Elevated cerebrospinal fluid cytokine levels in tuberculous meningitis predict survival in response to dexamethasone

Adjunctive treatment with anti-inflammatory corticosteroids like dexamethasone increases survival in tuberculosis meningitis. Dexamethasone responsiveness associates with a C/T variant in Leukotriene A4 Hydrolase (LTA4H), which regulates expression of the pro-inflammatory mediator leukotriene B4 (LTB4). TT homozygotes, with increased LTB4, have the highest survival when treated with dexamethasone and the lowest survival without. While the T allele is present in only a minority of the world’s population, corticosteroids confer modest survival benefit worldwide. Using Bayesian methods, we examined how pre-treatment levels of cerebrospinal fluid (CSF) pro-inflammatory cytokines affect survival in dexamethasone-treated tuberculous meningitis. LTA4H TT homozygosity was associated with global cytokine increases, including TNF. Association between higher cytokine levels and survival extended to non-TT patients, suggesting that other genetic variants may also induce dexamethasone-responsive pathological inflammation. These findings warrant studies that tailor dexamethasone therapy to pre-treatment CSF cytokine concentrations, while searching for additional genetic loci shaping the inflammatory milieu.


INTRODUCTION
Tuberculous meningitis is the most lethal form of tuberculosis with a mortality of 25-40% in drug-sensitive HIV uninfected adults [1][2][3]. Drug resistant infection and HIV coinfection leads to even higher mortality [1,3]. Because multiple investigations suggest that dysregulated inflammation plays a role in mortality from this disease, corticosteroids, which are broadly acting anti-inflammatory drugs, are now routinely used as adjunctive therapy to anti-tubercular antibiotics [4][5][6]. The relatively modest reduction of mortality with corticosteroids suggests that tuberculous meningitis may elicit different inflammatory responses, with corticosteroids helping those with high levels of inflammation. Genetic variation is likely to control these heterogeneous responses and a common functional variant in the Leukotriene A4 Hydrolase (LTA4H) gene is associated with responsiveness to dexamethasone, a potent corticosteroid [7-9]. LTA4H is a key enzyme in arachidonic acid metabolism that catalyzes the production of leukotriene B4, a pro-inflammatory lipid mediator with pleiotropic inflammatory effects [10,11] . A C/T transition in the promoter modulates LTA4H gene and thereby protein expression. Consistent with its expression mediating an inflammatory milieu, CC and TT homozygotes have the lowest and highest LTA4H expression, respectively, with intermediate expression in CT heterozygotes. TT homozygotes have the greatest survival benefit from dexamethasone while suffering the highest mortality among those not given this drug [8].
The role of LTA4H in controlling inflammation and survival in the context of mycobacterial infections was first identified in a zebrafish forward genetic screen where animals with both low and high LTA4H expression were more susceptible to Mycobacterium marinum infection than their wild type counterparts [8,12]. In the zebrafish, high LTA4Hmediated susceptibility is due to its increased product LTB 4
Because tuberculous meningitis is characterized by a necrotizing granulomatous reaction and macrophage-rich meningeal exudates [15,16], we wanted to determine if the LTA4H TT genotype mediates increases in cerebrospinal fluid (CSF) cytokines and if these increases are associated with dexamethasone responsiveness. Consistent with the findings of Tobin et al. [8], analysis of a second Vietnam tuberculous meningitis cohort, where all individuals had been treated with dexamethasone, showed that LTA4H TT HIV-uninfected individuals had increased survival over their non-TT counterparts [7,9]. The same study also determined CSF cytokine levels to test the prediction that TT individuals have a hyperinflammatory CSF profile reflected by elevated cytokines [7]. The original analysis of the cytokine profiles in this study was conducted using linear trend tests and found that the median levels of all 10 assayed cytokines were increased in TT compared to CC and CT patients, but in a multiple comparison test only the interleukins IL-1β, IL-2, and IL-6 increases achieved statistical significance [7].
Here, we have re-analyzed the cytokine data from HIV-uninfected adults with tuberculous meningitis using Bayesian methods. Bayesian analysis can detect significant results and relationships not detected by frequentist methods because they do not impose a penalty for multiple comparisons and can effectively detect significant differences that are hidden by Type 2 errors in frequentist analysis [9]. Moreover, we don't know the class of distribution (e.g. Normal, Gamma, etc) from which the cytokine values come. Bayesian methods can identify the likely class of distribution and take this information into account for the analyses even without having surety about the correct class.
Using Bayesian methods (detailed in Appendix 1 and Appendix 2), we find that survival in response to dexamethasone is associated with significant increases in all cytokines tested before or at the start of treatment, representing innate pro-inflammatory, helper T cellassociated and immunomodulatory classes. While the LTA4H TT genotype was associated with increases in these cytokines, we also found that increased cytokines are associated with survival in an LTA4H-independent manner in this dexamethasone-treated cohort.

RESULTS
In tuberculous meningitis patients, LTA4H TT genotype is associated with increased CSF levels of multiple cytokines, including TNF noncentral-t distribution) was the preferred distribution class both in the dataset as a whole and for the various subsets considered in our analyses (Appendix 2). Comparisons were performed using restricted geometric means as is appropriate for such heavy-tailed approximately logarithmically distributed data (Box 1 and Appendix 2). Furthermore, unlike the previous analysis, we made no assumption that there would be a linear trend with the number of T-alleles in a given patient. Using this method of analysis we found that TT patients had significant increases in all measured CSF cytokines, except Interferonγ (IFNγ) and IL-4, compared to both CC and CT patients who had similar levels to one another ( Figure 1A and Supplementary Table 1). Similarly, a comparison of cytokine levels in TT patients to those in combined non-TT (CT and CC) patients showed that these levels in TT patients were significantly higher for all cytokines except for IFN-γ and IL-4 ( Figure 1B).
The finding that a single T allele does not have a discernible influence on inflammatory pathways is consistent with the CC and CT patients in this cohort having similarly lower survival than TT patients when all patients were receiving dexamethasone therapy [7,9].
Thus, TT homozygosity is associated with increased cytokine concentrations across the board, including cytokines that are associated with an acute inflammatory response (TNF, IL-1β, and IL-6); with T cell activation and regulatory T cell homeostasis (IL-2), innate and adaptive type-1 immunity (IL-12 and IFNγ); innate and adaptive type-2 immunity (IL-4, IL-5, and IL-13), and immune modulation (IL-10). Importantly, TNF, which drives the pathogenesis caused by LTA4H excess in the zebrafish model of TB [8, 13,14] is significantly increased in TT patients.
The LTA4H TT genotype exerts a compensatory regulation on CSF cytokine levels in more severe disease.
Tuberculous meningitis patients can present with a wide-ranging disease severity, reflected by the presence or absence of focal neurological signs, or a generalized decrease in responsiveness including coma [17]. The modified British Medical Research Council tuberculous meningitis grading system categorizes patients into three grades in increasing order of severity [3,17]. Prior analysis of this cohort found that a disease grade was associated with a trend to increased cytokines across the board with a significant increase for only one, IFNγ [7]. Our analysis found all to be increased with increasing grade, with significant increases between grades for seven of the ten (Figure 2A and Supplementary   Table 1). Since there were increased cytokine levels for both the TT genotype and for higher disease grades, we predicted that these levels would be highest in TT patients in the higher disease grades. Whereas non-TT patients had a similar pattern of increased cytokines with increasing disease grade as the overall cohort ( Figure 2B), we were surprised to find that in TT patients, the pattern was reversed. The majority of the cytokines were lower in Grades 2 and 3 than in Grade 1, significantly so in many cases ( Figure 2C). The major shift occurred between Grades 1 and 2. Grade 3 cytokines were not lower than Grade 2; the levels were either similar in these two grades or Grade 2 levels were non-significantly lower. These findings suggest the existence of compensatory mechanisms in TT patients that limit extreme increases in cytokine levels driven by increased disease severity. Consistent with this hypothesis, when we compared cytokine levels in non-TT to those in TT patients stratified by disease grade, cytokine levels in TT patients were higher in all grades, with increases that were the greatest and most significant in Grade 1, rather than in Grades 2 and 3 ( Figure 2D- Both LTA4H TT-dependent and -independent CSF cytokine increases are associated with survival in response to dexamethasone Thuong et al. [7] compared cytokine levels independent of LTA4H genotype in tuberculous meningitis survivors to non-survivors following adjunctive dexamethasone treatment and found that survivors had increased cytokine levels. Our re-analysis confirmed this result -all cytokines were significantly increased in survivors compared to those who died ( Figure 3A). Because TT patients had significantly increased survival with dexamethasone as compared to non-TT patients [7,9], we hypothesized that the increased cytokine levels in survivors overall would be restricted to TT patients. However, even among non-TT patients only, survivors had significantly increased cytokines across the board when compared to those who died ( Figure 3B). We could not compare TT survivors to nonsurvivors as the four TT patients who died did not have CSF cytokine measurements.
Comparison of TT survivors to non-TT survivors revealed that most cytokines were significantly higher in TT survivors than in the non-TT survivors ( Figure 3B). CC and CT survivors each also had higher cytokines than non-survivors overall and in all three disease grades ( Figure 3-figure supplement 1). Together these analyses show that the default inflammatory response to tuberculous meningitis includes global increases in CSF cytokines, and suggest they are associated with a survival benefit from dexamethasone. Overlaid on these are further increases mediated by the LTA4H TT genotype.
The finding that the LTA4H TT patients have increased CSF cytokines over their non-TT counterparts provides an explanation for why they survive better when treated with dexamethasone than their non-TT counterparts [8]. However, we had now shown in this study that among dexamethasone-treated tuberculous meningitis patients, LTA4H TTindependent cytokine increases are also associated with survival, raising the question of whether non-TT patients might also benefit from dexamethasone. By the time the cytokine analysis study was undertaken, adjunctive dexamethasone had become standard-of-care treatment so that all patients were given this drug [7]. Therefore, to answer the question, we re-analyzed the survival data from the Tobin et al. study (Table 1), using recently-described Bayesian methods, which had compared survival of patients of the three LTA4H genotypes with and without adjunctive dexamethasone [8,9].
The survival of dexamethasone-treated CT heterozygotes was different between the two studies, appearing more similar to that of the TT patients in the Tobin et al. study but more similar to that of the CC patients in the Thuong et al. study [7,8]. So we re-analyzed both studies using Bayesian methods, separating the non-TT patients into the individual CC and CT genotypes. In the Tobin study, in the absence of dexamethasone treatment, TT patients had worse survival than both CC and CT patients with the difference being just short of being significant (maximum posterior probability 0.946) ( Figure 4A). There was no significant difference between CC and CT patients ( Figure 4A). Among dexamethasone treated patients, TT survival was significantly higher than CC survival ( Figure 4B). CT survival was in between the two, significantly higher than CC and non-significantly lower than TT ( Figure 4B). In the Thuong study, CT survival was significantly worse than TT and not significantly different from CC ( Figure 4C). We confirmed this shift in CT survival between the two studies by a direct comparison of the Tobin and Thuong studies. CT survival was significantly worse in the Thuong study whereas CC and TT survival were not significantly different in the two studies ( Figure 4 D-F).
Finally, we asked whether and how dexamethasone influenced the survival of each genotype in the Tobin study. Directly comparing survival of each of the three genotypes with and without dexamethasone, we found that TT patients derived the greatest benefit from dexamethasone, CT patients had a smaller but still significant benefit, and CC patients were neither helped nor harmed by dexamethasone ( Figure 4G-I).
In sum, because the CC and TT patients survived similarly in response to dexamethasone in the Thuong and Tobin studies, we can use the survival with and without dexamethasone in the Tobin study together with the pre-treatment cytokine levels in the Thuong study to draw the following two conclusions: (1) TT patients benefit very substantially from dexamethasone, consistent with their higher pre-treatment cytokine levels; and (2) among dexamethasone-treated patients, CC survivors have higher pre-treatment cytokine levels than non-survivors. These two findings can be reconciled by a model ( Figure   5) where dexamethasone reduces the higher pre-treatment cytokine levels in TT patients to a level optimal for survival, whereas the lower cytokine pre-treatment levels in many of the CC patients are lowered further by this treatment to suboptimal levels, so that there is no apparent benefit of the drug to the cohort overall.

DISCUSSION
Dysregulated intracerebral inflammation has long been thought to be responsible for the high mortality and morbidity of tuberculous meningitis. Multiple cytokines can be major effectors of dysregulated immune responses, yet pre-treatment cytokine data from TBM patients are limited [18,19]. Therefore, the Thuong et al. study [7] where CSF cytokines were collected in 306 HIV-uninfected patients, all of whom were treated with dexamethasone, together with comprehensive clinical information and survival analyses, provided an unprecedented opportunity. This cohort allowed for an analysis of CSF cytokine concentrations in TBM with respect to disease severity on presentation and outcome following dexamethasone treatment. Moreover, this study confirmed the survival benefit of the LTA4H TT genotype, providing the opportunity to ask if this hyperinflammatory genotype was associated with increased cytokines. Comparison of CSF cytokines applying frequentist statistical methods (linear trends tests) to median cytokine values showed that increased pre-treatment levels of most (8 of the 10 tested) were significantly associated with survival with dexamethasone, indicating that these higher levels are pathogenic. All cytokines were increased with increased disease grade, but only one of these, IFNγ, was significantly increased in those with Grade 3 disease. The LTA4H TT genotype was also associated with global increases in cytokine concentrations across the board in comparison to the non-TT genotype patients, but the differences were significant in only 3 of the 10 cytokines. Given that most cytokines are induced by interrelated and often shared signal transduction networks, notably the NFκB family of transcription factors [20], these patchy statistically significant differences were more likely to represent a Type 2 statistical error than biologically relevant patterns. So we turned to Bayesian methods to re-analyze these data and looked for associations between pre-treatment cytokine levels with disease severity, survival and LTA4H genotype, not only singly, but also in combination.
Bayesian analysis shows that all cytokines tested, representing multiple functional classes -innate pro-inflammatory, Th1-and Th2-associated and immunomodulatory -are associated with survival in this dexamethasone-treated cohort. This global induction of cytokines is consistent with the induced inflammatory trigger mediating effects upstream of a common signaling axis for all of them without significant additional downstream regulation. The finding that LTA4H TT genotype further increases all ten cytokines is consistent with prior work showing that LTB 4 binding to its receptors activates the NFκB pathway [21]. Furthermore, TNF, which has been implicated in the pathogenesis of tuberculosis, including tuberculous meningitis [13,14,22], appears to be dysfunctionally increased both in an LTA4H-independent and -dependent manner. Finally, this work highlights the role of the previously described regulatory circuits that dampen LTA4H TT -mediated inflammation (i.e., leukotriene B 4 ) in the context of one of the most lethal infectious diseases of humans [11,23].

Association of increased CSF cytokines and survival even independent of LTA4H genotype
Our initial goal in performing these analyses was to ask whether the LTA4H TT genotype is associated with global increases in pre-treatment CSF cytokines, as would be predicted by its activation of the NFκB pathway [21]. LTA4H TT individuals with tuberculous meningitis have a striking survival benefit from dexamethasone, which causes a global reduction in cytokines, and we find that the TT genotype is indeed associated with higher pre-treatment CSF cytokines across the board. However, even in Europe and Africa where the LTA4H T allele is much rarer (10% frequency) than in Asia (up to 33%; from https://tinyurl.com/y4c232e3) [8,9,24], multiple small studies have a modest survival benefit from corticosteroids comparable to that seen in Asia [4,5,25,26]. In fact, the earliest studies suggesting corticosteroid benefit, that gave the impetus for the larger randomized controlled trial in Vietnam [6], were done in patient populations that were mostly of European descent, in which the LTA4H TT homozygote genotype frequency would have been rare (1-4% of population) [24-26]. Perhaps our most important finding is that even among LTA4H non-TT individuals, higher CSF cytokines are associated with higher survival in response to dexamethasone. However, when we analyze a prior cohort that enables comparison of survival with and without dexamethasone (but lacks CSF cytokine analysis), we find that while TT patients gain a major survival advantage from dexamethasone, non-TT patients' survival is neither helped nor harmed by it. These two findings can be reconciled in two ways. The first is predicated on the idea that a major mechanism of dexamethasone's survival benefit is through its cytokine-reducing effect [27,28]. Non-TT pre-treatment CSF cytokines are lower on average than TT, and the currently-used dexamethasone dosages may decrease their cytokine levels to suboptimal levels ( Figure 5). In optimal concentrations, many of these cytokines play a role in the host responses that eliminate bacteria, and this function may be lost if they are lowered below a threshold. The model that dexamethasone optimizes TT cytokine levels while reducing non-TT levels too much can be tested when the results of an ongoing randomized clinical trial of dexamethasone for non-TT patients become available

Regulatory networks may keep LTA4H TT cytokine increases in check
The finding that disease severity increases cytokine levels in non-TT but not in TT patients suggests LTA4H TT-specific compensatory mechanisms that appear to dampen disease grade mediated cytokine increases. LTB 4 mediates its activity through two receptors

LTA4H TT-dependent and LTA4H-independent TNF dysregulation
Our finding that dexamethasone is associated with increased survival in both LTA4H TT and non-TT patients is tantalizing from a therapeutic standpoint. In the zebrafish, high LTA4H increases disease severity through increasing TNF, which in excess causes increased disease pathogenesis through a newly-identified programmed macrophage necrosis [13,14].
We were particularly interested in this question because in the zebrafish, several pathway-specific drugs that inhibit macrophage necrosis without being broadly anti-inflammatory have been identified, all of which are have a decades-long history of use in humans for other conditions [13,14]. Therefore, unlike glucocorticoids, these drugs would be beneficial to those with excessive TNF while being neutral to the other patients, as they target the downstream effects of excess TNF, without broadly reducing overall cytokines to levels which are detrimental to survival.

Conclusions and Implications for future studies
The use of Bayesian methods has enabled important insights into the induction and regulation in tuberculous meningitis and the possible detrimental effects of their dysregulation. On-going randomized control trials that will enroll >1200 participants are examining the role of adjunctive dexamethasone adults with tuberculous meningitis, stratifying participants according to HIV infection and LTA4H genotype [29, 33]. Pre-and post-treatment CSF cytokine analysis will be performed in all participants [29,33]. These trials will allow for validation of the analyses presented here as well as test the models and hypotheses that have arisen from them. Finally, some (though not all) studies done over decades across the globe have found adjunctive corticosteroid treatment to have a modest early benefit in the contagious and most common form of tuberculosis, that involving the lung, both in reducing inflammation and bacterial burdens [34][35][36][37]. Spatial studies of human tuberculous granulomas combining mass spectrometry find that necrotic tuberculous granulomas are enriched for LTA4H and TNF as compared to non-necrotic granulomas from the same lung, making it plausible that corticosteroids exert their effects through these determinants [38]. The analytical methods developed here could be readily tailored to examine the role of corticosteroids as drugs that may improve outcome in pulmonary TB as well.

MATERIALS AND METHODS
The anonymized tuberculous meningitis patient cohort CSF cytokine level data used here has been previously described [7,9] (Table 1). The tuberculous meningitis cohort used for survival analysis has also been previously described [6,8] (Table 1) Patients with mild disease (Grade 1) received 2 weeks of IV treatment (0.3mg/kg/day week 1, 0.2mg/kg/day week 2), followed by oral therapy for 4 weeks, starting at 0.1mg/kg/day in week 3, then a total of 3mg/day in week 4, decreasing by 1 mg/week. All patients in the cohort used for cytokine analysis were treated with adjunctive dexamethasone as above and CSF specimens collected on enrollment from 306 HIV-negative and 219 HIV-positive patients [7]. CSF for cytokine measurements was available in a smaller proportion of patients admitted to Hospital 1 than Hospital 2 (55.9% vs. 96.6%) (Supplementary Table 2). Further analysis showed that Hospital 1 TT patients were significantly less likely to have had cytokine measurements than non-TT patients (Supplementary Table 3). There was no such bias in relation to survival status or disease grade severity (Supplementary Table 3). To determine if the Hospital 1 collection bias in the TT vs. non-TT patients changed the analysis, all data were analyzed for the whole cohort and Hospital 2 alone (Supplementary Table 1

Definitions
• Posterior probability -the probability after seeing the data • Survivors (as opposed to nonsurvivors) -those who were not known to have died during the period that they were observed, i.e. those who either were censored or died. • Restricted geometric mean -the geometric mean of a distribution, restricted to an interval excluding only extreme tail events; see Appendix 2 for reasons for using this concept • Dithering -A process that spreads out data points concentrated on the maximum and minimum seen values, as these likely come from a limit in the measurement procedure

Abbreviated and example for Figures 1-3 and associated text
"A was significantly greater than B" -the posterior probability given the dithered data that the restricted geometric mean of the variable A in the relevant population was greater than that of the variable B was at least 0.95 (and similarly for "significant increase", etc.) "A was not significantly different from B" -the posterior probability given the dithered data that the restricted geometric mean of the variable A in the relevant population was greater than that of the variable B was between 0.05 and 0.95 (and similarly for "no significant difference", etc.) "A was greater than B" -the (restricted) geometric mean of the undithered data in A was greater that of B (note that the presence or absence of "restricted" makes no difference in this definition as all the observed data is within the range of restriction) "The probability that A was greater than B was p" -the posterior probability given the dithered data that the restricted geometric mean of the variable A in the relevant population was greater than that of the variable B was p           Log CSF cytokine [pg/ml], geometric mean TNF  IL1b  IL6  IL12  IFNg  IL2  IL4  IL5  IL13  Log CSF cytokine [pg/ml], geometric mean TNF  IL1b  IL6  IL12  IFNg  IL2  IL4  IL5  IL13  Log CSF cytokine [pg/ml], geometric mean TNF  IL1b  IL6  IL12  IFNg  IL2  IL4  IL5  IL13  Log CSF cytokine [pg/ml], geometric mean TNF  IL1b  IL6  IL12  IFNg  IL2  IL4  IL5  IL13  Log CSF cytokine [pg/ml], geometric mean TNF  IL1b  IL6  IL12  IFNg  IL2  IL4  IL5  IL13

Definitions -Figure 4 survival analysis
• Posterior probability -the probability after seeing the data • Mean posterior survival probability at time T -the expectation after seeing the data of the fraction of patients that will still be alive at time T • Hazard rate -the fraction of those still surviving that will die per unit time. A high hazard rate at a particular time indicates that patients are at high risk of dying at that time • Mean posterior hazard rate at time T -the expectation after seeing the data of the hazard rate at time T

Abbreviated and example usages
Onwardsfor the rest of the 270-day observation period Throughout -for the entire 270-day observation period "A was significantly greater than B at time T" -the posterior probability that A was greater than B, at time T, was at least 0.95 "A was not significantly different from B" -the posterior probability that A was greater than B was between 0.05 and 0.95 throughout "Group A survival was 30% greater than group B" -the mean posterior survival probability at 270 days, p A , was 30% absolute greater than the corresponding probability p B for group B. ("absolute" here meaning that p A = p B + 0.3, and not that p A = p B × 1.3) "Probability that group A survival was better than group B at time T was 0.97" -the posterior probability that group A survival probability at time T was greater than group B survival probability at time T was 0.97 (Note that this is not a reference to the mean posterior probability.) "The hazard rate ratio for group A over group B peaked at X at time T and remained greater than Y throughout" or "Group A had an X-fold higher relative risk of death at time T which remained greater than Y throughout"the mean posterior hazard rate for group A, divided by that for group B, peaked at a value of X at time T and remained greater than Y at all times up to 270 days.
"The probability that hazard rate for group A was greater than that of group B was 0.97 at time T" -the posterior probability was 0.97 that the hazard rate for group A at time T was greater than that for group B at time T (Note that this is not a reference to the mean posterior hazard rate.) "The hazard rate for group A was greater than that for group B at time T" -the mean posterior hazard rate for group A was greater than that for group B at time T.