Introduction Cryptococcal meningitis (CM), caused by the fungal organism Cryptococcus neoformans (commonly termed simply “cryptococcus”) is the most common fatal central nervous system (CNS) infection in persons with AIDS and is the initial AIDS-defining illness in 20%–30% of AIDS patients in Africa, causing 20%–40% of AIDS-attributable mortality [1]–[4]. The US Centers for Disease Control (CDC) estimates approximately 1 million CM cases occur annually, with 70% in sub-Saharan Africa [5]. The incidence of CM among Ugandans with CD4 counts 1 event, thus the number of events is more than the total listed as the N for the group. b Non-IRIS deaths include: four suspected IRIS events. Two deaths are excluded in persons with virologic suppression with suspected clinical IRIS who refused LPs, with one of these deaths attributed to suicide. Association between IRIS and Death In the absence of ART, CM in Africa is associated with high mortality [6],[45]. We analyzed mortality data in our cohort to determine whether the development of IRIS was associated with increased mortality. Overall, 28 of 101 CM patients died (27%) after initiating ART. Mortality was 36% (16/45) in patients with IRIS and 21% (12/56) in those without IRIS (Figure 2). Causes of death were cryptococcal IRIS (n = 16), virologic failure and CM relapse (n = 1), paradoxical TB-IRIS (n = 1), aspiration after AZT-induced severe anemia and lactic acidosis (non-IRIS) (n = 1), renal failure (n = 1), cerebrovascular event (n = 1), suspected but unverified cases of CM-IRIS (n = 4), suspected pulmonary emboli (n = 2), and one unknown cause of death occurring at home. By time-to-event analysis, mortality was higher among those experiencing cryptococcal IRIS (HR = 2.3, 95% CI 1.1–5.1, p = 0.04) than the remainder of the cohort, alone or after adjustment for baseline CD4 (p = 0.11 for CD4). Our findings suggest that the development of IRIS is associated with an increased risk for death in patients with recent CM. 10.1371/journal.pmed.1000384.g002 Figure 2 Cumulative survival in persons with prior cryptococcal meningitis newly initiating HIV therapy in Uganda is stratified by the occurrence of cryptococcal IRIS. IRIS was associated with increased mortality (HR = 2.4, 95% CI 1.1–5.3, p = 0.035). Included with the controls are three suspected, but unproven cases of CM-IRIS, three unknown causes of death (two suspected pulmonary emboli). Two deaths have been excluded of persons with known virologic suppression with clinical IRIS who refused lumbar punctures to exclude alternative etiologies of their deterioration, of which one of these deaths was attributed to suicide. Demographic Risk Factors for IRIS We sought to determine the risks for IRIS by analyzing demographic and standard HIV parameters. For participants with recent CM, we identified no statistically significant difference among those who did or did not develop IRIS in baseline number or change in CD4+ T cell number, in baseline plasma HIV RNA, nor in virologic suppression at 12, 24, 36, or 48 wk (Table1). CD4 recovery is displayed in Figure S2. The lack of differences in these variables pertained to univariate or multivariate analyses, controlling for CRP, body mass index (BMI), cytokine profiles, as well as time-to-event analysis or comparison of quartiles. Thus, the degree of immunosuppression, rate of immune recovery after ART, level of viremia, and degree of viral suppression on ART were not predictors of CM-IRIS in this prospective cohort with very severe immunosuppression. Median serum CRAG titers when starting ART were 4-fold higher in participants with IRIS (1∶512 cases versus 1∶128 controls, p = 0.006). This result is in marked contrast to our previous CSF report from 85 members of this cohort, in whom there were no differences in CSF CRAG titer at time of their initial CM diagnosis between IRIS cases and controls (median CRAG 1∶1024 for both groups) [46]. A pre-ART serum CRAG ≥1∶512 was associated with later CM-IRIS (OR = 4.2, 95% CI 1.7–10.3; p = 0.002). Though serum was not available for direct comparison at time of their initial CM diagnosis, the implication is that patients with IRIS likely had poorer antigen clearance in the 5-wk median interval between CM diagnosis and ART initiation. A concern among many clinicians is that early initiation of ART in patients with CM could lead to an increased incidence of IRIS and increased mortality [47]. We identified no statistical significant difference in the incidence of IRIS in those initiating ART 11–28 d from CM-diagnosis compared with those who waited ≥28 d, (44% versus 55% respectively; p = 0.4) nor a statistical difference in survival (p = 0.9). Thus, earlier initiation of ART following the diagnosis of CM was not a risk factor for IRIS. Predictive IRIS Model Based on a training set derived from pilot data through 2008 [40], we considered nine possible biomarkers to incorporate into a predictive IRIS model. We used penalized logistic regression with penalties defined by the lasso method to develop a classification model with the best overall fit. Applying this model to our cohort, we were able to stratify persons into high-, moderate-, and low-risks groups with IRIS incidences of 82%, 47%, and 22%, respectively (Figure 3), based on seven serum biomarkers collected on the day of ART initiation (IL-4, IL-17, G-CSF, GM-CSF, CCL2 [MCP-1], TNF-α, and VEGF). The predictive performance (i.e. C-statistic) as quantified by the area under the curve (AUC) of a receiver operating characteristic (ROC) curve is 0.82 for correct prediction (p 32 mg/l levels (highest quartile) experienced IRIS with an increased incidence of 74% (OR = 3.9, 95% CI 1.3-11.3, p = 0.01) and shorter time-to-event (HR = 3.4, 95% CI 1.8–6.6, p 0.1) with the possible exception of D-dimer being more elevated in non-CNS events (median 3.68 versus 2.22 µg/ml, unadjusted p = 0.031). Interestingly, the inflammatory profile distinguished relapse from CM-IRIS in the four cases of culture-positive CM relapse. In the relapse cases, CRP was normal ( 32 mg/l with an AUC of 0.84 for correct classification with 78% sensitivity, 81% specificity, 60% positive predictive value, and 91% negative predictive value. A more parsimonious three-biomarker model of increasing IL-17 or CRP >32 mg/l, and decreasing GM-CSF, predicted increasing mortality (AUC = 0.76). A weighted score based on these three biomarkers was used to classify high- and low-risk mortality groups (Figure 5). Persons scored as high risk had 69% mortality (i.e., 69% positive predictive value) whereas those with low risk had 83% survival (i.e. 83% negative predictive value), demonstrating the potential clinical utility of the tool to identify patients at increased and decreased risk for death. CRP alone was also clinically useful for predicting risk of death (Figure 5). Participants with a CRP >32 mg/l had 8-fold higher odds of all-cause mortality compared to those with CRP 32 mg/l (p = 0.004) (AUC = 0.76). With a categorical cutoff point for the model with high risk of probability >0.5 to 1.0, the positive predictive value was 69% and negative predictive value 83% with 48% sensitivity and 92% specificity. (C) Survival stratified by baseline, pre-ART CRP level of 500 cells/µl) was present in over one-quarter of the cohort; however there were not statistical differences in eosinophils or total serum IgE between cases and controls. Overall, our data suggest that Th2 responses to cryptococcus are associated with poor outcomes, including increased risk for IRIS and death. Risk factors for IRIS extend beyond the Th1:Th2 paradigm. Increased Th17 responses (i.e. IL-17) were a pre-ART risk for both IRIS and mortality on ART. Proinflammatory Th17 cells have been previously hypothesized as important in IRIS pathogenesis [60],[61]. In normal immune homeostasis, a balance between Th17 cells and regulatory T cells (Treg) is crucial. A Th17 imbalance can cause autoimmune diseases. In this balance, IL-6 plays a key role in naïve T cell differentiation inducing Th17 differentiation and inhibiting Treg differentiation in the presence of TGF-β [62]. The copious IL-6 present before IRIS events may alter the balance between Treg and Th17 cells, suppressing Treg differentiation or function during ART-related immune reconstitution. Additionally, other cytokines that we identified may play a role in IRIS pathogenesis by contributing to ineffective macrophage activation. For example, low pre-ART levels of TNF-α or VEGF were associated with increased IRIS risk. TNF-α, secreted by macrophages and T cells, is particularly important in activating antigen-presenting cells. Absence of TNF-α causes failure of mature dendritic cell activation and recruitment, thereby blunting further recruitment of T cells [63]. Lack of TNF-α could indicate failure to present and/or process antigen and could contribute to ineffective macrophage activation in patients at risk for IRIS. In cryptococcosis, VEGF is secreted by a variety of leukocytes but especially by CD4+ T cells during antigen-specific responses to cryptococcal mannoprotein being presented by MHC-II molecules [64]. Decreased VEGF may reflect greater immune dysfunction due to failure of CD4+ T cell antigen recognition from antigen-presenting cells. The downstream effects of decreased VEGF would further diminish both chemotaxis and the costimulatory activity of VEGF on IFN-γ–secreting Th1 memory T cells [65]. Thus, decreased VEGF may then diminish the Th1 response and shift the Th1:Th2 balance toward Th2 responses. The decreased levels of TNF-α and VEGF in patients at risk for IRIS may be causes or reflections of impaired macrophage function. Overall, the inappropriate Th2 responses and IL-4 production, coupled with ineffective macrophage activation, suggest that a failure in pathogen recognition and clearance could set the stage for paradoxical IRIS by promoting antigen persistence. Upon eventual immune restoration of more appropriate antigen-specific responses, these responses are exaggerated because of the abundance of uncleared foreign antigen and promotion by IL-6 of a proinflammatory state. Study Strengths Strengths of this study include its prospective design, careful and complete follow-up, and integration of pathophysiologic analyses in the context of characterizing clinical phenotypes. As such, we believe the results reported here are generalizable to cryptococcal IRIS as a pathophysiologic entity. In general, the diagnoses of other forms of IRIS are somewhat subjective, but in CM-IRIS, the identification of IRIS in the CNS is very objective due to the ability to examine the CSF, where the inflammation occurs. A problem arises with non-CNS manifestations of IRIS, which often represent diagnoses of exclusion and depend on the diagnostic capabilities available. In this cohort, we identified several (n = 20) non-CNS IRIS events, of which 60% were associated with later CNS-IRIS events or with supportive/definitive histopathology (Text S1). Eight cases (retinitis, optic neuritis, keratoconjunctivitis, pneumonitis; n = 5) were probable IRIS diagnoses and unsupported by histopathology or culture (17% of total IRIS cases). Although cryptococcus most often causes meningitis, cryptococcosis is a disseminated disease with systemic CRAG measurable for months in peripheral blood [66]. In our experience here, dead C. neoformans causing IRIS reactions were identified in tissue by histopathology of brain, lymphatics, gut, skin, and tongue and were isolated via bronchoalveolar lavage. Thus, non-CNS cryptococcal IRIS events do occur, allowing these “probable events” as likely, but not definitely, attributable to IRIS. In separately considering these non-CNS IRIS events, there were no statistical differences in the inflammatory profiles observed at time of IRIS as compared to CNS-IRIS events, except for possibly D-dimer. In Uganda, there is likely high and ongoing environmental exposure to cryptococcus [1],[4], thus pneumonitis-IRIS could equally be triggered by persisting original antigen or by environmental re-exposure manifesting effectively as an acute hypersensitivity pneumonitis reaction. Pulmonary cryptococcosis often goes unrecognized and/or is suspected as smear-negative TB [67]. Study Limitations This prospective observational study has allowed us to characterize the statistical association between cytokine profiles and IRIS. Of course, this does not prove causality; however, the anomalies we report are biologically plausible within the known immunology of responses to cryptococcus. A challenge in pathogenesis studies such as ours is that the inflammatory profiles associated with IRIS are heterogeneous, representing a continuum that is likely dependent on the duration of symptoms, the antigen, and the robustness of the event. In cytokine profiling, six of our participants had minimal immunologic perturbation in peripheral blood even though they clinically deteriorated, presenting with headache and elevated intracranial pressure several months into ART. Thus, some patients fulfilling the IRIS clinical case definition either have localized immune responses that were not detectable in peripheral blood, or they may not have had true immunologic IRIS but instead had delayed complications of CM that were clinically indistinguishable from CM-IRIS. This heterogeneity may have implications for IRIS management and the response to anti-inflammatory therapies. Although this prospective study of CM-IRIS is the largest to date, our biomarker data should be viewed as hypothesis-generating and will require validation in future cohorts. We plan to validate these biomarkers in subjects enrolled in the multi-site Cryptococcal Optimal ART Timing (COAT) trial (NCT 01075152; http://clinicaltrials.gov/ct2/show/NCT01075152) and investigate these biomarkers with regard to timing of ART initiation after CM diagnosis. This validation will give insight on whether altering the timing of ART initiation is a potential intervention to alter high risk or whether other strategies should be pursued. Once validated, creating diagnostics such as multiplex ELISAs or bead-based techniques could move this into clinical use. Conclusions Given the high incidence, morbidity, and mortality associated with CM-IRIS, identifying patients at risk for IRIS may enable interventions to improve management. Three distinct phases of IRIS pathogenesis can be identified (Table 5). First, before ART, a paucity of innate inflammatory responses or inappropriate (Th2) responses promote ineffective antigen clearance. Second, after ART initiation, copious antigen presentation promotes proinflammatory signaling (e.g., IL-6, CRP, IL-7). Third, at the time of IRIS, a generalized cytokine storm occurs. The biomarkers identified here, although requiring validation, may help target interventions to decrease IRIS risk, such as (1) early adjunctive GM-CSF or IFN-γ to increase macrophage function, (2) antiparasitic therapy to eliminate coinfections that promote inappropriate Th2 bias, (3) anti-IL-6 receptor antibody (tocilizumab) therapy to blunt inflammatory signaling, or (4) delaying ART initiation. This study suggests that prediction of IRIS or death may be possible with measurement of pre-ART serum biomarkers. 10.1371/journal.pmed.1000384.t005 Table 5 Summary of paradoxical cryptococcal-IRIS pathogenesis hypothesis. Phase Immunologic Activity Evidence in CM-IRIS Patients Before ART • Paucity of appropriate inflammation for cryptococcosis and/or • ↓ TNF-α, G-CSF, GM-CSF, VEGF in serum↓ IFN-γ, G-CSF, TNF-α, IL-6 in CSF [46] • Inappropriate (Th2) responses resulting in: • ↑IL-4 pre-ART • Poor antigen clearance, pre-ART • Similar CSF CRAG at initial infection [46]Higher CRAG pre-ART After starting ART • Increasing proinflammatory signaling from APCs due to persisting antigen burden and failure to clear antigen • ↑ IL6 from macrophages [56] then downstream ↑ CRP production; ↑ IL-7 from APCs • Secondary activation of coagulation cascade • ↑ D-dimer At IRIS • Cytokine storm of multiple immune pathways of innate and adaptive immune systems • Th1 ↑ INF-γ, VEGF; TH17 ↑ IL-17Innate: ↑ IL-8, G-CSF, GM-CSF • Activation of coagulation cascade • ↑ D-dimer • Neuronal cell activation and damage • ↑ FGF-2 Supporting Information Alternative Language Abstract S1 Translation of the abstract into Spanish by Dr. Jose Debes. (0.03 MB DOC) Click here for additional data file. Alternative Language Abstract S2 Translation of the abstract into French by Dr. Anali Conesa Botella. (0.03 MB DOC) Click here for additional data file. Alternative Language Abstract S3 Translation of the abstract into Portuguese by DLW and Dr. Jaime Luís Lopes Rocha. (0.03 MB DOC) Click here for additional data file. Alternative Language Abstract S4 Translation of the abstract into Russian by Dr. Irina Vlasova-St. Louis. (0.03 MB DOC) Click here for additional data file. Alternative Language Abstract S5 Translation of the abstract into Japanese by Dr. Kosuke Yasukawa. (0.03 MB DOC) Click here for additional data file. Figure S1 The 45% cumulative incidence of paradoxical CM-IRIS events through 1 y of ART. All patients had prior CM that was diagnosed a median of 5 wk before initiating ART. Censored events are time through non-IRIS deaths (n = 5), suspected but unproven IRIS deaths (n = 4), unknown causes of death (n = 3 of which two were suspected pulmonary emboli), and voluntary ART discontinuation (n = 1). (0.13 MB TIF) Click here for additional data file. Figure S2 CD4+ T cell response in participants with CM-IRIS versus CM controls without IRIS. Shown are the mean ± SD of the absolute CD4+ T cell counts in 101 patients with prior CM, of whom 47 were women and 53 were men with a mean age of 36±8 y. The baseline median CD4+ T cell count was 19 (IQR 7–36, range: 1–179) cells/µl increasing by 12 wk of ART to 69 (IQR 44–115) cells/µl, p<0.001) with a gradual increase thereafter to a median of 124 (62–175) cells/µl at 48 wk. Plasma HIV RNA of 5.3±0.5 log10 copies/ml at baseline achieved suppression (<400 copies/ml) in 70% by 12 wk, 90% at 24 wk, and 87% at 48 wk. There were no statistical significant differences in CD4+, CD4 change, or HIV-1 viral load between those who did develop IRIS and those who did not and had uneventful immune reconstitution. (0.13 MB TIF) Click here for additional data file. Figure S3 ROC Curve for parsimonious IRIS prediction models. ROC curve for more exhaustive or parsimonious models for IRIS prediction using log2 transformed biomarkers. AUC ranges from 0.72 to 0.825. IRIS probability = , where z is calculated as follows: (0.22 MB TIF) Click here for additional data file. Figure S4 ROC curves for predictive mortality models. (0.24 MB TIF) Click here for additional data file. Table S1 Serum cytokine profiles at time of suspected IRIS events. (0.05 MB PDF) Click here for additional data file. Table S2 Time-to-event analysis of hazard of IRIS by cytokine profile. (0.06 MB DOC Click here for additional data file. Text S1 Clinical spectrum of IRIS events in the Ugandan cohort. (0.07 MB DOC) Click here for additional data file.
