Frequency, Severity, and Prediction of Tuberculous Meningitis Immune Reconstitution Inflammatory Syndrome

Tuberculous meningitis (TBM) immune reconstitution inflammatory syndrome is a severe complication of antiretroviral therapy in human immunodeficiency virus–associated TBM. We found that high cerebrospinal fluid neutrophil counts and Mycobacterium tuberculosis culture positivity at TBM presentation characterize, and cytokine concentrations predict, this syndrome.

We therefore investigated clinical and laboratory findings in ART-naive HIV-infected patients who presented with TBM. We undertook serial cerebrospinal fluid (CSF) sampling in patients who did and did not develop TBM-IRIS.

Setting
This prospective, observational study was performed at GF Jooste Hospital, a public sector referral hospital in Cape Town. The hospital serves a low-income, high-density population in which the tuberculosis notification rate exceeds 1.5% per year with 70% of tuberculosis cases coinfected with HIV [19].

Participants
ART-naive HIV-infected patients aged ≥18 years presenting with meningitis from March 2009 through October 2010 were screened for study inclusion. HIV infection was diagnosed using 2 rapid HIV antibody tests and confirmed by HIV load. Definite TBM was diagnosed when acid-fast bacilli were seen, or when M. tuberculosis was cultured from CSF. Probable TBM was diagnosed when a patient showed clinical, laboratory, and radiological features of TBM in the absence of other infective causes for presentation [20]. Paradoxical TBM-IRIS was diagnosed according to a published definition for tuberculosis IRIS modified for meningitis [7,8]. The definition had 3 components: (1) TBM diagnosis before starting ART and improvement on tuberculosis treatment prior to ART initiation; (2) onset of TBM-IRIS manifestations (ie, new, recurrent, or worsening clinical features of TBM) within 3 months of ART initiation; and (3) exclusion of alternative causes for clinical deterioration.
Patients were ineligible if they had a contraindication to lumbar puncture, including unequal pressures between individual brain compartments on brain imaging, or severe TBM (ie, modified British Medical Research Council [BMRC] grade III disease severity) [21]. The University of Cape Town Research Ethics Committee approved the study and written informed consent was obtained from all patients or their relatives.

Procedures
Demographic data and history of tuberculosis disease, HIV infection, and other systemic illnesses and medications were recorded. Patients underwent general physical and neurological examination. Chest radiography, phlebotomy, and lumbar puncture were performed. In patients with suspected raised intracranial pressure or focal neurological deficits, brain computed tomography scanning was performed prior to lumbar puncture. CSF analysis included biochemistry, cytology, microbiology (including microscopy and culture for fungi and pyogenic bacteria), syphilis serology, HIV load, and Cryptococcus latex agglutination titer. Ziehl-Neelsen staining of sediment and M. tuberculosis culture was performed. If mycobacteria were cultured from CSF, tuberculosis polymerase chain reaction (PCR; Genotype MTBDRplus, Hain Lifesciences) was performed to determine susceptibility to rifampicin and isoniazid. CSF varicella zoster virus PCR was performed if the etiology was suspected. CSF was also stored at −80°C and analyzed for a range of inflammatory markers on the Bio-Plex platform (Bio-Rad Laboratories, Hercules, CA) using customized Milliplex kits (Millipore, St Charles, MO) according to the manufacturer's instructions.
At TBM diagnosis, patients started tuberculosis treatment according to national guidelines [22] and prednisone (1.5 mg/ kg/day). After 2 weeks of treatment and prior to initiation of ART, patients were assessed for improvement on tuberculosis treatment. The initial ART regimen was stavudine, lamivudine, and efavirenz. Later during the study, tenofovir replaced stavudine according to revised national guidelines. CSF investigations performed at TBM diagnosis were repeated at the time of ART initiation, 2 weeks later, and at time of TBM-IRIS presentation and 2 weeks thereafter. Prednisone was reduced to 0.75 mg/kg/day 4 weeks after starting ART and discontinued 2 weeks thereafter, unless the patient developed TBM-IRIS. At TBM-IRIS presentation, investigations were performed to exclude alternative causes of deterioration. Prednisone was either recommenced or the dose increased. Patients were followed for the duration of tuberculosis treatment (9 months); routine visits were at 2 weeks, 4 weeks, 6 weeks, 12 weeks, 6 months, and 9 months after TBM diagnosis. Patients were seen more frequently during deterioration.

Statistical Analysis
Statistical analysis was performed using the GraphPad Prism version 5, R version 2.14.1, and StatXact version 9 software packages. Categorical variables were compared using χ 2 or Fisher exact test. Continuous variables were compared between groups and time points within groups, using the Wilcoxon rank sum and Wilcoxon matched pairs tests, respectively. Adjusted relative risks (RRs) were evaluated using Cochran-Mantel-Haenszel tests and tests of homogeneity when considering categorical risk factors. Log-binomial models were fitted to continuous risk factors. Significance testing was done using 2-sided P values with P < .05 taken as significant.
The predictive accuracy of CSF neutrophil counts at TBM diagnosis for TBM-IRIS was assessed using nonparametric area under the receiver operating characteristic curve (AUC). Additionally, a model to predict TBM-IRIS risk was developed from 5 prespecified cytokines measured in CSF at time of TBM diagnosis. Interleukin 6 (IL-6), interleukin 10, interleukin 12p40, interferon gamma (IFN-γ), and tumor necrosis factor alpha (TNF-α) were selected as candidate markers of TBM-IRIS based on previous studies [23,24]. We prespecified the analysis for evaluating the multivariate cytokine model as follows. Significant cytokines comparing TBM-IRIS and non-TBM-IRIS using Wilcoxon rank sum tests were selected for a logistic regression model. Nonsignificant cytokines were dropped, resulting in a final model. The entire model building process was evaluated using leave-one-out cross-validation, the bootstrap method [25], and the permutation test to provide a crossvalidated (nonparametric) estimate of the AUC, P values, and 95% confidence intervals (CIs). As a secondary analysis, we examined whether the addition of CSF neutrophils and/or lymphocyte counts would improve the model's predictive ability.

TBM Presentation
The final diagnoses and reasons for exclusion of patients with meningitis are shown in Figure 1 Tables 1 and 2 show the demographic and baseline characteristics of these 2 groups. Baseline characteristics were similar between groups, although patients who developed TBM-IRIS had a longer duration of symptoms (median, 19 vs 9 days) and more frequent features of extrameningeal tuberculosis, such as chest symptoms (81% vs 44%) and chest radiographic abnormalities (81% vs 50%). Five TBM-IRIS patients (31%) and 3 non-TBM-IRIS patients (17%) developed features of extrameningeal tuberculosis IRIS.

Serial Blood and CSF Findings
Baseline blood investigations were similar between TBM-IRIS and non-TBM-IRIS patients with the exception of serum sodium concentrations, which were lower in TBM-IRIS patients (median, 123 vs 130 mmol/L, P = .01, Table 1). Table 2 and Figure 2 show serial blood and CSF findings. There was a significant rise in CD4 counts between starting ART and 2 weeks later in both TBM-IRIS (median, 93-158 cells/μL, P = .009) and non-TBM-IRIS (median, 145-178 cells/μL, P = .04) patients. Between these time points, blood and CSF HIV loads decreased significantly (P < .001) in both groups. At TBM diagnosis, TBM-IRIS patients had higher CSF cell counts, in particular neutrophils (median, 50 vs 3 cells ×10 6 /L, P = .02, Table 2). Similarly, neutrophil percentages from individual samples were higher in TBM-IRIS patients compared with non-TBM-IRIS patients (median, 36% vs 0%, P = .009). CSF to blood glucose ratios were lower in TBM-IRIS patients (median, 0.24 vs 0.47, P = .005). In both groups, CSF parameters initially improved on tuberculosis treatment (Table 2 and Figure 2). However, at TBM-IRIS presentation, TBM-IRIS patients showed findings of recurrent inflammation. In this group, lymphocyte and neutrophil counts at TBM-IRIS presentation were similar, and protein concentrations higher, compared with the same parameters at TBM diagnosis (median protein, 3.11 g/L at TBM-IRIS vs 2.70 g/L at TBM diagnosis, P = .007).
Mycobacterium tuberculosis was cultured from the CSF of 15 TBM-IRIS patients (94%) and 6 non-TBM-IRIS patients (33%) at TBM diagnosis; the risk of developing TBM-IRIS if CSF M. tuberculosis culture was positive at this time point was 71.4% (15/21), compared with a risk of 7.7% (1/13), corresponding to an RR of 9.3 (95% CI, 1.4-62.2, P = .004). Additional analyses considered the RRs of culture positivity adjusting for the following known or potential risk factors for tuberculosis IRIS: baseline viral load (median, ≥330 000 copies/mL, equivalent to log 10 = 5.52, vs <330 000), CD4 count (median, ≤137 cells/μL vs >137), evidence of disseminated disease in the form of an abnormal chest radiograph suggesting miliary disease, and duration of illness (median duration, ≥2 weeks vs <2 weeks). Mycobacterium tuberculosis culture positivity remained a significant risk factor after adjusting for these factors (Supplementary Table 1). There was no evidence of differences in the RRs according to these factors after considering culture status. Some TBM-IRIS patients remained culture positive after 2 weeks (n = 7) and 4 weeks (n = 2) of tuberculosis treatment. No non-TBM-IRIS patients were culture positive after starting tuberculosis treatment. All cultures were fully drug sensitive with the exception of one, which was monoresistant to isoniazid.

Analysis of Baseline CSF Neutrophil Count and Cytokine Concentrations to Predict TBM-IRIS
Concentrations of the prespecified cytokines from which the model to predict TBM-IRIS was developed are shown in Supplementary Table 2 and Supplementary Figure 2. The final multivariate logistic regression model included IFN-γ and TNF-α and produced a cross-validated AUC of 0.91 (95% CI, .53-.99, P = .02), indicating high diagnostic accuracy when jointly considering these 2 cytokines to differentiate TBM-IRIS from non-TBM-IRIS at time of TBM diagnosis. The odds ratio (OR) for TNF-α was 1.85 ( per 10 pg/mL, P = .006), indicating an 85% increase in the odds of IRIS for every 10 pg/mL increase in TNF-α (after adjusting for IFN-γ). The OR for IFN-γ was 0.64 ( per 100 pg/mL, P = .01), indicating decreased odds of IRIS with increasing IFN-γ (after adjusting for TNF-α). Figure 3 provides a heatmap representation of the predicted probabilities of the resulting model with the observed values overlaid. Neutrophil counts produced an AUC of 0.72 (95% CI, .54-.90, P = .03), indicating modest discriminatory accuracy for TBM-IRIS. However, including CSF neutrophil and lymphocyte counts in the model did not improve its ability to predict TBM-IRIS.

Management and Outcome
In 13 patients prescribed prednisone (0.75-1.5 mg/kg/day), the dose was increased at TBM-IRIS diagnosis. Prednisone was restarted at a dose of 1.5 mg/kg/day in the other 3 TBM-IRIS patients. The median total duration of corticosteroid treatment in TBM-IRIS patients was 109 days (IQR, 69-141) compared with 35 days (IQR, 20-43) in non-TBM-IRIS patients. ART was interrupted during TBM-IRIS in 1 patient because of brainstem involvement. This patient made a full recovery and had no recurrent symptoms after recommencement of ART under prednisone cover. At 9 months' followup, all non-TBM-IRIS patients were alive (including 1 who had defaulted study follow-up but continued tuberculosis treatment from a primary care tuberculosis clinic), 2 had marked cognitive impairment (international HIV dementia scale <10) [26], and 1 patient had marked cognitive impairment with residual hemiparesis. Twelve IRIS patients (75%) were alive at 9 months' follow-up, 2 patients showed marked cognitive impairment, 1 patient defaulted study follow-up but was alive, and 1 had marked cognitive impairment, residual hemiparesis, and hearing impairment. Death occurred in 4 (25%) TBM-IRIS patients at 33, 53, 60, and 118 days after TBM diagnosis and was related to TBM-IRIS in 2 patients. The 2 other deaths were due to a road traffic accident (n = 1) and unknown cause (n = 1). Kaplan-Meier survival analysis by TBM-IRIS vs non-IRIS (with a log-rank hypothesis test of the difference in survival between these 2) was nonsignificant.

DISCUSSION
This is the first prospective study of TBM-IRIS. In our cohort, tuberculosis IRIS presenting as TBM-IRIS (47%), as well as tuberculosis IRIS involving any organ system (56%), was more frequent than in previous studies [2][3][4]27]. Extrapulmonary tuberculosis is a risk factor for tuberculosis IRIS [14,28], and our cohort included only patients with TBM. A shorter interval between starting tuberculosis treatment and ART (which was 2 weeks in our study) similarly increases the risk of tuberculosis IRIS [2][3][4]. The high TBM-IRIS incidence we observed is striking considering that all TBM-IRIS patients were taking prednisone (1.5 mg/kg/day) at time of ART initiation and 13 of these patients (81%) were taking prednisone (0.75-1.5 mg/ kg/day) at the time of developing TBM-IRIS. Adjunctive corticosteroids have been shown to reduce mortality in TBM and tuberculosis pericarditis, presumably by reducing pathological host immune responses [21,29]. In paradoxical tuberculosis IRIS, the symptomatic benefit of corticosteroids was demonstrated in a randomized trial in which prednisone was compared to placebo [30]. For these reasons, we anticipated that prednisone would decrease the risk of tuberculosis IRIS.
TBM-IRIS was associated with a poor outcome; 2 patients (13%) died as a result of TBM-IRIS, all-cause mortality at 9 months was 25%, and 3 of 11 (27%) survivors examined at 9 months' follow-up were severely disabled, compared with no deaths and 3 of 17 (18%) patients with severe morbidity in the non-TBM-IRIS group. Our findings are similar to a previous study performed in neurological tuberculosis IRIS [8]. The poor outcome in at least 44% of TBM-IRIS cases emphasizes the need to predict and prevent, and improve the treatment of, TBM-IRIS.
Low serum sodium concentration is associated with death in HIV-associated TBM [31]. In our study, serum sodium concentration was lower in TBM-IRIS patients compared with non-TBM-IRIS patients at the time of TBM diagnosis. This may reflect the higher degree of tuberculosis dissemination observed in TBM-IRIS patients, which could have contributed to their risk of developing TBM-IRIS.
Our finding of an association between higher CSF neutrophils at TBM presentation and subsequent development of TBM-IRIS provides important and novel insight into the pathogenesis of tuberculosis IRIS. Not only were neutrophil counts higher in TBM-IRIS patients compared with non-TBM-IRIS patients, but neutrophil percentages for individual patients were similarly raised in TBM-IRIS. The neutrophil counts showed dynamic fluctuations over time in TBM-IRIS patients with a marked decrease on tuberculosis treatment, and a striking increase at TBM-IRIS onset. Similar changes in lymphocyte counts were not observed. Studies of tuberculosis IRIS have hitherto focused on the contribution of helper T-cell type 1 lymphocyte responses [32,33]. However, a role for myeloid cells in tuberculosis IRIS is suggested by a case report of a patient who died from unmasking pulmonary tuberculosis IRIS; postmortem histological examination of diseased lung showed a marked macrophage infiltrate [34]. We have found cytokines of predominantly myeloid origin (IL-6 and TNF-α) to be consistently elevated in patients with tuberculosis IRIS, compared with those who did not develop IRIS [23]. Oliver et al [6] also reported an association between plasma cytokines (interleukin 18) and chemokines (CXCL-10) of the innate immune system and tuberculosis IRIS. In an animal model, immune reconstitution following transfer of mycobacteriaspecific CD4 T cells to T-cell-deficient mice infected with Mycobacterium avium was associated with marked increases of both blood and lung CD11b cells (likely representing inflammatory monocytes and neutrophils) [35]. Our results suggest that neutrophils contribute to tuberculosis IRIS pathogenesis. The combination of high CSF TNF-α and low IFN-γ concentrations at the time of TBM diagnosis predicted TBM-IRIS in this cohort. Simmons et al [36] reported a negative correlation between CSF IFN-γ and mortality in HIV-infected patients with TBM. Conversely, a positive correlation was found between CSF IFN-γ and TNF-α concentrations and TBM disease severity by others [37].
Several studies have shown an association between disseminated and extrapulmonary tuberculosis and subsequent tuberculosis IRIS [14,28,38,39]. At TBM diagnosis, CSF M. tuberculosis culture positivity, which reflects mycobacterial antigen load, was a major risk factor for developing TBM-IRIS (RR = 9.3). Furthermore, 7 TBM-IRIS patients (44%) were persistently CSF M. tuberculosis culture positive after 2 weeks of tuberculosis treatment and 2 patients (13%) remained culture positive after 4 weeks of tuberculosis treatment. This strongly supports the inference that a high M. tuberculosis bacillary load at time of starting ART is a risk factor for tuberculosis IRIS [40]. The findings suggest it important to optimize tuberculosis treatment prior to starting ART in patients at high risk of developing TBM-IRIS.
We acknowledge several limitations. Because of the relatively small sample size, the study may not have been powered to detect further differences between IRIS and non-TBM-IRIS patients. Only patients with less severe disease (BMRC TBM grade 1 and 2) and those without contraindications to lumbar puncture were enrolled, resulting in the exclusion of a significant proportion of TBM patients presenting in our setting [41]; our results may therefore not be generalizable to ARTnaive patients presenting with severe HIV-associated TBM. The model to predict TBM-IRIS needs further validation and exploration with independent data.
In conclusion TBM-IRIS complicated the course of treatment of HIV-associated TBM in nearly half our patients, despite the use of adjunctive corticosteroid therapy. The manifestations were severe, fatal in 2 cases. The occurrence of TBM-IRIS associated with CSF M. tuberculosis culture positivity and a high neutrophil count at both baseline and at the time of TBM-IRIS. The baseline relationship between CSF TNF-α and IFN-γ predicted TBM-IRIS. These observations provide novel insight into the pathogenesis of this condition and provide rationale to individualize ART beyond 2 weeks in this devastating, partly iatrogenic, condition.

Supplementary Data
Supplementary materials are available at Clinical Infectious Diseases online (http://www.oxfordjournals.org/our_journals/cid/). Supplementary materials consist of data provided by the author that are published to benefit the reader. The posted materials are not copyedited. The contents of all supplementary data are the sole responsibility of the authors. Questions or messages regarding errors should be addressed to the author.