A Glucuronoxylomannan-Associated Immune Signature, Characterized by Monocyte Deactivation and an Increased Interleukin 10 Level, Is a Predictor of Death in Cryptococcal Meningitis

Background. Cryptococcal meningitis remains a significant cause of death among human immunodeficiency virus type 1 (HIV)–infected persons in Africa. We aimed to better understand the pathogenesis and identify immune correlates of mortality, particularly the role of monocyte activation. Methods. A prospective cohort study was conducted in Cape Town, South Africa. Patients with a first episode of cryptococcal meningitis were enrolled, and their immune responses were assessed in unstimulated and stimulated blood specimens, using flow cytometry and cytokine analysis. Results. Sixty participants were enrolled (median CD4+ T-cell count, 34 cells/µL). Mortality was 23% (14 of 60 participants) at 14 days and 39% (22 of 57) at 12 weeks. Nonsurvivors were more likely to have an altered consciousness and higher cerebrospinal fluid fungal burden at presentation. Principal component analysis identified an immune signature associated with early mortality, characterized by monocyte deactivation (reduced HLA-DR expression and tumor necrosis factor α response to lipopolysaccharide); increased serum interleukin 6, CXCL10, and interleukin 10 levels; increased neutrophil counts; and decreased T-helper cell type 1 responses. This immune signature remained an independent predictor of early mortality after adjustment for consciousness level and fungal burden and was associated with higher serum titers of cryptococcal glucuronoxylomannan. Conclusions. Cryptococcal-related mortality is associated with monocyte deactivation and an antiinflammatory blood immune signature, possibly due to Cryptococcus modulation of the host immune response. Validation in other cohorts is required.

Cryptococcal meningitis, caused by the encapsulated yeast Cryptococcus neoformans, is the commonest form of adult meningitis in sub-Saharan Africa and remains a significant cause of death among human immunodeficiency virus type 1 (HIV)infected individuals in this region [1][2][3]. Despite availability of amphotericin B and antiretroviral therapy (ART), mortality remains high, ranging between 17% and 32% at 14 days and between 34% and 59% at 12 weeks [4,5]. Reduced consciousness level, high fungal burden, and slower yeast clearance from the cerebrospinal fluid (CSF) have been identified as independent predictors for mortality [4]. However, some patients with a normal consciousness level at the time of diagnosis may deteriorate and die despite optimal treatment. Secondary infections such as bacterial sepsis or tuberculosis can also contribute [4,6].
The host immune response is central to the pathogenesis of cryptococcosis. The vast majority of cryptococcal infections occur among individuals with impaired cell-mediated immunity, and the marked susceptibility of persons with AIDS demonstrates the importance of CD4 + T lymphocytes in protection [7]. However, low CD4 + T-cell counts have not consistently been identified as a risk factor for mortality among persons with AIDS-related cryptococcosis, nor has the number of circulating Cryptococcus-specific CD4 + T lymphocytes [4,8]. Animal models of cryptococcosis demonstrate the microbiological and survival benefits of a T-helper type 1 (Th1) response, characterized by interferon γ (IFN-γ) production and classical macrophage activation, along with the detrimental effects of a Th2 response, characterized by interleukin 4 (IL-4) and interleukin 13 (IL-13) production with alternative macrophage activation [9][10][11]. Studies in humans have not demonstrated any association between a Th2 response and poor outcome but have observed a significantly higher proportion of circulating cryptococcalspecific CD4 + T cells that produce both IFN-γ and tumor necrosis factor α (TNF-α) in survivors [8]. This is consistent with previous studies that have observed higher CSF concentrations of such cytokines in survivors and significantly improved CSF fungal clearance when adjunctive IFN-γ was given alongside antifungal therapy [12,13]. Despite their importance in the murine immune response, the relationship between the activation state of macrophages and their blood precursor monocytes with clinical outcome has not been studied in persons with HIV-associated cryptococcosis. To address this, we examined immune responses in blood, using ex vivo stimulation assays, flow cytometry, and cytokine analysis. We hypothesized that a less activated monocyte phenotype would be associated with mortality.

Study Design
A prospective cohort study was conducted between April 2012 and January 2014 at GF Jooste, Khayelitsha, and Mitchell's Plain Hospitals, Cape Town, South Africa. Consecutive patients aged ≥18 years with a first episode of HIV-associated cryptococcal meningitis were enrolled within 48 hours of starting antifungal therapy. At enrollment, CSF samples were collected and quantitative culture performed to quantify fungal burden, as previously described [14]. A peripheral blood specimen was obtained to assess the immune response. Antifungal therapy comprised intravenous amphotericin B (1 mg/kg/day) and oral fluconazole 800 mg/day for 14 days, followed by fluconazole 400 mg/day for a further 10 weeks and then maintenance fluconazole. Additional lumbar punctures with therapeutic CSF drainage were performed at attending clinicians' discretion to control raised intracranial pressure. ART was initiated 4 weeks after meningitis diagnosis according to national guidelines, unless patients were receiving it at enrollment [15]. The primary end point was all-cause mortality at 14 days; the secondary end point was mortality at 12 weeks. All participants provided written informed consent; surrogate consent was sought from the next of kin for patients with impaired consciousness. Ethical approval was obtained from the research ethics committees of the University of Cape Town (reference numbers 408/2010 and 371/2013) and Liverpool School of Tropical Medicine (reference number 11.92).

Whole-Blood Flow Cytometry
Flow cytometry was performed on unstimulated blood specimens from 56 participants. Fresh whole-blood specimens were stained with commercially available conjugated monoclonal antibodies to assess phenotype and activation of both T-cell and monocyte subsets. Variables measured included the relative frequency of circulating neutrophils, monocytes, CD4 + T cells, CD8 + T cells, and CD4 − CD8 − T cells (recorded as the percentage of white blood cells [WBCs]); the proportion of CD16 − neutrophils (recorded as the percentage of total neutrophils); the proportion of classical (CD14 ++ CD16 − ), intermediate (CD14 ++ , CD16 + ), and nonclassical (CD14 + CD16 ++ ) monocytes; the expression of HLA-DR, CD38, and PD-1 on CD4 + , CD8 + , and CD4 − CD8 − T cells; and the expression of HLA-DR, CD163, PD-1, CCR2, CCR5, CD80, and Toll-like receptor 4 (TLR4) on monocytes (determined for the whole population and for the 3 monocyte subsets described above). Further details of the antibodies used are given in the Supplementary Methods; the gating strategy is illustrated in Figure 1A.

Antigen-Stimulation Assay
Fresh whole-blood specimens were stimulated for 6 and 24 hours at 37°C with lipopolysaccharide (LPS; a TLR4 agonist; Invivogen, San Diego, California); the imidazoquinoline compound R848 (a TLR7/8 agonist; Invivogen); Cryptococcus capsule glucuronoxylomannan (GXM); heat-killed, mechanically disrupted C. neoformans H99; or an equivalent volume of phosphate-buffered saline (unstimulated). Supernatants from wholeblood specimens stimulated for 24 hours were removed and stored at −80°C for later cytokine analysis (see below). Cytokine production was assessed in whole-blood specimens stimulated for 6 hours, using intracellular cytokine staining. Variables measured for each of the 4 stimulations included relative frequency of neutrophils and monocytes (recorded as the percentage of WBCs); neutrophil expression of CD66 (recorded as the mean fluorescence intensity [MFI]) and CD16 (recorded as the percentage of cells); and expression of HLA-DR (recorded as the MFI) and production of interleukin 6 (IL-6), interleukin 10 (IL-10), interleukin 12 (IL-12), and TNF-α by circulating monocytes (recorded as the percentage of monocytes positive for cytokine). HLA-DR expression was recorded as absolute MFI value and relative expression compared to unstimulated. Results from antigen-stimulation assays were available for 41-48 participants (depending on stimulation). Owing to autofluorescence of the cryptococcal preparations, IL-12 and IL-10 expression could not be measured accurately for cryptococcal stimulations. Further details regarding the preparation of GXM and the antibodies and fluorochromes used are given in the Supplementary Methods; the gating strategy is illustrated in Figure 1B.

Data Analysis
Baseline characteristics were summarized using proportions or median values with interquartile ranges (IQRs). Characteristics between survivors and nonsurvivors were compared using the Wilcoxon rank sum test, t test, or Fisher exact test, as appropriate. Data from flow cytometry and biomarker analysis on unstimulated and stimulated blood were combined and analyzed using principal component analysis (PCA) and unsupervised hierarchical cluster analysis as described elsewhere [16]. Before PCA, skewed variables were log 2 transformed and scaled, such that the geometric mean equaled 0 and variance equaled 1. Variables were filtered using statistical tests prior to incorporation into PCA plots and cluster analysis, such that only variables with a statistically significant association with the dependent variable (14-day mortality) were used. Statistical significance was defined as a P value of <.05 and a Q value of <.1 (equivalent to a 10% false-discovery rate [FDR], using the Benjamini-Hochberg procedure for multiple-testing correction [17]). For initial PCA, missing values were imputed using the K nearestneighbors technique [18]; subsequent logistic regression analysis was restricted to 37 participants who had all 205 baseline variables measured. Analysis was performed using Stata, version 12. (Stata, College Station, Texas), and Qlucore Omics Explorer, version 3.0 (Qlucore, Lund, Sweden).

Differences in Blood Immune Response Between Subjects Who Died and Those Who Survived
A clear difference in peripheral blood immune response was noted between nonsurvivors and survivors. PCA showed distinct clustering of participants according to day 14 outcome, with persons who died having significantly higher values for principal component 1 (PC1; Figure 2A). Twenty-three variables contributed to PC1, and levels of all differed significantly between survivors and nonsurvivors (P < .05 and Q < 0.1), resulting in a distinct immune signature ( Figure 2B and Table 2). This was characterized by an increased proportion of circulating neutrophils; higher serum concentrations of IL-6, IL-10, and CXCL10; decreased expression of HLA-DR on circulating monocytes (both resting and stimulated with LPS or R848); a decreased proportion of monocytes producing TNF-α when stimulated with LPS; and reduced concentrations of IL-12 and IFN-γ when whole blood was stimulated with LPS or R848 ( Table 2). Although HLA-DR ratios (stimulated/unstimulated) were significantly higher in nonsurvivors, this was due to low expression by unstimulated cells. Unsupervised hierarchical clustering using the same 23 variables also demonstrated clear grouping of participants by day 14 outcome (Figure 2C). A similar immune signature was identified in participants who died by week 12, but this was not significant when adjusted for multiple comparisons (data not shown).

Predictive Ability of Blood Immune Signature
To further explore the association between this immune signature and poor outcome, the first principal component from the analysis (PC1) was added as a summative variable to a logistic regression model predicting death by day 14. Analysis was restricted to 37 individuals for whom full immunological data were available (28 survivors and 9 nonsurvivors  or excluding the 3 noncryptococcal deaths did not significantly alter these findings (Supplementary Tables 1 and 2). Adding PC1 or limited PC1 to the base model also resulted in increased area under the curve on receiver operating characteristic curve analysis, but this did not reach statistical significance (P = .107 and .074, respectively; Supplementary Figure 1).

Reasons for an Antiinflammatory Response in Nonsurvivors
We hypothesized that the immune phenotype observed in nonsurvivors was a compensatory downregulation of cellular responses following a period of immune activation. Possible causes of this immune activation included disseminated cryptococcal infection, HIV infection, or occult bacterial sepsis. To explore the first 2 possibilities, associations between PC1 scores and parameters approximating these factors were examined. A statistically significant positive correlation was noted between limited PC1 scores and serum GXM titers (Spearman r = 0.28; P = .041; Figure 3A) but not with plasma HIV load (Spearman rho = 0.08; P = .561). When the relationship between GXM and individual components of limited PC1 were examined, a positive correlation was noted with the percentage of neutrophils in blood (Spearman rho = 0.33; P = .010), serum CXCL10 concentrations (Spearman rho = 0.31; P = .015), and serum IL-10 concentrations (Spearman rho = 0.39; P = .002; Figure 3B).

DISCUSSION
In this prospective cohort study, we examined clinical and immune correlates of fatal outcome in people with HIV-associated cryptococcal meningitis. Similar to other cohorts, nonsurvivors were significantly more likely to have a reduced consciousness level at presentation, a higher CSF fungal burden, and increased numbers of circulating blood leukocytes [8]. PCA and unsupervised hierarchical clustering analysis demonstrated clear differences in the baseline blood immune response between subjects who survived or died by day 14, with nonsurvivors having significantly higher scores for PC1. A breakdown of the variables contributing to PC1 revealed an immune signature associated with mortality, characterized by a higher proportion of circulating neutrophils, increased levels of soluble markers of immune activation (IL-6 and CXCL10), higher concentrations of antiinflammatory mediators (IL-10), reduced Th1 cytokine (IL-12 and IFN-γ) production in supernatant following incubation of whole blood with TLR4 or TLR7/8 agonists, decreased HLA-DR expression on circulating monocytes (both stimulated and unstimulated), and decreased monocyte production of TNF-α following LPS stimulation. This immune signature remained an independent determinant of 14-day mortality even when  adjusted for baseline fungal burden, consciousness level, CD4 + T-cell count, ART status, and amphotericin B status, and even if noncryptococcal deaths were excluded. The combination of decreased monocyte expression of HLA-DR and reduced monocyte TNF-α response to LPS has been referred to by some authors as "monocyte anergy" [19][20][21]. It has previously been described in bacterial sepsis, acute pancreatitis, and severe trauma, in which it is frequently associated with a worse clinical outcome, such as increased frequency of nosocomial infections and increased mortality [19,[22][23][24][25]. Monocyte anergy is one component of a more widespread downregulation of the immune response known as the compensatory antiinflammatory response syndrome (CARS), which typically occurs following a severe proinflammatory insult. Additional features of CARS include increased lymphocyte apoptosis, cutaneous anergy, reduced T-cell proliferation and proinflammatory cytokine (IL-2 and TNF-α) responses, and increased concentrations of IL-10 [21,[25][26][27]. Increased plasma IL-10 levels were noted among nonsurvivors in our study and may explain the observed alterations in ex vivo cellular immune responses. In vitro studies have shown that IL-10 directly inhibits proinflammatory cytokine production and antigen presentation by macrophages, resulting in additional impairment of CD4 + T-cell production of IFN-γ [28,29]. Murine studies have also shown that significantly impaired T-cell proliferation and decreased IFN-γ production occur when monocyte anergy is induced through preexposure to LPS, an IL-10-dependent phenomenon known as endotoxin tolerance [30][31][32]. Although T-cell responses were not examined specifically in this study, subjects who died by day 14 had decreased production of IL-12 and IFN-γ when stimulated with TLR4 or TLR7/8 agonists in whole-blood assays, suggesting a decreased ability to mount a Th1 response. IFN-γ is widely recognized to be a central component of the protective Th1 host response against C. neoformans. Low CSF concentrations of IFN-γ and a blood CD4 + T-cell phenotype characterized by a lack of IFN-γ and TNF-α production have been observed in HIV-infected individuals who died from cryptococcal meningitis [8,12]. Furthermore, IFN-γ therapy has been shown to increase the rate of fungal clearance when added to amphotericin B and flucytosine in a clinical trial [13]. IL-12 is produced by monocytes, macrophages, and dendritic cells and acts during antigen presentation to promote IFN-γ production by natural killer and CD4 T cells [30,[33][34][35]. The low levels of IL-12 production following whole-blood stimulation along with decreased expression of the major histocompatibility complex class II molecule HLA-DR on deactivated (or "anergic") monocytes observed in this study suggest that impaired antigen presentation may be an explanation for the impaired Th1 responses previously observed in nonsurvivors. In addition to impairing anticryptococcal responses, such an immune phenotype may also predispose to further infections with bacterial pathogens, as observed in patients with severe sepsis [27].
There are a number of possible explanations why a CARS-like immune response was observed among HIV-infected individuals who died from cryptococcosis. The significant correlations between serum titers of GXM (the major component of the cryptococcal capsule), PC1, and serum IL-10 provide supportive evidence that this immune phenotype may have been driven by the cryptococcal organism and its antigens. This hypothesis is further supported by in vitro studies showing that GXM provokes a proinflammatory cytokine response in neutrophils (IL-1β, IL-6, IL-8, and TNF-α) but leads to an antiinflammatory response in monocytes ( production of IL-10 and impaired antigen presentation) [36][37][38]. Given the massive tissue burden of cryptococci observed in HIV-associated cryptococcosis [39], it is plausible that the immune phenotype observed among nonsurvivors in this study was a direct effect of GXM, possibly mediated by the actions of IL-10. This is supported by in vitro work demonstrating that the addition of GXM to monocytes in cell culture results in impaired antigen presentation and reduced T-cell responses via an IL-10-dependent mechanism [38].
An alternative explanation is that the immune signature represents immune exhaustion due to persistent HIV-associated immune activation, an increasingly recognized component of HIV pathogenesis [40,41]. IL-10 levels are known to increase as HIV infection progresses, and high levels of serum IL-6 (along with C-reactive protein) have been repeatedly associated with fatal outcome [42][43][44][45]. Possible mechanisms for this include translocation of bacterial products through the intestinal wall [46,47] or direct effects of HIV on the host immune response, possibly through direct stimulation of monocytes by HIV proteins [48]. However, there was no significant association noted between PC1 and HIV load, and sCD14-a marker of monocyte activation following stimulation with LPS and frequently used as a surrogate marker of gut translocation [47]did not form part of the immune signature identified among nonsurvivors.
Finally, the observed immune phenotype could conceivably be due to coexistent but clinically unrecognized bacterial infection. Although no blood cultures performed at study enrollment identified any bacterial pathogens, the widespread use of ceftriaxone on admission to the hospital will have significantly reduced the sensitivity of this.
There are a number of limitations to this study. It was exploratory in nature with a relatively small sample size and thus was underpowered for certain analyses. As the cohort was not split into training and testing sets, the predictive ability of this immune signature remains unvalidated. It is therefore important to repeat these observations in independent cohorts to assess reproducibility. The study also suffered from a degree of missing data, as an ambitious number of immune assays were performed, the majority of them in real time on fresh samples of peripheral blood. However, this did not appear to alter the final results. The immune signature remained significantly associated with 14-day mortality either when restricted to the 37 persons with a full set of observations, when a limited signature was applied to 55 persons, or when an accepted method of imputing missing data was used for the whole data set.
Despite these limitations, this study presents new insights regarding the systemic immune dysfunction associated with death in HIV-associated cryptococcal meningitis. A clear immune signature was identified at baseline in the blood of nonsurvivors, characterized by increased inflammation, monocyte deactivation (or anergy), and downregulation of Th1 responses. This immune signature remained an independent determinant of mortality even after adjustment for other baseline factors. Given the known immune modulating effects of GXM and the strong association between serum GXM titer and IL-10 concentration observed in this cohort, it is plausible that this may be a direct consequence of disseminated infection with C. neoformans. Further work is required to define the mechanism and develop strategies to reverse the immune phenomenon.

Supplementary Data
Supplementary materials are available at http://jid.oxfordjournals.org. Consisting of data provided by the author to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the author, so questions or comments should be addressed to the author.