Background

Progressive HIV infection [1, 2] leads to loss of body fat, altering subjects' appearance and resulting in social stigma and negative body image, particularly in women [3]. Caloric intake and adiposity are regulated by adipokines, neurotransmitters, hormones and their receptors that modulate appetite, fat metabolism and energy homeostasis [4]. Leptin is produced by adipocytes, and contributes to appetite reduction via a negative hypothalamic feedback [5], as well as exerting immunomodulatory activity (e.g., inhibition of cell proliferation and promotion of type 1 adaptive responses [6–8]).

A correlation between adipose tissue and leptin is maintained in antiretroviral-treated HIV-infected individuals, independent of viremic status or disease stage [9–12], even though leptin receptor expression and phosphorylation are increased in peripheral blood mononuclear cells [13]. A number of factors (chronic fever, increased TNF-α and IL-6 levels, opportunistic infections and other secondary causes [1, 14, 15]) have been considered to contribute to fat loss in advanced disease; however, a direct link to viral replication has not been established.

Women account for more than 50% of HIV-1 prevalence worldwide [16]. Here, we explore the relationship between viral load, cellular activation, innate immunity effector levels and adipose tissue-related measures in a cohort of antiretroviral therapy (ART)-naïve South African women.

Methods

Study subjects

A cross-sectional convenience sample of 83 ART-naïve, HIV-1 infected women without evidence of prior or ongoing opportunistic infection was enrolled at the Clinical Research HIV-1 Unit, Themba Lethu Clinic (University of the Witwatersrand, Johannesburg, South Africa). Screening CD4+ T cell count was 200-350 cells/mm3. Medical history was obtained from the clinic record and by interview. Informed consent was obtained from all participants as per University of the Witwatersrand Ethics Committee and Wistar Institute IRB-approved study protocol.

Flow cytometry

CD4+ T cell counts were assessed using the single platform method described by Scott and Glencross [17]. Expression of lineage, differentiation and activation markers on CD3+ T cells (CD4, CD8, CD38, CD95, HLA-DR, CD28), CD3- NK cells (CD16, CD56, HLA-DR) and Lin-1- Dendritic cells (Lin-1, HLA-DR, CD123, CD11c) was assessed on whole peripheral blood using lyophilized mAb panels (BD Biosciences, Palo Alto, CA). Samples were tested using a Faxcalibur Analyzer, followed by analysis using CellQuest software (BD Biosciences).

Adipose tissue measurements

Body mass index (BMI) was calculated as weight (kg)/height (m)2. Dual Energy X-ray Absorptiometry (DEXA) scans were performed using a Hologic QDR-2000 scanner, assessing limb and trunk fat and lean mass. Bone mineral density (g/cm2) was also measured.

Magnetic resonance imaging (MRI) scans were performed using a Toshiba Flexart 0.5 T; a single L4-L5 axial section was analyzed. Variables collected were sagittal diameter, visceral, subcutaneous abdominal and perirenal fat. The analysis was conducted using V3.51*R553 software.

Clinical laboratory testing

Serum from fasting blood draws was tested for:

  • Leptin: ELISA, BioVendor Laboratory Medicine, Inc, Czech Republic

  • High density lipoprotein (HDL)- and low density lipoprotein (LDL)-associated cholesterol, triglycerides, glucose: Roche Integra analyzer 400, Roche Diagnostics, Mannheim, Germany

  • Insulin: Immulite 1000 analyzer Diagnostics Corp, Los Angeles. CA

  • Proinsulin: ELISA, Dako-Cytomation Ltd (UK)

  • Free fatty acids (FFA): half-micro test, Roche Diagnostics, Mannheim, Germany

  • HIV-1 infection (confirmation): rapid antibody testing and/or Ultra-sensitive (US) PCR, (Roche COBAS Ampliprep/COBAS Amplicor v1.5 methods)

  • Homeostatic Model Assessment for insulin resistance (HOMA2-IR) was determined using the HOMA2 calculator, v. 2.2, from the Diabetes Trials Unit, University of Oxford, GB http://www.dtu.ox.ac.uk/homa.

Statistical analysis

Study subjects were divided into two groups: Group 1 (low viral load, LVL): 41 subjects, viral load (VL) < sample median; and Group 2 (high viral load, HVL): 42 subjects, VL ≥ sample median. Differences in means between groups for each of the variables listed in Table 1 were tested using two-sample t-tests. Corresponding p-values and q-values based on the positive false discovery rate (FDR) are reported; q-values of less than 0.2 were considered meaningful. Pairwise correlations were tested by determining the Pearson correlation coefficient (r).

Table 1 Expression of metabolic, anthropometric and immunologic markers in high and low viral load subjects
Full size table

The effect of selected variables on log10VL (univariate analysis) was assessed by fitting a linear model using the least square method. Additionally, a multivariable linear model was fitted to log10VL using a step-wise approach based on likelihood ratio tests [18]. For the final model, Wald test statistics with p-values < 0.1 were considered relevant. Goodness-of-fit was evaluated based on the adjusted coefficient of determination (R2): briefly, variables were added individually to the model, and those variables that did not lead to an increase of R2 were rejected. All statistical tests were performed using r vers. 2.10.0 ("the r Foundation for Statistical Computing", http://www.r-project.org/).

Results

Cohort clinical and immunological characterization

Eighty-three HIV-1-infected, ART-naïve women, aged 19 to 55 (mean age 34.5 years, see Table 1), and of black (n = 81) or mixed race (n = 2) were studied. The cohort median serum HIV RNA (VL) was 60,400 copies of HIV RNA/ml (log10VL = 4.78; IQR = 1.21). Based on BMI, our cohort was composed of three underweight (BMI < 18.5 kg/m2), 33 normal weight (BMI 18.5-24.9 kg/m2), 26 overweight subjects (BMI 25-29.9 kg/m2) and 20 obese subjects (BMI > 30 kg/m2), in keeping with prior reports on similar South African cohorts [19]. The cohort mean CD4+ T cell count was 260 cells/mm3, indicative of moderately advanced disease. Due to enrolment restrictions of 200-350 CD4+ T cell count/mm3, entry CD4+ T cell counts were only slightly lower in HVL (Table 1) than LVL subjects. However, confirming previous reports [20, 21], cellular activation was higher in HVL subjects, as assessed by the expression of CD38 on CD4+ and CD8+ T cells, and of HLA-DR on CD4+ T cells (Table 1). Likewise, we did not detect significant differences in the levels of mature or activated NK cells, or of plasmacytoid (PDC) or myeloid (MDC) dendritic cells.

Association of viral load with adipose tissue

HVL women had lower BMI and waist circumference than LVL women (Table 1). Direct MRI assessments of abdominal fat content confirmed that HVL subjects had considerably less subcutaneous abdominal fat than LVL women, but similar amounts of visceral and perirenal fat (Table 1). HVL women also had slightly lower DEXA scan-based fat and lean mass measurements than LVL women, but only trunk fat mass reached the level of significance, confirming that the difference in adipose tissue between HVL and LVL women is due to differential representation of central fat. Bone density was similar in the two groups (not shown).

We tested a number of markers associated with fat and glucose regulation (Table 1); of all the indicators assessed, only leptin levels were significantly lower in HVL subjects.

As expected, leptin levels were directly correlated with BMI (r = 0.6991; p < 0.0001), MRI-measured subcutaneous (Figure 1) or visceral abdominal fat (r = 0.7755 and 0.5417, respectively, both p < 0.0001) or DEXA-based total fat mass, (r = 0.7637; p < 0.0001)].

Figure 1
figure 1_486

Association of serum leptin levels with subcutaneous fat. Linear regression modelling of the association between observed serum leptin levels and MRI-assessed subcutaneous fat area. Circles represent individual observations. P < 0.0001; adjusted R2 = 0.5983.

Full size image

Unlike direct measures of adiposity, serum lipids and glucose were similar in HVL and LVL women.

Visceral fat was positively associated with levels of insulin (r = 0.3861, p = 0.0010), C-peptide (r = 0.5331, p < 0.0001) and HOMA2-IR (r = 0.2774, p = 0.0241), but not proinsulin/insulin ratio; similar results were obtained for subcutaneous abdominal fat. Reported recent weight loss rates and free fatty acids (FFA) levels were similar in both groups (Table 1), and FFA showed no association with the amount of adipose tissue (r = 0.1084, p = 0.3868).

Modelling of leptin levels as a predictor of viral load, independent of fat accumulation

To determine the relationship between metabolic parameters and viral replication, we first assessed the relationship of all individual variables (see Table 1) with log10 VL by fitting a linear model using all available data points (i.e., no censoring of subjects missing individual variable measurements). As illustrated in Table 2 and Figure 2, of the 35 variables tested, only eight had a significant (p < 0.05) effect on log10VL based on univariate analysis. As expected, the direction of this effect was positive for parameters associated with activation (CD4+ T cells expressing CD38 or HLA-DR, CD8+ T cells expressing CD38), indicating that subjects with higher viral load also have higher cellular activation. Conversely, a negative effect was observed for PDC counts, mature NK cell frequency, as well as BMI, subcutaneous abdominal fat and leptin (Figure 2), supporting the observations of differences between HVL and LVL women, as we have described. Importantly, the negative association between leptin serum levels and log10VL was maintained (n = 65; effect estimate = -0.0186653; p = 0.0289), even after adjusting for subcutaneous abdominal fat area in multivariate analysis, supporting a direct association between leptin levels and viral replication, independent of the amount of adipose tissue.

Table 2 Variable association with log10VL, univariate analysis*
Full size table
Figure 2
figure 2_486

Association of metabolic and immunologic parameters with HIV viral load. Linear regression modelling of the association between log10 serum viral load and significantly associated variables; individual p values and adjusted R2 for the associations are indicated. Circles represent individual observations.

Full size image

The eight variables with univariate association to log10VL were further tested in a subset of subjects with complete datasets (no missing variable measurements; n = 45). Based on their ability to improve the model predictivity (as assessed by testing the model likelihood ratio), five significant variables were selected to build the additive model: the estimate terms for the model and corresponding test statistics are provided in Table 3. Leptin levels, mature NK frequency and PDC count have a negative effect on log10VL that were significant at the 10% level, whereas the positive effect of the expression of CD38 on CD8+ T and HLA-DR on CD4+ T cells was not significant. Taken together, our analysis indicates that leptin levels, together with mature NK and PDC frequency, remain negatively associated with log10VL after adjustment for multiple metabolic and activation parameters, suggesting an independent association.

Table 3 Effect* of selected variables on viral load: multivariate analysis
Full size table

Discussion

We show for the first time that leptin levels are associated with viral load after adjusting for fat measurements. Body fat changes in HIV-infected individuals have been the subject of a number of studies, many of which have focused on ART-associated lipodistrophy [22–26]. Yet to our knowledge, no study has directly sought to determine the relationship between fat, leptin and HIV replication. Importantly, since opportunistic infections (OIs), in association with lower CD4+ T cell count, might also contribute to low adiposity in chronic HIV infection (via LPS-induced TNF, limited food intake, malabsorption, etc.), we chose to study OI-free, ART-naïve women within a narrow CD4+ T cell count range (200-350 CD4+ T cell count/mm3 at screening); however, a contribution of prior or subclinical OIs and other potential confounding factors cannot be categorically excluded.

The narrowness of the cohort's CD4+ T cell count range (which limits the confounding effects of this variable) might explain the observed lack of a significant effect of CD4+ T cell count levels on VL, which has been consistently reported in larger, unrestricted cohorts [27]. However, as expected, in our cohort, viral load was positively associated with immune activation (CD38 and HLA-DR expression) and negatively associated with the frequency of innate immunity effectors (PDC and mature NK cells) in peripheral blood, confirming prior observations [28–32].

HVL women showed lower BMI, waist circumference and subcutaneous abdominal fat than LVL women, with a significant negative association between VL and several measures of central fat. Leptin levels observed in LVL women were similar to those observed in a cohort of 50 healthy women with similar age, ethnicity and provenance (median BMI 27 kg/m2, IQR 10.85; median serum leptin 36.15 pg/ml, IQR = 38.95, N Crowther, unpublished results).

As predicted by the fact that leptin is mainly produced by subcutaneous fat adipocytes [33], this adipokine was also lower in HVL women, and its levels are inversely correlated to viral replication. Interestingly, however, HOMA2-IR, another marker usually highly correlated with adiposity, presented no association with VL in our cohort, and was only weakly associated with leptin levels (p = 0.0751). Taken together, these results indicate that: (a) adipose tissue is associated with both leptin levels and insulin resistance; and (b) only leptin levels are inversely associated with viral replication, suggesting the hypothesis that the inverse relationship between viral replication and leptin levels may be independent of the amount of adipose tissue. Formal testing of this hypothesis in a multivariate model demonstrated that the effect of leptin is in fact independent of direct measures of fat (e.g., subcutaneous fat area).

With the exception of leptin, serum molecules associated with adipose tissue and obesity (e.g., total or LDL cholesterol, triglycerides, glucose, insulin) did not independently correlate with VL in our cohort, in contrast with prior reports that VL could predict BMI in HIV-infected women [2], and correlate negatively with LDL and HDL cholesterol and positively with triglycerides, but not with insulin or glucose levels [34, 35]. Indeed, we confirmed the selection of leptin over any other adipose tissue measure that appeared to be individually correlated with log10VL by introducing BMI or subcutaneous abdominal fat in a multivariate model with leptin: neither carried a significant independent association (p = 0.3667 and 0.884, respectively), and both actually resulted in making our model less accurate by reducing the adjusted R2 (0.063 and 0.05, as compared to 0.066 for leptin alone).

Therefore, among measures and correlates of adipose tissue, leptin remained the variable best associated with viral replication, suggesting the possibility that leptin may play a role in the observed significant association between fat and viral replication. Interestingly, prior studies (FRAM cohort [36]) testing whether viral replication was associated with serum adipokines did not evidence a significant association between adiponectin and leptin levels and HIV viral load: the fact that our cohort is composed only of ART-naïve women with a narrow CD4 range, and possibly the high levels of adiposity in our cohort, might contribute to this discordance.

Based on leptin's known immunomodulatory activity [6–8, 13, 37, 38], it is interesting to speculate that in HVL women, lower leptin levels may contribute to chronic immune activation, in keeping with the observed increased expression of CD38 and HLA-DR in T cell subsets. Obesity is associated with chronic low-level inflammation and high levels of TNF-α [39], which is produced by subcutaneous adipocytes [40]. This condition would be expected to promote viral replication since TNF-α promotes HIV replication via NFκB activation [41].

The negative association between leptin and viral replication that we report here suggests that leptin, which is produced by adipocytes in response to exposure to TNF-α[42], may be part of a negative regulatory feedback that attenuates the pro-inflammatory and pro-replicative effects of TNF-α. Interestingly, this effect is likely lost in chronic inflammatory conditions where a negative association between TNF-α and leptin production has been observed [43, 44]. Conversely, increased body fat may attenuate viral load via the effects of other mediators (e.g., MIP-1α) known to suppress HIV infection [45]). Another potential mechanism could be that high viral replication and cellular activation may result in chronic inflammation, affecting adipocytes and causing lipoatrophy and lower leptin levels.

The interpretation of these results should be qualified in light of the cross-sectional design of the study, which does not allow the interpretation of cause-effect relationships in the variables studied. Further longitudinal studies focusing on these factors will be required to determine whether fat changes directly contribute to alterations in viral replication via adipocyte-mediated leptin secretion.

Conclusions

Our data provide the first demonstration of a relationship between VL and leptin in African women and suggest that lower leptin levels associated with the loss of adipose tissue may contribute to disease progression.