Country of residence is associated with distinct inflammatory biomarker signatures in HIV-infected patients.

BACKGROUND: Inflammation and coagulation biomarkers are independent predictors of morbidity and mortality in HIV-infected patients. The impact of country of residence on these biomarkers is unknown and was investigated in persons at similar stages of HIV infection.
METHODS: Cryopreserved plasma specimens were analysed from 267 ART-naive patients with CD4 cell counts <100 cells/μl from Mexico (n=124) and South Africa (n=143). Biomarkers were compared and dimension reduction analyses were performed to highlight biosignatures according to nationality, gender and tuberculosis co-infection.
RESULTS: Mexican patients were significantly different from South Africans with regard to age, gender, CD4 cell count, haemoglobin, presence of AIDS-defining illness and prevalence of active tuberculosis. After adjusting for baseline characteristics, patients from Mexico had higher levels of IFN-γ, IL-8, and CXCL-10 whereas patients from South Africa had higher levels of fibrinogen, LTB4, P-selectin, protein S, and sCD40 ligand. The effect of country on the profile of biomarker expression was stronger than gender differences and tuberculosis co-infection.
CONCLUSION: Inflammation and coagulation biomarkers vary significantly by country. Further studies are needed to evaluate how these differences may contribute to HIV pathogenesis and prognosis in diverse populations and how they can be accounted for in studies using biomarkers as surrogate end points.

Introduction

Biomarkers of inflammation and coagulation have emerged as independent predictors of morbidity and mortality in HIV-infected patients in several studies [1-6]. In particular IL-6 and D-dimer [3,7-9], CRP [8] and sCD14 [5] have been strongly associated with all-cause mortality and cardiovascular mortality [10]. IL-6, CRP and D-dimers have also been linked with cardiovascular disease [1]. Most of these studies included large numbers of patients from several countries, but predominantly from North America and Europe.

Many studies focused on cardiovascular disease have shown that the inflammatory profile may vary by geographic location or ethnicity [11-18]. Nationality- or ethnicity-driven variation of biomarker concentrations in the blood have been explored in the context of infectious diseases such as tuberculosis [12]. Nevertheless, the potential effects of country of residence on the inflammatory profile of HIV-infected persons remain largely unknown. Recognising a possible association may have important implications for the potential use of biomarkers as surrogate markers or clinical trial endpoints.

In the present study, we examined the potential effect of nationality on circulating biomarkers by studying HIV-infected, ART naive patients participating in a randomised clinical trial in two different countries (CADIRIS trial, NCT00988780), a trial in which biomarkers predictive of IRIS were recently analysed [19]. Our aim was to evaluate whether patients living with HIV in different regions (country of residence), but at similar stages of infection, have significant differences in plasma levels of biomarkers of inflammation and coagulation.

Methods

Study outline

The CADIRIS trial (CCR5 Antagonism to Decrease the Incidence of the Immune Reconstitution Inflammatory Syndrome in HIV-infected Patients, NCT00988780) was a double-blinded trial that randomly allocated 276 ART-naive patients with AIDS (CD4 cell count <100 cells/μL) to receive maraviroc or placebo in combination with a standard antiretroviral regimen (efavirenz, tenofovir and emtricitabine) to assess whether maraviroc could decrease the incidence of IRIS [20,21]. Patients were screened from 2009 to 2012 at clinical sites in Mexico and South Africa. The primary endpoint was IRIS diagnosis by week 24 of ART initiation. All patients provided written informed consent. A full description of the clinical trial and outcomes has been previously published [21]. This study was sponsored by the Instituo Nacional de Cicencias Medicas y Nutricion Salvador Zubiran and approved by the Ministry of Health and Federal Commission for Sanitary Risks Protection of Mexico, and the Medicines Control Council and Human Research Ethics Committee of South Africa. We evaluated levels of biomarkers based upon nationality, gender, and TB status of the CADIRIS population in 267 available cryopreserved plasma specimens from the pre-ART timepoint.

Biomarker measurements

All plasma specimens were obtained at the time of enrolment, prior to ART initiation, and were stored at −80°C. Biomarkers and coagulation markers were measured in plasma. Specimens were collected in EDTA or citrate (for protein C, S, von Willebrand factor and vitamin D measurement). Levels of IFN-γ, IL-10, IL-12p70, IL-1β, IL-6, TNF-α, CRP, SAA, P-selectin, CXCL10 and IL-17 were measured by electrochemiluminescence using a multiplex plate (MESO scale discovery). Levels of sCD14, sCD40L, sCD163, vWF, fibrinogen, protein C and S were measured using enzyme-linked immunosorbent assays (ELISAs) according to the manufacturers’ instructions (R&D Systems, AdipoBioscience, Zymutest). Levels of 25-hydroxyvitamin D were also measured using ELISA (ALPCO). D-dimer was measured using an enzyme-linked fluorescent assay (Biomeriuex) and hyaluronic acid (HA) was measured using an enzyme coagulation kit (Corgenix).

Statistical methods

Baseline characteristics were summarised with frequencies or medians and interquartile ranges (IQR) as appropriate. Demographic data were analysed using Wilcoxon rank-sum test and Fisher's test. Data were logarithmically transformed, and comparisons of biomarker distributions by country, gender and TB status were performed using the Student's t-test. Confounding factors were chosen based upon significant differences in baseline characteristics and adjusted for using a multivariate generalised linear model. To compensate for multiple comparisons, we defined statistical significance as a P-value ≤0.01.

Dimension reduction analyses

A principal component analysis (PCA) model, utilising a combination of plasma concentrations of several biomarkers that exhibited statistical differences assessed in univariate analyses between patients from Mexico and South Africa, was employed to test whether individuals from these different countries could be distinguished. A similar approach has been successfully used previously [22-24]. Unsupervised two-way hierarchical cluster analyses (Ward's method) with bootstrap, were used to test whether males and females from the different countries exhibiting differential levels of biomarkers in plasma could be grouped separately. The same analysis was performed to assess the overall biomarker profile between patients with TB versus those without TB co-infection. This tested whether gender differences or differences based on TB would influence the potential differences observed in the analysis by country.

The inferential networks (host interactome) were generated from Spearman correlation matrices containing values of each biomarker measured in the plasma samples and blood parameters. In this analysis, each mediator is selected as a target, and the software program searches among the other mediators for those that are correlated, with the target calculating a correlation matrix using Spearman rank tests. As a result, the features related to the selected target are linked. The links shown in the networks represent highly statistically significant Spearman rank correlations (P<0.001). To compensate for differences in the number of individuals in the distinct study groups (e.g. male vs female or TB vs no TB), we employed the bootstrap method (100X). Only correlations remaining with P<0.001 in at least 40 out of 100 times were considered.

Sparse canonical correlation analysis (CCA) modelling was employed to assess whether combinations of circulating biomarkers can discriminate between patients from Mexico or South Africa. The CCA model was chosen because many variables were studied. This model is able to perform dimensionality reduction for two co-dependent data sets (biomarker profile and baseline characteristics profile) simultaneously so that the discrimination of the clinical endpoints represents a combination of variables that are maximally correlated. Thus, trends of correlations between parameters in different clinical groups rather than their respective distribution within each group are the key components driving the discrimination outcome. In our CCA model, simplified and adapted from previously reported investigations of biomarkers for TB diagnosis [25,26], linear regression graphs represent coefficients from different combinations of plasma factors and baseline characteristics. In the biomarker profile dataset, we included values of variables that exhibited significant differences (P<0.05) in the multivariate analysis (CXCL10, fibrinogen, hyaluronic acid, IFN-γ, IL-8, LTB4, P-selectin, protein S, vitamin D and sCD40L; see Appendix 1 for details). The diagnostic class prediction values obtained were calculated using receiver operator characteristics curve analysis.

Results

Baseline characteristics

The 267 participants who had baseline pre-ART samples available had similar baseline demographics in the maraviroc and placebo arms and the incidence of IRIS did not differ in the two randomisation arms [21]. One hundred and twenty-four patients were enrolled in Mexico (five clinical sites) and 143 patients were enrolled in South Africa (one clinical site). The Mexican sites were three HIV tertiary care clinics in university-affiliated hospitals, one sexually transmitted diseases clinic and one HIV clinic, and the South African site was an HIV-referral clinic based at a tertiary care university-affiliated hospital in Johannesburg.

When comparing baseline demographics between the two nations, we found that patients from the Mexican cohort were younger compared to the South African cohort (35 vs 38 years, respectively, P=0.006; Table 1) and included a lower proportion of women (12% vs 54% female, respectively, P<0.0001). Patients from South Africa were predominantly black (95.8%) with a small proportion of patients who were Asian (1.4%) and mixed race (2.8%). Most patients from Mexico were of mixed race (88.71%), while the remaining patients were white (11.29%)

Table 1.

Baseline characteristics of study participants based upon country of origin

	Mexico (n=124)	South Africa (n=143)	P-valuea
Age (years)	35[29–41]	38[33–44]	0.006
Gender (% female)	12	54	<0.001*
CD4 cell count (cells/μL)	31[15–53]	37[21–64]	0.030
CD8 cell count (cells/μL)	478[361–786]	473[326–680]	0.320
Plasma HIV RNA (log10 copies/mL)	5.4[5.1–5.7]	5.4[5–5.8]	0.560
Haemoglobin (g/dL)	12.4[11–14]	12[10.8–13]	0.030
AIDS-defining conditions	79%(98)	43%(61)	<0.001
HIV-associated wasting	52%(64)	5%(7)	<0.001
Tuberculosis	17%(21)	28%(40)	0.040

Data are reported as median [IQR] or as percentage

Significance was determined using Wilcoxon rank-sum test or Fisher's test*

Patients in the Mexican cohort also had a lower median CD4 cell count than did patients from South Africa (31 vs 37 cells/μL, P=0.034) and more AIDS-defining illnesses (79 vs 43% of patients, respectively, P≤0.001). Baseline HIV viral load and CD8 T cell count did not differ significantly by country. The patients in the South African cohort had lower haemoglobin levels than did the Mexican cohort (12 vs 12.4 g/dL, P=0.030).

The most common AIDS-defining illnesses in the CADIRIS trial at enrolment were HIV-associated wasting and tuberculosis [21]. The Mexican cohort had a higher prevalence of HIV-associated wasting than did the South African cohort (52% vs 5%, respectively, P≤0.0001), whereas the South African cohort had a higher prevalence of active TB infection (28% vs 17% of patients, P=0.04; Table 1). At enrolment, 89% of patients with active TB infection were on therapy, while 100% of patients in the Mexican cohort on therapy, and 83% of patients in the South African cohort on therapy (P=0.08). The remaining opportunistic infections included CMV disease, Kaposi's sarcoma, oesophageal candidiasis, cryptococcosis, histoplasmosis, cryptosporidiosis, Pneumocystis jirovecii pneumonia, progressive multifocal leukoencephalopathy and isoporiasis [21].

Biomarkers by country

Median levels of the baseline cytokines and biomarkers from each country were compared using univariate analysis with a Student's t-test. As shown in Figure 1, patients from Mexico had significantly higher levels of hyaluronic acid, IFN-γ, IL-8, CXCL10 and sCD14 than did patients from South Africa, whereas patients from South Africa had higher levels of vitamin D, fibrinogen, LTB4, P-selectin, protein S, and sCD40 ligand (Figure 1A–C and Appendix 1). To explore the hypothesis that HIV viral clades (Mexican vs South African) could have different effects on plasma markers, we performed correlation analyses between viraemia and all biomarkers assessed separately in Mexicans and South Africans (Figure 1D). We found that in Mexico, HIV viraemia was robustly positively correlated with hyaluronic acid, IL-10, CXCL10 and vWF (Figure 1D). In South Africa, HIV viraemia remained positively correlated with CXCL10; however, additional positive correlations became significant such as those with sCD14 and IFN-γ (Figure 1D). Moreover, viraemia was negatively correlated with circulating levels of IL-12p70 in South Africa but not in Mexico. These findings indicate that viral clades could influence distinct aspects of the host immune responses in different countries.

Figure 1.

Differences in expression profile of plasma biomarkers between Mexico and South Africa. (A, B) Plasma concentrations of several biomarkers were compared between patients from Mexico (n=124) or South Africa (n=143) with univariate analysis using a generalised linear model. (C) Principal component analysis was employed using data from the biomarkers displayed in (A) and (B). Normal ellipsoids represent area estimated to contain 50% of the population of a group. (D) Spearman correlations between HIV viremia and several biomarkers are shown for patients from Mexico or South Africa. P-values of significant correlations are shown

In order to further assess the relationship of country with biomarker levels, sparse canonical correlation analysis (sCCA)was performed to demonstrate the distribution of biomarkers between the two countries as shown in Figure 2. This model had 90.3 % sensitivity and 86.5 % specificity to accurately predict the country based upon the biomarker profile. When these values were adjusted for significant differences in baseline characteristics, IFN-γ, IL-8 and CXCL10 remained significantly elevated in the Mexican cohort, and vitamin D, fibrinogen, LTB4, P-selectin, protein S, and sCD40 ligand remained significantly elevated in the South African cohort (Appendix 1). With adjustments for multiple comparisons, IL-8 and CXCL10 remained significantly elevated in the Mexican cohort, and fibrinogen, LTB4, P-selectin, protein S and sCD40 ligand remained elevated in the South African cohort (Appendix 1).

Figure 2.

Discrimination of country of origin using combination of plasma biomarkers. In an exploratory approach, a sparse canonical correlation analysis (sCCA) was employed to test whether patients from Mexico or South Africa could be distinguished. The sCCA performs dimensionality reduction for two co-dependent data sets (biomarker profile and baseline characteristics profile) so that the trends of correlations between parameters from both countries represent a combination of variables that are maximally correlated. In the biomarker profile dataset, values of variables that were exhibited significant differences (P<0.05) in the multivariate analysis (CXCL10, fibrinogen, IFN-γ, IL-8, LTB4, P-selectin, protein S, vitamin D and sCD40L; see Appendix 1 for details). The overall performance of the model in discriminating participants from Mexico versus South Africa is shown

Biomarkers by gender

Based upon findings of significant differences among biomarkers by gender in the cardiovascular literature [13,14,27-29] and the significant skew of gender between the two countries, we decided to investigate the biomarker profile by gender in our CADIRIS cohort.

Baseline characteristics are demonstrated in Appendix 2. In addition to the significant difference of gender proportions by country, haemoglobin levels were found to be significantly different by gender with lower haemoglobin in females (11.45 vs 12.8 g/dL, P<0.001). Age, CD4 cell count, CD8 cell count, viral load, and TB status did not significantly differ by gender. Biomarkers did not significantly differ by oral contraceptive use in female patients.

Female patients had significantly higher median levels of fibrinogen, vitamin D, sCD40 ligand and vWF, whereas male patients had significantly higher median values of IL-6, IL-8 and CXCL10 (Figure 3B,C; Appendix 2). When these biomarkers were adjusted for country of origin, there was no significant difference between genders (Appendix 2).

Figure 3.

Gender differences in expression of plasma biomarkers. (A) Network analysis shows significant correlations (P<0.001, after bootstrap of 100X) between several biomarkers assessed by Spearman correlation matrices in male (n=175) and female (n=92) patients. Network densities were compared between the groups using the permutation test. (B) Plasma concentrations of selected biomarkers were compared between male and female patients with univariate analysis using a generalised linear model. (C) A hierarchical clustering analysis (Ward's method) of the plasma expression profile of several biomarkers was employed to test whether male and female patients from either Mexico or South Africa could be clustered separately. The colours represent the level of expression relative to the median

Because differences in biomarkers by gender have been reported in the literature and yet we found none in our study, we decided to perform a network analysis to investigate whether the relationships of the biomarkers, rather than their respective concentrations values per se, could be different by gender. As shown in Figure 3A, we found positive correlations between inflammatory markers including D-dimer, IL-6, TNF-α and IFN-γ in addition to negative correlation with haemoglobin, which can be suggestive of a tightly coordinated inflammatory response in men. Interestingly this observation is not seen in the female cohort, which may indicate that inflammatory responses may vary by gender. Moreover, hierarchical cluster analysis of the overall biomarker expression profile in plasma underlined that male and female patients from either Mexico or South Africa exhibit a distinct biosignature (Figure 3C), further highlighting the role of nationality in driving the differences detected in the biomarker values.

Biomarkers by TB status

We then investigated whether TB, the most prevalent opportunistic infection in this cohort, which was also significantly more prevalent in South Africa, had an important effect on the biomarker profile.

Patients infected with tuberculosis had higher median levels of CRP, D-dimer, hyaluronic acid, IFN-γ, CXCL10, TNF-α and sCD14 than did patients without tuberculosis as shown in Figure 4B (see also Appendix 3).

Figure 4.

Differential expression of plasma biomarkers in tuberculosis. (A) Network analysis shows significant correlations (P<0.001, after bootstrap of 100X) between several biomarkers assessed by Spearman correlation matrices in patients with no TB (n=206) or with a TB diagnosis (n=61). Network densities were compared between the groups using the permutation test. (B) Plasma concentrations of selected biomarkers were compared between patients with (green) or without TB (pink) with univariate analysis using a generalised linear model. (C) A hierarchical clustering analysis (Ward's method) of the plasma expression profile of several biomarkers was employed to test whether patients with or without TB and from either Mexico or South Africa could be clustered separately. The colours represent the level of expression relative to the median

When these biomarkers were adjusted for country of origin, haemoglobin and viral load, IFN-γ, CRP, hyaluronic acid, sCD14 and sCD40 ligand remained significantly higher in patients with tuberculosis (Appendix 3). After adjustments for multiple comparisons, IFN-γ and sCD14 remained significantly higher in patients with TB (Appendix 3).

Network analysis was performed, see Figure 4A. We observed that in patients without TB, inflammatory markers such as D-dimer, IL-6, TNF-α and IFN-γ were tightly correlated. In addition, significant negative correlations involving haemoglobin levels were observed in both groups of patients. A new hierarchical cluster analysis of the overall biomarker expression profile in plasma demonstrated that patients with or without TB from either Mexico or South Africa exhibit a distinct biosignature (Figure 4C). These findings suggest nationality as the central driver of certain observed differences in pro-inflammatory biomarker expression in plasma of HIV patients.

Discussion

In this study, we evaluated the effects of country of residence on the inflammatory and coagulation profile in HIV infection based upon nationality prior to ART initiation. To our knowledge this is one of the first studies to show a significant role of country of residence on levels of certain relevant biomarkers during HIV infection, which may have important pathogenesis implications and affect use of biomarkers as clinical trial endpoints.

In our study, which included patients from Mexico and South Africa at similar stages of HIV infection, we demonstrated a unique biomarker profile by country of origin during HIV infection prior to ART initiation. Although both cohorts had elevations of markers associated with immune activation, the South African cohort appeared to have higher baseline levels of coagulation markers (fibrinogen and protein S), platelet activation (sCD40 ligand) and leukocyte adhesion (LTB4 and P-selectin).

Elevated levels of P-selectin, fibrinogen, and hyaluronic acid have been associated with morbidity and mortality in HIV [4,6,30-32]. Boulware et al. has demonstrated that higher levels of hyaluronic acid levels pre-ART identified patients at higher risk of AIDS or death following ART initiation, and Musselwhite et al. found that elevated levels of P-selectin and hyaluronic acid were associated with increased risk of venous thromboembolism [4,30]. In addition, Tien et al. demonstrated that fibrinogen was associated with an increased risk of mortality in HIV-patients [6].

In studies including HIV-seronegative patients, there have been conflicting reports regarding fibrinogen levels in patients of different ethnicities [15,33,34], particularly when comparing Caucasians with black individuals. These studies were performed in different countries so it is possible that the country of origin plays an important role in addition to race.

P-selectin, protein S and fibrinogen have been evaluated in different ethnic groups in HIV-seronegative patients [15,35]. Miller et al. demonstrated significantly lower levels of P-selectin in people of African origin compared to Caucasians in the United Kingdom [15]. Dunne et al. investigated protein S levels in healthy individuals by ethnicity, and found that black Africans had significantly lower protein S levels than black Caribbeans, and Caucasians in the United Kingdom [35].

The Mexican cohort had significantly elevated levels of chemokines, specifically CXCL10, which has been shown to be involved with the Type 1 helper cell (TH1) response as well as lymphocyte recruitment, and IL-8, a chemokine produced by myeloid cells to promote chemotaxis and phagocytosis. During acute HIV infection, increased CXCL10levels have been associated with low CD4 counts later in the chronic phase31 [31] and elevated levels of IL-8 have been associated with mortality during an opportunistic infection [36] and IRIS [37].

In the context of the current literature on biomarkers and nationality, we found that levels of CXCL10 have been compared in studies; however, no studies compared South Africans and Mexicans. Sekikawa et al. found that Japanese individuals had higher levels of CXCL10 than did Caucasians and Americans of Japanese origin [38].

We further investigated biomarkers based on gender and TB status, as the gender distribution and TB status differed between the two countries. We did not see a significant difference when looking at biomarkers by gender in HIV infection prior to ART initiation, in contrast to observations in cardiovascular disease [13,14,39]; however, our network analysis is suggestive of distinguishable relationships among these inflammatory markers by gender.

When we investigated biomarkers by TB status, we found that sCD14 and IFN-γ were significantly higher in patients with TB. Our network analysis also demonstrated less coordination of biomarkers during TB infection, which is suggestive of a perturbation of the inflammatory environment with TB. During HIV infection, sCD14 has been associated with morbidity and mortality [4,32]. Other studies have shown an elevation of sCD14 [40] during TB infection, and increased IFN-γ production is a known feature of TB infection.

The finding of a distinguishable biomarker profile by country is intriguing. What could be contributing to these differences? There may be many factors driving the difference amongst biomarkers between these two countries. Foremost, there could be underlying genetic differences by country, specifically genetic differences in the biomarker, its receptor, or the determinants of expression, distribution or elimination. For example, P-selectin [41,42] and IL-8 [43,44] are known to have polymorphisms conferring disease risk that vary among different ethnic populations. The difference by geographic location suggests a role for viral clades influencing biomarker status. We explored the notion that HIV viral clades could have an impact on the plasma markers, and we found distinct biomarkers correlated with viraemia by country. However, these biomarkers were different when compared to biomarkers associated with country, thus making the influence of viral clade on the country's biomarker profile unclear.

Different countries may have unique endemic infectious organisms that could influence immunological responses and biomarker levels. Additionally, the microbiome could affect biomarker status, as diet and flora can vary by location. Finally, underlying behavioural and psychosocial factors may play a role as well.

Our study had several strengths including comparable populations by HIV disease stage and large numbers. The data could be generalisable to settings where patients with HIV receive routine clinical care, as the patients were accrued in clinical centres in two different regions of the world with different epidemics and distributions of opportunistic infections. Limitations include the lack of an uninfected control group to compare biomarkers by country of residence. Additionally, despite randomisation we did have baseline differences between countries requiring adjustment. We also had no data on BMI, race, smoking status or cardiovascular or other comorbidities, which could be potential confounding factors of our findings.

Our data suggest that inflammatory and coagulation biomarkers may vary significantly by region and that country-specific data may be needed in studies using biomarkers as predictors or clinical trial endpoints. Further questions remain, including whether these elevated biomarkers constitute an individual's normal range or reflect differences in HIV pathogenesis and adverse cardiovascular outcomes. In this light, if these baseline biomarkers reflect a persistent state of inflammation, this could contribute to an abnormal immune reconstitution following ART. In this regard, elevated plasma levels of inflammatory and coagulation indices both before and after ART-mediated suppression of viraemia are similarly predictive of non-AIDS morbidities and mortality after ART [45]. Overall these findings suggest that the country of residence could play a role in HIV pathogenesis, and further studies are needed to evaluate how these differences may contribute to HIV treatment, response and prognosis in diverse populations.

	Mexico (n=124)	South Africa (n=143)	Univariate P-value*	Multivariate P-value†
CRP (mg/L)	3.36[1.59–9.14]	2.93[1.17–11.13]	0.482	0.600
D-dimer (mg/L)	0.87[0.51–1.58]	1.10[0.58–1.61]	0.200	0.140
Fibrinogen (mg/dL)	723[448–1300]	1229[715–1945]	<0.001	<0.001
HA (ng/mL)	61.5[37.6–98.1]	48.1[16.9–92.3]	0.001	0.098
IFN-γ (pg/mL)	4.09[2.20–7.41]	2.19[1.17–5.58]	<0.001	0.034
IL-10 (pg/mL)	12.2[8.24–18.3]	10.4[7.63–15.7]	0.186	0.877
IL-12p70 (pg/mL)	1.06[0.42–2.72]	1.30[0.85–3.28]	0.602	0.450
IL-17 (pg/mL)	0.33[0.16–0.57]	0.31[0.18–0.59]	0.762	0.923
IL-1β (pg/mL)	0.28[0.00–0.81]	0.20[0.10–0.51]	<0.001	0.002
IL-6 (pg/mL)	2.25[1.54–3.57]	1.56[1.12–2.78]	0.080	0.445
IL-8 (pg/mL)	11.8[7.77–18.1]	5.79[3.40–8.31]	<0.001	<0.001
CXCL10 (pg/mL)	2575[439–1780]	1691[999–2693]	<0.001	<0.001
LTB4 (pg/mL)	10.2[10.2–37.2]	47.8[16.5–76.6]	<0.001	<0.001
P-Selectin (ng/mL)	48.5[37.2–64.7]	64.7[50.1–82.6]	<0.001	<0.001
Protein C (%)	3797[3267–4323]	3768[3112–4378]	0.993	0.348
Protein S (%)	3550[3037–3960	4376[3658,5430]	<0.001	<0.001
SAA (mg/L)	5.29[2.03–12.79]	5.54[1.52–17.19]	0.742	0.632
TNF-α (pg/mL)	17.3[13.5–25.4]	15.5[12.0–22.1]	0.611	0.811
Vitamin D (ng/mL)	7.72[4.44–13.4]	10.4[6.90–17.1]	0.001	0.026
sCD14 (μg/mL)	2221[1895–2686]	1957[1530–2615]	0.005	0.123
sCD163 (ng/mL)	705[471–957]	629[419–903]	0.521	0.882
sCD40 L(pg/mL)	586[227–1096]	1157[870–1716]	<0.001	<0.001
vWF (%)	9162[5861–10,880]	10,990[9259–13,770]	<0.001	<0.001

Data are reported as median[IQR] or as percentage.

Univariate analysis was determined with a generalised linear model.

Multivariate analysis was determined using a generalised linear model and was adjusted for age, CD4 cell count, haemoglobin, gender, and AIDS-defining conditions.

	Females		Males
	Mexico	South Africa	Mexico	South Africa	Univariate	Multivariate
	(n=15)	(n=77)	(n=109)	(n=66)	P-value*	P-value†
Agea (years)	38.0	38.0	35.0	38.0	0.249
	[30.0–44.0]	[31.0–44.5]	[29.0–40.5]	[34.0–43.2]
CD4 cell counta (cells/μL)	15.0	37.0	32.0	36.5	0.752
	[3.00–36.0]	[21.5–62.5]	[16.0–53.5]	[21.0–66.3]
CD8 cell counta (cells/μL)	475	465	480	476	0.337
	[351–725]	[337–673]	[362–824]	[304–718]
Plasma HIV RNAa (log10 c/mL)	5.44	5.32	5.38	5.41	0.221
	[5.23–5.65]	[4.90–5.69]	[5.04–5.72]	[5.00–5.82]
HgBa (g/dL)	11.7	11.4	12.9	12.8	<0.001
	[10.8–12.5]	[10.5–12.6]	[11.05–14.1]	[11.4–13.7]
TBb (%)	19.6 (18)		24.6 (43)		0.443
Countryb (%)	16.3 Mexico		62.3 Mexico		<0.001
CRP (mg/L)	1.83	2.30	3.57	3.67	0.086	0.117
	[0.70–2.38]	[0.88, 10.5]	[1.81–10.3]	[1.37–12.3]
D-dimer (mg/L)	0.71	0.99	0.95	1.11	0.488	0.903
	[0.45–1.23]	[0.60–1.78]	[0.52–1.59]	[0.53–1.52]
Fibrinogen (mg/dL)	670	1288	726	1009	0.011	0.533
	[403–1004]	[781–2288]	[463–1318]	[581–1730]
HA (ng/mL)	55.3	46.1	63.4	48.1	0.122	0.952
	[34.1–85.5]	[20.4–88.5]	[40.7–101]	[9.00, 104]
IFN-γ (pg/mL)	3.93	2.64	4.16	1.74	0.119	1.000
	[2.83–8.11]	[1.38–4.71]	[2.14–7.31]	[0.98–8.78]
IL-10 (pg/mL)	8.97	10.0	12.3	11.0	0.203	0.453
	[5.43–18.0]	[7.52–16.3]	[8.42–18.6]	[7.97–15.2]
IL-12p70 (pg/mL)	0.49	1.31	1.17	1.29	0.382	0.476
	[0.16–1.34]	[0.76–3.27]	[0.52–2.74]	[0.90–3.28]
IL-17 (pg/mL)	0.26	0.31	0.34	0.30	0.211	0.211
	[0.05–0.52]	[0.15–0.57]	[0.17–0.57]	[0.21–0.64]
IL-1β (pg/mL)	0.28	0.20	0.28	0.21	0.065	0.607
	[0.09–0.44]	[0.10–0.45]	[0.00–0.82]	[0.08–0.52]
IL-6 (pg/mL)	2.01	1.52	2.40	1.68	0.042	0.165
	[1.36– 2.25]	[1.13–2.67]	[1.57–3.69]	[1.12–2.86]
IL-8 (pg/mL)	10.8	5.83	11.8	5.76	<0.001	0.416
	[5.65–15.0]	[3.69–833]	[8.32–18.5]	[3.32–8.23]
CXCL10 (pg/mL)	2791	1710	2522	1565	0.012	0.755
	[1954–4130]	[999–2748]	[1703–4207]	[1001–2426]
LTB4 (pg/mL)	10.2	44.5	10.2	54.8	0.153	0.087
	[10.2–53.4]	[14.7–67.6]	[10.2–34.5]	[30.1–83.2]
P-Selectin (ng/mL)	47.4	56.8	48.7	75.8	0.631	0.079
	[34.9–61.2]	[48.1–77.7]	[37.2–65.8]	[51.1–88.2]
Protein C (%)	3984	3739	3747	3792	0.332	0.281
	[3523–4512]	[3208–4426]	[3224–4322]	[3029–4349]
Protein S (%)	3560	4286	3550	4474	0.132	0.128
	[3040–3682]	[3467–5334]	[3034–4018]	[3806–5634]
SAA (mg/L)	2.35	5.47	5.60	6.38	0.122	0.117
	[0.67–7.92]	[1.21–17.3]	[2.32–15.6]	[1.82–15.8]
TNF-α (pg/mL)	16.9	15.3	17.4	16.14	0.109	0.125
	[12.1–19.7]	[10.8–22.4]	[13.5–26.3]	[13.0– 2.0]
Vitamin D (ng/mL)	6.35	13.9	8.07	9.18	0.007	0.160
	[4.31–9.59]	[7.95–19.7]	[4.45–13.7]	[5.83–14.6]
sCD14 (μg/mL)	2263	2034	2214	1904	0.461	0.533
	[1978–2625]	[1542–2604]	[1877–2713]	[1500–2626]
sCD163 (ng/mL)	735	607	700	670	0.358	0.482
	[511–982]	[380, 895]	[462–953]	[424–956]
sCD40 L(pg/mL)	481	1165	601	1148	<0.001	0.774
	[214–983]	[902–1749]	[257–1140]	[817–1695]
vWF (%)	9181	11,139	9143	10,792	0.002	0.339
	[7442–10,564]	[9065–14,653]	[7090–10,944]	[9576–13,077]

Data are reported as median [IQR] or as percentage.

Significance of baseline characteristics was obtained using Wilcoxon analysisa or Fisher's T testb.

Univariate analysis was determined with generalised linear model to compare biomarkers by gender.

Multivariate analysis was determined with generalised linear model to compare biomarkers by gender and was adjusted for country.

	Tuberculosis			No Tuberculosis
	Mexico	South Africa		Mexico		South Africa	Univariate	Multivariate
	(n=21)	(n=40)		(n=103)		(n=103)	P-value*	P-value†
Agea	35.0	37.5		35.0		38.0	0.600
(years)	[26.5–38.5]	[34.0–42.0]		[29.0–42.0]		[32.0–45.0]
CD4 cell counta	31.0	32.0		30.0		40.0	0.760
(cells/μL)	[21.5–60.5]	[20.3–64.0]		[14.0, 52.0]		[22.0–66.0]
CD8 cell counta	423	414		526		478	0.091
(cells/μL)	[286–538]	[290–707]		[378–823]		[341–680]
Plasma HIV RNAa	5.23	5.64		5.40		5.19	0.014
(log10copies/mL)	[5.08–5.68]	[5.22–5.94]		[5.05–5.72]		[4.92–5.63]
HgBa	11.3	11.3		12.7		12.1	<0.001
(g/dL)	[9.75–12.9]	[10.2–13.1]		[11.2–14.1]		[11.1–13.0]
Genderb (%)	30 Female		36 Female		0.443
Countryb (%)	34.4 Mexico		50 Mexico		0.0404
CRP (mg/L)	6.00	7.37		2.89		2.21	0.006	0.026
	[2.50–28.0]	[2.09–19.5]		[1.37–8.16]		[1.12–9.20]
D-dimer (mg/L)	1.26	1.48		0.83		0.96	0.013	0.223
	[0.66–2.15]	[0.77–2.18]		[0.50–1.55]		[0.54–1.56]
Fibrinogen (mg/dL)	575	1449		743		1150	0.190	0.511
	[477–1667]	[757–2324]		[446–1291]		[707–1876]
HA (ng/mL)	120	60.2		54.3		46.1	0.002	0.017
	[65.9–223]	[20.8–145]		[34.1–87.8]		[10.5–91.4]
IFN-γ (pg/mL)	8.11	5.64		3.22		1.74	<0.001	<0.001
	[4.05–17.6]	[2.59–11.5]		[2.06–6.23]		[0.91–3.80]
IL-10 (pg/mL)	12.6	11.0		12.0		10.3	0.846	0.672
	[7.07–15.9]	[8.11–16.7]		[8.27–18.4]		[7.39–14.6]
IL-12p70 (pg/mL)	0.80	1.23		1.10		1.30	0.300	0.408
	[0.32–2.34]	[0.90–3.27]		[0.41–2.74]		[0.77–3.37]
IL-17 (pg/mL)	0.34	0.29		0.31		0.31	0.610	0.410
	[0.16–0.54]	[0.16–0.38]		[0.15–0.59]		[0.19–0.64]
IL-1β (pg/mL)	0.26	0.20		0.28		0.21	0.168	0.179
	[0–1.18]	[0.08–0.42]		[0–0.72]		[0.10–0.55]
IL-6 (pg/mL)	3.54	1.65		2.22		1.47	0.247	0.752
	[1.60–5.73]	[1.19–3.20]		[1.40–3.23]		[1.11–2.68]
IL-8 (pg/mL)	13.08	7.08		11.6		5.44	0.987	0.977
	[9.04–17.8]	[3.35–9.65]		[7.37–18.5]		[3.50–8.27]
CXCL10 (pg/mL)	2917	2156		2480		1515	0.032	0.066
	[2068–4739]	[1527–3462]		[1703–4152]		[947–2292]
LTB4 (pg/mL)	13.9	52.2		10.2		45.2	0.591	0.584
	[10.2–44.6]	[10.2–78.1]		[8.1–40.0]		[18.0–76.6]
P-Selectin (ng/mL)	34.9	68.9		51.4		62.2	0.120	0.053
	[27.7–47.6]	[42.2–82.8]		[40.3–66.4]		[50.4–82.6]
Protein C (%)	3494	3803		3849		3768	0.195	0.256
	[3080–4058]	[3147–4482]		[3323–4412]		[3088–4312]
Protein S (%)	3420	4421		3561		4329	0.424	0.910
	[2764–4015]	[3683–5744]		[3045–3948]		[3574–5299]
SAA (mg/L)	10.6	10.1		4.70		5.35	0.825	0.858
	[2.27–57.3]	[0.86–21.8]		[1.88–10.3]		[1.52–14.1]
TNF-α (pg/mL)	22.0	19.7		16.7		15.2	0.115	0.465
	[14.6–29.0]	[14.3–26.0]		[13.3–23.8]		[11.3–20.7]
Vitamin D (ng/mL)	7.55	10.4		7.96		10.6	0.534	0.818
	[5.00–12.7]	[6.10–19.9]		[4.28–13.7]		[7.29–16.8]
sCD14 (μg/mL)	3452	2906		2096		1800	<0.001	<0.001
	[2576–3882]	[2247–3305]		[1784–2407]		[1424–2174]
sCD163 (ng/mL)	713	840		704		607	0.451	0.526
	[360–953]	[425–1133]		[480–980]		[381–851]
sCD40 L(pg/mL)	429	1124		604		1229	0.256	0.030
	[109–772]	[609–1695]		[263–1141]		[906–1744]
vWF (%)	9575	12,491		9057		10,668	0.012	0.022
	[8169–11,324]	[10,112–14,793]		[7023–10,790]		[8776–13,317]

Data are reported as median [IQR] or as percentage.

Significance of baseline characteristics was obtained using Wilcoxon analysisa or Fishers T testb.

Univariate analysis was determined with generalised linear model to compare biomarkers by TB status.

Multivariate analysis was determined with generalised linear model to compare biomarkers by TB status and was adjusted for country, haemoglobin and viral load.