.593 | 0.214 | .001 | (1.223–2.073) |
| ȃKIM1 | 0.075 | 0.023 | .001 | (.029–.12) | 1.417 | 0.162 | .002 | (1.133–1.772) |
| ȃPGF | 0.14 | 0.051 | .006 | (.041–.24) | 1.937 | 0.485 | .008 | (1.185–3.164) |
| ȃPRSS8 | 0.182 | 0.059 | .002 | (.067–.297) | 2.459 | 0.745 | .003 | (1.358–4.451) |
| ȃSPON2 | 0.315 | 0.094 | .001 | (.13–.499) | 4.914 | 2.506 | .002 | (1.808–13.352) |
| ȃTRAILR2 | 0.113 | 0.039 | .003 | (.037–.188) | 1.797 | 0.52 | .043 | (1.019–3.169) |
| ȃVEGFD | 0.108 | 0.054 | .048 | (.001–.214) | 1.629 | 0.421 | .059 | (.982–2.704) |
| Cardiovascular panel III | ||||||||
| ȃAXL | 0.119 | 0.057 | .035 | (.008–.23) | 1.756 | 0.471 | .036 | (1.038–2.971) |
| ȃCCL15 | 0.085 | 0.043 | .046 | (.001–.169) | 1.5 | 0.304 | .046 | (1.008–2.231) |
| ȃCHI3L1 | 0.046 | 0.022 | .037 | (.003–.09) | 1.235 | 0.126 | .038 | (1.011–1.508) |
| ȃCHIT1 | 0.034 | 0.013 | .009 | (.008–.06) | 1.196 | 0.099 | .031 | (1.017–1.407) |
| ȃCPA1 | 0.076 | 0.032 | .02 | (.012–.139) | 1.419 | 0.221 | .025 | (1.045–1.926) |
| ȃCPB1 | 0.071 | 0.032 | .026 | (.009–.133) | 1.385 | 0.207 | .029 | (1.034–1.856) |
| ȃCSTB | 0.073 | 0.034 | .032 | (.006–.14) | 1.412 | 0.228 | .033 | (1.029–1.937) |
| ȃCTSD | 0.088 | 0.039 | .026 | (.011–.165) | 1.518 | 0.281 | .024 | (1.056–2.182) |
| ȃCTSZ | 0.139 | 0.057 | .015 | (.027–.252) | 1.92 | 0.53 | .018 | (1.117–3.298) |
| ȃCXCL16 | 0.18 | 0.064 | .005 | (.055–.306) | 2.347 | 0.736 | .006 | (1.27–4.338) |
| ȃEPHB4 | 0.196 | 0.069 | .004 | (.061–.331) | 2.534 | 0.893 | .008 | (1.27–5.057) |
| ȃFAS | 0.138 | 0.069 | .046 | (.003–.273) | 1.906 | 0.627 | .05 | (1–3.633) |
| ȃGAL4 | 0.108 | 0.037 | .003 | (.035–.18) | 1.649 | 0.295 | .005 | (1.162–2.342) |
| ȃGDF15 | 0.078 | 0.024 | .001 | (.031–.126) | 1.447 | 0.172 | .002 | (1.147–1.826) |
| ȃGRN | 0.174 | 0.069 | .012 | (.038–.309) | 2.29 | 0.79 | .016 | (1.165–4.501) |
| ȃIGFBP7 | 0.19 | 0.046 | <.001 | (.099–.281) | 2.467 | 0.621 | <.001 | (1.507–4.04) |
| ȃIL-18BP | 0.184 | 0.055 | .001 | (.075–.292) | 2.396 | 0.675 | .002 | (1.379–4.161) |
| ȃIL-2RA | 0.155 | 0.043 | <.001 | (.07–.24) | 2.104 | 0.469 | .001 | (1.36–3.256) |
| ȃIL-6RA | 0.118 | 0.06 | .049 | (0–.236) | 1.757 | 0.508 | .051 | (.997–3.096) |
| ȃLTBR | 0.201 | 0.056 | <.001 | (.09–.311) | 2.629 | 0.819 | .002 | (1.428–4.84) |
| ȃMCP1 | 0.141 | 0.055 | .01 | (.033–.25) | 1.96 | 0.522 | .011 | (1.164–3.302) |
| ȃOPG | 0.175 | 0.064 | .006 | (.05–.3) | 2.29 | 0.724 | .009 | (1.233–4.255) |
| ȃPCSK9 | 0.143 | 0.055 | .009 | (.035–.25) | 1.962 | 0.525 | .012 | (1.161–3.315) |
| ȃPLC | 0.137 | 0.057 | .016 | (.025–.248) | 1.917 | 0.525 | .017 | (1.121–3.28) |
| ȃRARRES2 | 0.15 | 0.058 | .01 | (.036–.264) | 2.087 | 0.625 | .014 | (1.16–3.755) |
| ȃSCGB3A2 | 0.066 | 0.023 | .003 | (.022–.111) | 1.358 | 0.145 | .004 | (1.102–1.673) |
| ȃST2 | 0.087 | 0.037 | .019 | (.014–.16) | 1.5 | 0.274 | .027 | (1.048–2.146) |
| ȃTFF3 | 0.13 | 0.053 | .015 | (.026–.234) | 1.862 | 0.509 | .023 | (1.09–3.181) |
| ȃTNFR1 | 0.179 | 0.049 | <.001 | (.083–.276) | 2.385 | 0.635 | .001 | (1.416–4.018) |
| ȃTNFR2 | 0.149 | 0.046 | .001 | (.059–.239) | 2.026 | 0.475 | .003 | (1.279–3.207) |
| ȃTNFSF13B | 0.149 | 0.05 | .003 | (.051–.246) | 2.023 | 0.5 | .004 | (1.245–3.285) |
| ȃUPAR | 0.133 | 0.051 | .01 | (.032–.233) | 1.862 | 0.472 | .014 | (1.133–3.062) |
Table 3.
Summaries of Significant Associations Between Protein Levels From the Cardiovascular Disease-Related Panels and Cardiovascular Disease Outcomes for Linear and Logistic Regression Models
| Protein | Linear Model | Logistic Model | ||||||
|---|---|---|---|---|---|---|---|---|
| Slope | SE | P Value | (95% CI) | OR | SE | P Value | (95% CI) | |
| Immune response panel | ||||||||
| ȃAREG | 0.139 | 0.044 | .001 | (.054–.225) | 1.922 | 0.404 | .002 | (1.272–2.903) |
| ȃCCL11 | 0.139 | 0.047 | .003 | (.047–.231) | 1.927 | 0.443 | .004 | (1.227–3.025) |
| ȃCD83 | 0.117 | 0.053 | .027 | (.013–.222) | 1.749 | 0.45 | .03 | (1.056–2.896) |
| ȃCKAP4 | 0.165 | 0.044 | <.001 | (.079–.252) | 2.168 | 0.509 | .001 | (1.369–3.435) |
| ȃCLEC4C | 0.106 | 0.045 | .02 | (.017–.195) | 1.67 | 0.372 | .021 | (1.079–2.584) |
| ȃCLEC4D | 0.078 | 0.04 | .047 | (.001–.156) | 1.437 | 0.261 | .046 | (1.006–2.052) |
| ȃCLEC6A | 0.134 | 0.031 | <.001 | (.072–.195) | 2.027 | 0.38 | <.001 | (1.404–2.928) |
| ȃDPP10 | 0.091 | 0.045 | .045 | (.002–.179) | 1.501 | 0.313 | .051 | (.998–2.259) |
| ȃFAM3B | 0.159 | 0.044 | <.001 | (.073–.245) | 2.143 | 0.481 | .001 | (1.381–3.325) |
| ȃHNMT | 0.076 | 0.03 | .013 | (.016–.135) | 1.423 | 0.199 | .012 | (1.082–1.872) |
| ȃIL-6 | 0.119 | 0.027 | <.001 | (.066–.171) | 1.754 | 0.238 | <.001 | (1.344–2.288) |
| ȃITM2A | 0.075 | 0.033 | .025 | (.009–.141) | 1.417 | 0.228 | .03 | (1.035–1.942) |
| ȃKRT19 | 0.078 | 0.027 | .004 | (.024–.131) | 1.431 | 0.183 | .005 | (1.114–1.837) |
| ȃLAMP3 | 0.086 | 0.03 | .004 | (.027–.144) | 1.501 | 0.22 | .006 | (1.126–2.002) |
| ȃLILRB4 | 0.086 | 0.038 | .023 | (.012–.159) | 1.493 | 0.267 | .025 | (1.051–2.121) |
| Inflammation panel | ||||||||
| ȃCCL11 | 0.13 | 0.044 | .003 | (.044–.216) | 1.848 | 0.399 | .004 | (1.211–2.821) |
| ȃCCL20 | 0.066 | 0.02 | .001 | (.027–.105) | 1.358 | 0.128 | .001 | (1.13–1.633) |
| ȃCCL25 | 0.098 | 0.03 | .001 | (.038–.157) | 1.565 | 0.228 | .002 | (1.175–2.083) |
| ȃCDCP1 | 0.075 | 0.033 | .023 | (.01–.139) | 1.411 | 0.214 | .023 | (1.048–1.899) |
| ȃCSF1 | 0.167 | 0.062 | .007 | (.045–.289) | 2.653 | 1.057 | .014 | (1.215–5.791) |
| ȃCST5 | 0.108 | 0.043 | .012 | (.023–.192) | 1.674 | 0.346 | .013 | (1.116–2.511) |
| ȃCX3CL1 | 0.132 | 0.047 | .005 | (.039–.225) | 1.852 | 0.419 | .006 | (1.188–2.885) |
| ȃFGF19 | 0.06 | 0.026 | .019 | (.01–.111) | 1.329 | 0.165 | .022 | (1.041–1.696) |
| ȃFGF21 | 0.043 | 0.016 | .006 | (.012–.073) | 1.222 | 0.091 | .007 | (1.057–1.413) |
| ȃFGF23 | 0.097 | 0.036 | .008 | (.026–.169) | 1.574 | 0.3 | .018 | (1.082–2.288) |
| ȃFLT3L | 0.164 | 0.05 | .001 | (.066–.263) | 2.224 | 0.564 | .002 | (1.352–3.656) |
| ȃHGF | 0.207 | 0.036 | <.001 | (.135–.278) | 2.777 | 0.603 | <.001 | (1.815–4.249) |
| ȃIL-10 | 0.067 | 0.03 | .025 | (.009–.126) | 1.365 | 0.196 | .03 | (1.031–1.808) |
| ȃIL-10RB | 0.253 | 0.061 | <.001 | (.133–.373) | 3.344 | 1.055 | <.001 | (1.803–6.205) |
| ȃIL-17C | 0.084 | 0.028 | .003 | (.029–.14) | 1.501 | 0.218 | .005 | (1.129–1.995) |
| ȃIL-6 | 0.127 | 0.028 | <.001 | (.073–.181) | 1.831 | 0.262 | <.001 | (1.383–2.424) |
| ȃIL-8 | 0.075 | 0.028 | .007 | (.02–.13) | 1.414 | 0.19 | .01 | (1.086–1.84) |
| ȃLIFR | 0.224 | 0.069 | .001 | (.088–.36) | 2.933 | 1.072 | .003 | (1.433–6.003) |
| ȃMCP1 | 0.147 | 0.049 | .003 | (.051–.243) | 2.027 | 0.493 | .004 | (1.259–3.266) |
| ȃMMP10 | 0.123 | 0.037 | .001 | (.052–.195) | 1.78 | 0.325 | .002 | (1.245–2.544) |
| ȃOPG | 0.199 | 0.062 | .001 | (.078–.321) | 2.595 | 0.808 | .002 | (1.409–4.777) |
| ȃTNFRSF9 | 0.085 | 0.042 | .041 | (.003–.167) | 1.493 | 0.297 | .044 | (1.011–2.204) |
| ȃUPA | 0.154 | 0.059 | .009 | (.039–.269) | 2.058 | 0.579 | .01 | (1.185–3.572) |
| ȃVEGFA | 0.095 | 0.043 | .027 | (.011–.18) | 1.545 | 0.308 | .029 | (1.045–2.283) |
To better understand the interplay between these biomarkers, models were fit that included all of them in addition to relevant covariates. Data from cardiovascular panel III (IGFBP7), the immune response panel (CLEC6A), and the inflammation panel (HGF, IL-6, IL-10RB) were merged for this analysis. Proteins were dropped according to the backwards selection scheme described above. This resulted in a final model with CLEC6A (odds ratio [OR] = 1.48; P = .037; 95% confidence interval [CI], 1.02–2.13), HGF (OR = 1.83; P = .012; 95% CI, 1.14–2.95), and IL-6 (OR = 1.45; P = .016; 95% CI, 1.07–1.96) increasing the risk for the CVD composite outcome. Because proteins are on the log2 scale the OR corresponds to the change in the odds ratio associated with a doubling of the protein level. The final model also indicated differences in CVD risk across the study arms.
Mendelian Randomization Analysis
With 385 proteins, we can take all significance levels in the sequential testing strategy to be .05 and still control the family-wise error rate at 5%. This is what is reported here: Figure 1 provides a graphical presentation of the following results. As noted above, 86 proteins were associated with CVD at a significance level of .05, hence these 86 proteins were selected during the first stage of the sequential procedure. Of these 86 proteins, 10 had no SNPs in the region of its corresponding structural gene and 29 were associated with at least 1 SNP in the corresponding gene. The numbers of such SNPs from the proteins identified as being associated with CVD outcomes in the previous section are as follows (in parentheses): CLEC6A (41), HGF (0), IL-6 (0), IL-10RB (0), and IGFBP7 (4). Supplementary Table 17 provides a summary of the variants that pass various levels of filtering for each protein associated with the CVD composite outcome with at least 1 variant in its coding region. The filtering displayed in Supplementary Table 17 is cumulative, showing the resulting number of variants after each application of the sequential filtering strategy. Of the 29 proteins that were associated with at least 1 SNP in their coding region, the sequential procedure detected significant associations between 5 proteins and CVD outcomes: these were AXL (P = .021), CHI3L1 (P = .045), GAS6 (P = .015), IL-6RA (P = .049), and SCGB3A2 (P = .038). Table 4 presents results of all MR tests conducted. All F statistics used for diagnosis of weak instruments exceeded 10, indicating the appropriateness of the approximations used for inference. While only 1 SNP served as an instrument for AXL, all other proteins used a haplotype involving multiple SNPs as an instrument with the numbers of SNPs for each protein as follows (numbers of SNPs in parentheses): CHI3L1 (4), GAS6 (2), IL-6RA (68), and SCGB3A2 (27). While all these proteins exhibited a positive association with the composite CVD outcome in the previously described linear and logistic models, CHI3L1 and GAS6 have negative associations with the composite CVD outcome in the MR analysis. Additional detailed results can be found in the Supplementary Materials.
Figure 1.
Schematic representation of analysis pipeline. All 5 panels had 92 proteins and the number of proteins passing quality control in each panel is presented on the far left. The next column of boxes from the left shows the number of proteins that differ between cases and controls at a significance level of .05 and the number of these proteins that have at least 1 variant. On the far right, information on quality control of genetic variants is presented. Following the description of overall quality control, the number of variants in a protein that differs between cases and controls for a panel is provided. For example, there are 8 proteins in the cardiometabolic panel that differ and 6 of these proteins have at least 1 variant. Across these 6 proteins there are 205 variants in their coding region for a mean of 34 variants per protein. The middle column displays the number of proteins from each panel that have single-nucleotide polymorphisms that are significantly associated with protein levels: these are the proteins for which a MR test was conducted. The protein IL-10 appears in 2 panels (immune response and inflammation) hence the number of MR candidates in this figure exceeds the number of rows in Table 4 by 1. Duplication of proteins across panels and differences in analytical results across the duplicated proteins also creates differences between the results in Tables 2 and 3 and this figure (for the cardiovascular III, immune response, and inflammation panels). Abbreviations: CVD, cardiovascular disease; IL, interleukin; MR, mendelian randomization.
Table 4.
Summaries of Mendelian Randomization Tests Between Protein Levels With the Cardiovascular Disease Composite Outcome
| Protein | Estimate | SE | P Value | (95% CI) |
|---|---|---|---|---|
| ADAMTS13 | 1.286 | 1.722 | .455 | (−2.088 to 4.661) |
| AXL | 1.628 | 0.704 | .021 | (.248 to 3.008) |
| CCL18 | 0.182 | 0.174 | .295 | (−.159 to .523) |
| CHI3L1 | −0.150 | 0.075 | .045 | (−.296 to −.003) |
| CHIT1 | −0.160 | 1.367 | .907 | (−2.839 to 2.519) |
| CLEC4C | −0.014 | 0.303 | .964 | (−.608 to .581) |
| CLEC4D | −0.247 | 0.231 | .286 | (−.7 to .207) |
| CLEC6A | 0.424 | 0.230 | .065 | (−.027 to .875) |
| CPA1 | 0.266 | 0.367 | .469 | (−.454 to .986) |
| CPB1 | 0.004 | 0.493 | .993 | (−.962 to 0.97) |
| CST3 | −0.480 | 0.345 | .164 | (−1.155 to .196) |
| CSTB | 0.008 | 0.334 | .981 | (−.647 to .663) |
| CTSD | −0.268 | 0.259 | .301 | (−.777 to 0.24) |
| FAM3B | 0.722 | 0.900 | .422 | (−1.042 to 2.486) |
| FAS | −0.288 | 0.586 | .623 | (−1.437 to .86) |
| GAL9 | −1.018 | 0.852 | .232 | (−2.688 to .652) |
| GAS6 | −0.766 | 0.316 | .015 | (−1.386 to −.147) |
| GRN | 1.031 | 0.846 | .223 | (−.626 to 2.689) |
| HNMT | 0.211 | 0.117 | .072 | (−.019 to .441) |
| IGFBP7 | −0.121 | 0.417 | .771 | (−.939 to .696) |
| IL-10 | 0.243 | 0.396 | .539 | (−.533 to 1.018) |
| IL-2RA | 0.567 | 0.315 | .072 | (−.051 to 1.186) |
| IL-6RA | 0.177 | 0.090 | .049 | (.001 to .353) |
| LAMP3 | −0.109 | 0.227 | .630 | (−.554 to .336) |
| LILRB4 | 1.463 | 2.023 | .470 | (−2.502 to 5.427) |
| LTBR | 0.336 | 0.477 | .481 | (−.599 to 1.272) |
| PCSK9 | 0.172 | 0.389 | .658 | (−.59 to .935) |
| RARRES2 | −0.792 | 0.818 | .333 | (−2.395 to .81) |
| SCGB3A2 | 0.397 | 0.191 | .038 | (.023 to .771) |
DISCUSSION
Strong evidence was detected for an association between multiple protein biomarkers and adjudicated CVD outcomes in a global population of PWH. There were associations between IGFBP7, CLEC6A, IL-10RB, HGF, and IL-6 as individual biomarkers and CVD outcomes. Because biomarker levels are generally correlated, investigation of joint models examined how the collection of these proteins impact CVD risk. Joint models identified IL-6, HGF, and CLEC6A as each independently contributing to CVD risk given the effect of the other proteins. Higher levels of all 3 of these proteins are associated with increased risk for CVD. These proteins were at elevated levels over 6 years prior to the development of CVD in some individuals. This suggests that there may be an association between protein levels and subclinical atherosclerotic disease long before overt CVD.
A strategy of constructing haplotypes and using these haplotypes as instruments in MR analysis provided evidence for causal effects of 5 proteins on CVD outcomes: AXL, IL-6RA, CHI3L1, SCGB3A2, and GAS6. Our strategy was unable to find suitable genetic instruments for IL-6, IL-10RB, and HGF, thus we could not effectively test for causal effects for these biomarkers, which appear to at least be associated with CVD outcomes. One of the biomarkers with evidence for a causal role, IL-6RA, has been demonstrated to have a causal effect on CVD in the general population, hence finding this in a PWH population lends support to the hypothesis that at least some mechanisms of CVD development are common among PWH and the general population. GAS6 has been described as a ligand for AXL so finding both as being involved in the causal pathway for CVD implicates processes that involve this pathway in CVD development. It is intriguing that the direction of the effects for CHI3L1 and GAS6 differed in the MR analysis compared to what we found in the linear model. The most likely explanation is that the models that detect the positive associations are excluding critical covariates (perhaps the levels of other proteins or metabolites), and if these critical covariates had been measured and included the direction of the association would change in the linear models.
There is support in the literature for the proteins identified as being associated with CVD outcomes. The role of IL-6 in inflammatory processes relating to CVD among PWH is well established [6, 8] and there are reviews that describe its role in inflammatory processes in the general population [28]. There are also a number of publications that link HGF to various aspects of CVD in the general population [29, 30], but this appears to be the first such association described among PWH. This protein is thought to play a role in cardiovascular remodeling in response to endothelial injury. More specifically it is thought to modulate inflammatory responses in immune cells and to prevent fibrosis mediated by fibroblasts via transforming growth factor-β1 (TGF-β1)-dependent mechanisms (these have been reviewed [31]). The association between CLEC6A (also known as DECTIN-2) levels and CVD outcomes appears to be novel. CLEC6A is a C-type lectin receptor expressed by macrophages, dendritic cells, and monocytes with ligands derived from pathogens that induce inflammatory signals upon binding [32].
The literature on the proteins identified using the MR approach is mixed and not entirely consistent with the findings presented here. AXL and its primary ligand GAS6 have been found to be associated with CVD previously and they have been described as being upregulated in response to inflammation and also characterized as being involved in the resolution of inflammatory signals [33, 34]. Similarly, CHI3L1 has been found at elevated levels in patients with a variety of inflammatory conditions, including CVD and has been variously described as having pro- and anti-inflammatory roles [35, 36]. IL-6RA has been well documented to have a critical role in inflammation [6, 8]. SCGB3A2 has been described as having antifibrotic and anti-inflammatory properties [37, 38].
Our findings have some limitations. We do not have comprehensive data on established CVD risk factors across all trials. In particular, we do not have data on smoking or cholesterol levels, and this may induce some residual confounding in our tests for associations between protein levels and the composite CVD outcome. Moreover, these variables could act as effect modifiers, and we would not be able to detect this. However, our models did control for a diagnosis of diabetes and hypertension so some of the potential confounding was accounted for. Because we do not know that the CVD events described here were necessarily the first such event experienced by study participants there is some potential for reverse causality. On the other hand, many of the events took place years after the specimens examined here were obtained. In addition, we do not have data on the type of stroke or myocardial infarction across all trials, so further mechanistic understanding of the impact of these proteins is limited. Finally, it is difficult to assess the exclusion restriction assumption in MR analysis that provides the basis for a causal interpretation of the associations described here. To reduce the potential for this we restricted our consideration of SNPs to those near the coding region of the protein of interest.
Our findings indicate that final common pathways leading to the development of CVD among PWH are similar to the general population. This is particularly the case for IL-6– and GAS6-dependent pathways. We note, however, that the proteins included on the CVD panels were selected based on risk in an HIV-negative setting, thus these panels exclude proteins unique to CVD pathogenesis in PWH should such proteins exist. We also report for the first time associations between biomarkers of fibrotic/inflammatory pathways with CVD among PWH. Further investigations are warranted to (1) define whether this reflects mechanistic differences in the development of CVD between PWH and the general population, and (2) assess the clinical relevance of using these pathways as potential targets for new CVD therapies.
Supplementary Data
Supplementary materials are available at The Journal of Infectious Diseases online. These consist of data provided by the authors to benefit the reader. The posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.
Supplementary Material
Contributor Information
Cavan S Reilly, Division of Biostatistics, University of Minnesota, Minneapolis, Minnesota, USA.
Álvaro H Borges, Immunology, Statens Serum Institut, Copenhagen, Denmark.
Jason V Baker, HIV Medicine, Infectious Diseases, Hennepin County Medical Center, Minneapolis, Minnesota, USA.
Sandra E Safo, Division of Biostatistics, University of Minnesota, Minneapolis, Minnesota, USA.
Shweta Sharma, Division of Biostatistics, University of Minnesota, Minneapolis, Minnesota, USA.
Mark N Polizzotto, Department of Medicine, Australian National University, Canberra, Australia.
James S Pankow, Division of Epidemiology and Community Health, University of Minnesota, Minneapolis, Minnesota, USA.
Xiaojun Hu, Animal and Plant Inspection Service, US Department of Agriculture, Beltsville, Maryland, USA.
Brad T Sherman, Laboratory of Human Retrovirology and Immunoinformatics, Frederick National Laboratories, Frederick, Maryland, USA.
Abdel G Babiker, Epidemiology and Medical Statistics, University College London, London, United Kingdom.
Jens D Lundgren, Department of Infectious Diseases, University of Copenhagen, Copenhagen, Denmark.
H Clifford Lane, Division of Clinical Research, National Institutes of Allergy and Infectious Diseases, Bethesda, Maryland, USA.
Notes
Financial support . This work was supported by the National Institutes of Health (grant number HHSN261200800001E).
References
- 1. Freiberg MS, Chang CC, Kuller LH, et al. HIV Infection and the risk of acute myocardial infarction. JAMA Intern Med 2013; 173:614–22. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Baker JV, Lundgren JD. Cardiovascular implications from untreated human immunodeficiency virus infection. Eur Heart J 2011; 32:945–51. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Hsue PY, Deeks SG, Hunt PW. Immunologic basis of cardiovascular disease in HIV-infected adults. J Infect Dis 2012; 205(Suppl 3):S375–82. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Rasmussen LD, Helleberg M, May MT, et al. Myocardial infarction among Danish HIV-infected individuals: population-attributable fractions associated with smoking. Clin Infect Dis 2015; 60:1415–23. [DOI] [PubMed] [Google Scholar]
- 5. Duprez DA, Neuhaus J, Kuller LH, et al. Inflammation, coagulation and cardiovascular disease in HIV-infected individuals. PLoS One 2012; 7:e44454. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Nordell AD, McKenna M, Borges ÁH, et al. SILCAAT Scientific Committee. Severity of cardiovascular disease outcomes among patients with HIV is related to markers of inflammation and coagulation. J Am Heart Assoc 2014; 3:e000844. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Grund B, Baker JV, Deeks SG, et al. Relevance of interleukin-6 and D-dimer for serious non-AIDS morbidity and death among HIV-positive adults on suppressive antiretroviral therapy. PLoS One 2016; 11:e0155100. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Borges ÁH, O'Connor JL, Phillips AN, et al. Interleukin 6 is a stronger predictor of clinical events than high-sensitivity C-reactive protein or D-dimer during HIV infection. J Infect Dis 2016; 214:408–16. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Baker JV, Peng G, Rapkin J, et al. CD4+ count and risk of non-AIDS diseases following initial treatment for HIV infection. AIDS 2008; 22:841–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Baker JV, Peng G, Rapkin J, et al. Poor initial CD4+ recovery with antiretroviral therapy prolongs immune depletion and increases risk for AIDS and non-AIDS diseases. J Acquir Immune Defic Syndr 2008; 48:541–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Phillips AN, Neaton JD, Lundgren JD. The role of HIV in serious diseases other than AIDS. AIDS 2008; 22:2409–18. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Imamichi H, Smith M, Adelsberger JW, et al. Defective HIV-1 proviruses produce viral proteins. Proc Natl Acad Sci USA 2020; 117:3704–10. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Friis-Møller N, Reiss P, Sabin CA, et al. Class of antiretroviral drug and myocardial infarction. N Engl J Med 2007; 356:1723–35. [DOI] [PubMed] [Google Scholar]
- 14. Eyawo O, Brockman G, Goldsmith CH, et al. Risk of myocardial infarction among people living with HIV: an updated systematic review and meta-analysis. BMJ Open 2019; 9:e025874. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Shah ASV, Stelzle D, Lee KK, et al. Global burden of atherosclerotic cardiovascular disease in people living with HIV. Circulation 2018; 138:1100–12. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Kearns A, Gordon J, Burdo TH, Qin X. HIV-1-associated atherosclerosis: unraveling the missing link. J Am Coll Cardiol 2017; 69:3084–98. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Hsue PY. Mechanisms of cardiovascular disease in the setting of HIV infection. Can J Cardiol 2019; 35:238–48. [DOI] [PubMed] [Google Scholar]
- 18. Lind L, Siegbahn A, Lindahl B, Stenemo M, Sundström J, Ärnlöv J. Discovery of new risk markers for ischemic stroke using a novel targeted proteomics chip. Stroke 2015; 46:3340–7. [DOI] [PubMed] [Google Scholar]
- 19. Schiopu A, Bengtsson E, Gonçalves I, Nilsson J, Fredrikson GN, Björkbacka H. Associations between macrophage colony-stimulating factor and monocyte chemotactic protein 1 in plasma and first-time coronary events: a nested case-control study. J Am Heart Assoc 2016; 5:e002851. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Gonçalves I, Hultman K, Dunér P, et al. High levels of cathepsin D and cystatin B are associated with increased risk of coronary events. Open Heart 2016; 3:e000353. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Youngman LD, Keavney BD, Palmer A. Plasma fibrinogen and fibrinogen genotypes in 4685 cases of myocardial infarction and in 6002 controls: test of causality by ‘mendelian randomization’. Circulation 2000; 102:31–2. [Google Scholar]
- 22. Lawlor DA, Harbord RM, Sterne JAC, Timpson N, Davey Smith G. Mendelian randomization: using genes as instruments for making causal inferences in epidemiology. Stat Med 2008; 27:1133–63. [DOI] [PubMed] [Google Scholar]
- 23. Voight BF, Peloso GM, Orhu-Melander M, et al. Plasma HDL cholesterol and risk of myocardial infarction: a mendelian randomisation study. Lancet 2012; 380:572–80. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Swerdlow DI, Holmes MV, Kuchenbaecker KB, et al. The interleukin-6 receptor as a target for the prevention of coronary heart disease: a mendelian randomisation analysis. Lancet 2012; 379:1214–24. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. MacArthur RD, Novak RM, Peng G, et al. A comparison of three highly active antiretroviral treatment strategies consisting of non-nucleoside reverse transcriptase inhibitors, protease inhibitors, or both in the presence of nucleoside reverse transcriptase inhibitors as initial therapy (CPCRA 058 FIRST study): a long-term randomized trial. Lancet 2006; 368:2125–35. [DOI] [PubMed] [Google Scholar]
- 26. Abrams D, Lévy Y, Losso MH, et al. Interleukin-2 therapy in patients with HIV infection. N Engl J Med 2009; 361:1548–59. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. The Strategies for Management of Antiretroviral Therapy (SMART) Study Group . CD4+ count–guided Interruption of antiretroviral treatment. N Engl J Med 2006; 355:2283–96. [DOI] [PubMed] [Google Scholar]
- 28. Lundgren JD, Babiker AG, Gordin F, et al. Initiation of antiretroviral therapy in early asymptomatic HIV infection. N Engl J Med 2015; 373:795–807. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Hunter CA, Jones SA. IL-6 as a keystone cytokine in health and disease. Nat Immunol 2015; 16:448–57. [DOI] [PubMed] [Google Scholar]
- 30. Bell EJ, Larson NB, Decker PA, et al. Hepatocyte growth factor is positively associated with risk of stroke: the MESA. Stroke 2016; 47:2689–94. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Bell EJ, Decker PA, Tsai MY, et al. Hepatocyte growth factor is associated with progression of atherosclerosis: the multi-ethnic study of atherosclerosis (MESA). Atherosclerosis 2018; 272:162–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Gallo S, Sala V, Gatti S, Crepaldi T. Cellular and molecular mechanisms of HGF/met in the cardiovascular system. Clin Sci 2015; 129:1173–93. [DOI] [PubMed] [Google Scholar]
- 33. Sancho D, Reis e Sousa C. Signaling by myeloid C-type lectin receptors in immunity and homeostasis. Annu Rev Immunol 2012; 30:491–529. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Zagórska A, Través PG, Lew ED, et al. Diversification of TAM receptor tyrosine kinase function. Nat Immunol 2014; 15:920–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Carrera-Silva EA, Chan PY, Joannas L, et al. T cell-derived protein S engages TAM receptor signaling in dendritic cells to control the magnitude of the immune response. Immunity 2013; 39:160–70. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. Maegdefessel L, Spin JM, Raaz U, et al. miR-24 limits aortic vascular inflammation and murine abdominal aneurysm development. Nat Commun 2014; 5:5214. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37. Zhou Y, Peng H, Sun H, et al. Chitinase 3-like 1 suppresses injury and promotes fibroproliferative responses in mammalian lung fibrosis. Sci Transl Med 2014; 6:240ra76. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38. Cai Y, Kimura S. Secretoglobin 3A2 exhibits anti-fibrotic activity in bleomycin-induced pulmonary fibrosis model mice. PLoS One 2015; 10:e0142497. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. Kido T, Yoneda M, Cai Y, Matsubara T, Ward JM, Kimura S. Secretoglobin superfamily protein SCGB3A2 deficiency potentiates ovalbumin-induced allergic pulmonary inflammation. Mediators Inflamm 2014; 2014:216465. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.