Recent Developments in Tuberculous Meningitis Pathogenesis and Diagnostics [version 3; peer review: 3 approved]

The pathogenesis of Tuberculous meningitis (TBM) is poorly understood, but contemporary molecular biology technologies have allowed for recent improvements in our understanding of TBM. For instance, neutrophils appear to play a significant role in the immunopathogenesis of TBM, and either a paucity or an excess of inflammation can be detrimental in TBM. Further, severity of HIVassociated immunosuppression is an important determinant of inflammatory response; patients with the advanced immunosuppression (CD4+ T-cell count of <150 cells/μL) having higher CSF neutrophils, greater CSF cytokine concentrations and higher mortality than those with CD4+ T-cell counts > 150 cells/μL. Host genetics may also influence outcomes with LT4AH genotype predicting inflammatory phenotype, steroid responsiveness and survival in Vietnamese adults with TBM. Whist in Indonesia, CSF Open Peer Review


Introduction
The pathogenesis of Tuberculous meningitis (TBM) is poorly understood. Mechanisms by which Mycobacteria disseminate from lung to the brain, key factors driving a dysregulated host response, and the pathogen specific factors influencing presentation and severity, compared to other forms of TB, are not well described. In recent years application of contemporary molecular biology 'omics' techniques to clinical samples, greater availability of advanced neuroradiology, emphasis on immune-mediated contributions to pathology, and use of refined experimental models of TBM have better illuminated its pathogenesis. A better grasp of these processes may also lead to improved diagnostics, targeted immunotherapy as well as a biomarker-based, patient-specific approach to personalized treatment. Diagnosis has been traditionally insensitive (AFB smear) and slow (culture). This has improved with the addition of GeneXpert MTB/Rif (Xpert) which gave sensitivities similar to culture in 2 hours (versus 2-4 weeks with culture). Subsequently, GeneXpert MTB/Rif Ultra (Ultra), a re-engineered version, has shown better sensitivities than culture in some settings. Yet, none of these technologies has adequate negative predictive value to 'rule-out' TBM. In this article we review important recently published studies that have informed our current understanding of TBM pathogenesis and diagnostics. We do not seek to present a comprehensive review of the history of TBM pathogenesis and diagnostics as a number of detailed papers have addressed this recently [1][2][3] . Rather we provide a commentary of key studies published within the last 5 years and summarise knowledge gaps and future considerations to enable progress in the field.

TBM pathogenesis
Dissemination to the central nervous system Understanding of the microbial and immune processes that allow M. tuberculosis to disseminate from the respiratory epithelium to reach the meninges remains incomplete 2,4 . The foundations of what is known were laid through natural history and autopsy studies in the pre-chemotherapy era. The necessary steps to develop TBM include the pathogen surviving its initial encounter with the innate immune system at the respiratory epithelium and establishment of primary infection in the lung parenchyma with characteristic granulomatous inflammation 5-7 . Spread beyond the lungs likely occurs through the blood and may be preceded by local invasion to the lymphatic system. Donald and Schoeman have highlighted the possibility of coincident miliary TB in cases of TBM, particularly in young children, where tubercles of different sizes and ages have been described on the meninges and confirmed by magnetic resonance imaging (MRI) 8,9 . In children, miliary TB and TBM develop most often within 3 months of primary infection, when fresh anatomical changes are still found in the primary lung focus 10 . In addition to children, people living with HIV (PLWHIV) are another vulnerable group who may be unable to control the infection in the lungs and therefore at risk of coincident miliary TB and TBM secondary to haematogenous dissemination of M.tb 8,11 . The contemporaneous nature of TBM and miliary TB illustrates that the "Rich focus" model (of a single meningeal/sub-cortical granuloma rupturing years after initial haematogenous dissemination discharging acid-fast bacilli into the sub-arachnoid space) 12 does not apply to all TBM cases and there may be more than one pathway to the development of meningitis following M. tuberculosis infection.

Host immune response to TB infection in the CNS
The host immune response to TB bacilli in the sub-arachnoid space gives rise to a granulomatous inflammation predominantly affecting the basal meninges. Inflammatory exudates may obstruct the passage of cerebrospinal fluid (CSF), leading to hydrocephalus. Small and medium-sized intracerebral arteries can become inflamed and occluded, leading to cerebral infarcts. The majority of TBM pathology is believed to result from the host inflammatory response, which has been reviewed in depth elsewhere 2 ; several pro-and anti-inflammatory cytokines such as tumour necrosis factor-α (TNF-α), interferon-γ (IFN-γ), interleukin (IL) 1β, IL-6, IL-8, and IL-10 are shown to be induced in TBM 13,14 . Disequilibrium of pro-and anti-inflammatory cytokines influence the severity and course of TBM. Current understanding of key established mechanisms known to play a role in host immune response in TBM are summarised in Figure 1 2,15 .
In the recent literature, the long-standing belief that excessive inflammation is the cause of death in TBM was brought into question by a recent immunopathogenesis study in Vietnam. In HIV-negative adults, associations between death and both lower CSF cytokines concentrations and lower CSF leucocyte counts (median 59 × 10 3 cells/mL (IQR 13-240 × 10 3 cells/mL) in those who died versus 135 × 10 3 cells/mL (IQR, 48-298 × 10 3 cells/mL) in survivors) were noted 16 . These data support the notion that poor outcome from TBM, in the context of immunosuppressive treatment (adjunctive corticosteroids), is associated with an inadequate pretreatment inflammatory response in HIV-negative individuals. In a study of 120 Vietnamese adults with TBM included in a trial of adjunctive aspirin treatment, it was shown that there was an aspirin dose-dependent inhibition of thromboxane A 2 and upregulation of pro-resolving CSF protectins, resulting in potential reduction in new infarcts and deaths by day 60 of treatment in microbiologically confirmed TBM patients 17 . A further study investigated concentrations of host protective lipid mediators (specialized proresolving mediators, SPMs) in CSF. Prostaglandins and cysteinyl leukotrienes were found to be reduced in more severe cases, while the lipoxygenase 5-derived 13-series resolvin (RvT)2, RvT4, and 15-epi-lipoxin B4, were significantly increased in survivors. These data suggest SPMs may play an important role in TBM pathogenesis 18 . to seed the brain. The bacilli traverse the blood brain barrier (BBB) and blood cerebrospinal fluid barrier (BCSFB) through various virulence factors that enable the invasion of and migration through cerebral vascular endothelial cells, or are carried into the CNS by infected peripheral innate immune cells. B: In the CNS antigen recognition and internalization by microglia, neurons and astrocytes occurs, mediated by numerous host genetic factors. C: The resulting immune response stimulates the release of proinflammatory cytokines and chemokines and other immune mediators that contribute to the breakdown of the BBB and the influx of innate and adaptive immune cells from the periphery. D: A prolific inflammatory response ensues. The inflammatory exudate in the basal cisterns contributes to cerebral vascular pathology and the development of hydrocephalus and raised intracranial pressure. Vasogenic edema due to an influx of proteins through the leaky BBB, and cytotoxic edema as a result of cellular damage contribute to the raised pressure. The overall decrease in cerebral blood flow puts the brain at risk of ischemia, infarction and poor patient outcomes. In some cases the infection is controlled in discrete tuberculomas or abscesses, which may resolve with treatment and time.
(95% CI 1.03-1.10), respectively) were associated with mortality 19 . Flow-cytometry on blood in a subset of 160 HIV-negative adults with TBM showed lower αβT and γδT cells, NK cells and MAIT cells in TBM subjects compared to 26 pulmonary TB adults (2.4 to 4-fold, all p < 0.05) and 27 healthy controls (2.7-7.6-fold, p < 0.001), but higher neutrophils and classical monocytes (2.3 -3.0-fold, p < 0.001). CSF flow cytometry of TBM patients showed a predominance of αβT and NK cells, associated with better survival, as well as the presence of MAIT cells, previously undescribed in CSF 20 . Indonesian HIV-negative TBM patients showed a strong myeloid blood response and a remarkably broad lymphoid CSF response including innate lymphocytes, however there was little correlation between blood and CSF compartments 20 .
These recent studies in Vietnamese and Indonesian adults with TBM, aimed at gaining insights into mechanisms of the inflammatory response in disease pathogenesis, used novel and high-resolution methods to look at lipid mediator profiles and immune cell populations. Data indicated specific lipid mediator signatures and cell populations that are associated with disease severity before treatment and mortality; these should beconsidered for host-directed therapy of TBM.

Host genetic and metabolic factors
More efficient and cost-effective genomics platforms have enabled of late better understanding of variable host responses in TBM through the study of host genetics. Polymorphisms in CD43 encoding a surface glycoprotein involved in M.tb adhesion and proinflammatory cytokine induction and PKP3-SIGIRR-TMEM16J gene region encoding a negative regulator of TLR/IL-1R signalling have both been linked to survival in TBM 21,22 . However, the greatest interest has been around the role of leukotriene A4 hydrolase (LTA4H). LTA4H catalyzes the final step in the synthesis of leukotriene B4 (LTB4), a potent chemoattractant and pro-inflammatory eicosanoid. A common functional promoter variant rs17525495 in the LTA4H gene can predict survival and dexamethasone responsiveness in HIV-uninfected adults with TBM 16,23 . This human candidate gene association study was guided by findings in a zebra fish model where LTA4H was found to determine the balance of pro-inflammatory and anti-inflammatory eicosanoids in response to mycobacterial infection 16,24 . In a retrospective study in Vietnamese HIV-uninfected adults with TBM, while LTA4H rs17525495 TT and CC genotypes were both associated with susceptibility to mycobacterial infection, the associations involved opposing inflammatory states: high inflammation for the TT genotype and low inflammation for the CC genotype. CT genotype had an intermediate inflammatory response and were more likely to survive TBM. Dexamethasone treatment improved survival in TT genotype patients with hyperinflammatory response but was possibly harmful to CC patients with hypo-inflammatory response 23 . A later prospective study in Vietnam reported that in TBM HIV-uninfected adults, LTA4H genotype influences cytokine inflammatory response and correlates with TBM severity, Figure 2 16 . More importantly, this study confirmed that the LTA4H genotype determined corticosteroid responsiveness and survival. Interestingly, LTA4H genotype did not predict outcomes in Indonesian adults with TBM, but there was a trend towards improved survival with TT genotype compared to CC or CT genotype, Figure 2 19 . A clinical trial is currently underway in Vietnam (NCT03100786) to evaluate LTA4H genotype-directed corticosteroid therapy, an exciting example of personalised medicine in TBM 25 .
Although prior studies have considered sodium, glucose and lactate as related to TBM pathogenesis, recent developments in the application of targeted metabolomics have provided greater insight in the role of tryptophan, a potential key metabolite in TBM. This amino acid required for protein biosynthesis is a precursor to serotonin and melatonin (serotonin pathway) and kynurenine and quinolinic acid (kynurenine pathway). The latter is stimulated at the expense of the former by pro-inflammatory cytokine such as IL-6, TNF-alpha and IFN-gamma via indoleamine 2, 3-dioxygenase. In a recent study of serum and CSF metabolites, low levels of tryptophan were associated with survival 26 . One theory regarding this association could be the neuroprotective effects of the associated kynurenine pathway downstream metabolites. Either this pathway, or the 11 genetic foci related to CSF tryptophan metabolism could have novel clinical implications for TBM 26 .
HIV co-infection and immune reconstitution inflammatory syndrome HIV infection is a strong independent predictor of death from TBM (hazard ratio, 3.94; 95% confidence interval (CI), 2.79-5.56) 27 . The role of adjunctive corticosteroids in HIV-associated TBM is inconclusive (relative risk of death with adjunctive steroids, 0.78; 95% CI, 0.59 to 1.04; P=0.08) 28 and a randomized placebo-controlled trial is underway (NCT03092817) to address the use of steroids in HIV-associated TBM. Pathogenesis studies in PLWHIV are required to identify the unique pathogenic determinants of poor prognosis. Thuong et al. compared the pretreatment CSF cells and cytokine profiles of 764 HIV-positive and HIV-negative participants in Vietnamese TBM clinical trials. HIV-positive individuals had higher mean CSF neutrophil percentage (17% vs 5%; P < .0001) and global cytokine expression (aside from IL-10 which inhibits response to M. tuberculosis) than their HIV-negative counterparts. PLWHIV with CD4+ T-cell counts <150 cells/μL showed higher median CSF neutrophil percentage (25%), than those with a count ≥150 cells/μL (neutrophils 10%; P=0.021) and patients without HIV infection (neutrophils 5%; P<0.0001). Of patients with a CD4+ T-cell count of <150 cells/μL, 44% (105 of 238) died, compared with 13% (5 of 39) with a count of ≥150 cells/μL and 19% (83 of 439) without HIV infection 16 . These findings, amongst others, suggest a role for neutrophils in the immunopathogenesis of HIV-associated TBM 29 .
Marais et al. conducted longitudinal analyses of paired blood and CSF samples in South Africans with HIV-associated TBM, describing the relationships between the development of immune reconstitution inflammatory syndrome (IRIS) and CSF leucocytes, the concentrations of >30 blood and CSF inflammatory mediators, and blood transcriptional profiles. They found TBM-associated CNS IRIS to have an inflammatory signature characterized by neutrophil and inflammasome-mediated proinflammatory responses 30,31 . The neutrophil-dependent inflammatory activation could be detected in peripheral blood before the start of TB treatment and therefore has potential to predict who will develop IRIS.

Brain Injury Markers
The study of neurodegenerative-associated proteins to describe extent and type of brain injury post TBM has recently been explored through omics analysis, approaches which strive to understand genetic or molecular profiles of humans, particularly in paediatric TBM. In lumbar CSF of children with TBM, S100B and NSE (structural proteins of the CNS, and biomarkers of CNS tissue damage) at disease onset were associated with poor outcome, as was highest concentration overall and an increasing profile over time in S100B, NSE, and GFAP neuromarker concentrations increased over time in those who died (whilst inflammatory markers decreased), and were overall highest in those with cerebral infarction 32 . It is of interest that despite markers of inflammation reducing, proteins traditionally associated with neurodegenerative processes continued to rise.
In ventricular CSF of children with TBM, transcriptome analysis has revealed significant enrichment of transcripts associated with neuro-excititoxicity predominantly driven by glutamate release and NMDA binding and receptor uptake 33 . Upregulation of genes associated with nitric oxide, cytochrome c, brain injury proteins like myelin basic protein, and proteins including tau, amyloidbeta and apo-lipoprotein were also seen 33 ; many of which have also been described in in neurodegenerative conditions such as Alzhiemer's disease and traumatic brain injury 34 .
These findings raise the possibility of ongoing brain injury which in TBM seem to occur following ischaemic injury, despite resolving acute inflammation 32 . Further studies, including those which investigate the longer-term pathogenic processes in TBM are required to validate these results and understand further neurological sequelae including those which may indicate a post-infectious process in TBM.

Neuroimaging in pathogenesis studies
Technical advances and increasing availability of imaging modalities have recently enabled research in which imaging is used to assess pathogenic mechanisms in TBM in vivo in animal and human subjects. In a blood and CSF biomarker study of childhood TBM tuberculomas, magnetic resonance imaging has been used to note an association between tuberculomas and elevated interleukin (IL) 12p40, interferon-inducible protein 10, and monocyte chemoattractant protein 1 concentrations, whereas infarcts were associated with elevated TNF-α, macrophage inflammatory protein 1α, IL-6, and IL-8 32 . Specific sequences can also be used to describe morphology of structural damage and correlate this to meaningful clinical measures. For instance poorer Diffusion Tensor Imaging (DTI) parameters of white matter integrity in the anterior cingulate gyrus, parahippocampal gyrus and globus pallidus are associated with worse neuropsychological performance 35 . A further study by the same group used Diffeomorphic Anatomical Registration Through Exponentiated Lie Algebra (DARTEL) voxel-based morphometry (VBM) to assess the integrity of grey matter in these same TBM patients 36 . Patients with TBM performed significantly poorer on the digit symbol, similarities, block design, matrix reasoning, and letter-number sequencing subtests of the Wechsler Adult Intelligence Scale compared to healthy adults. These changes correlated with smaller grey matter volumes in the right thalamus, right superior temporal gyrus, right precuneus, right middle temporal gyrus, left putamen, right caudate nucleus, and right middle temporal gyrus 36 . These studies suggest that structural damage can be cortical as well as subcortical which may in turn be related to degree of long-term impairment. This has implications for understanding long term outcomes particularly neurocognitive impairment in TBM, which in light of these findings may share features with other forms of dementia (including vascular and HIV associated neurocognitive impairment) where a subcortical pattern of neurocognitive impairment (including frontal and executive functions) can be observed.
A rabbit model study of childhood TBM, utilized ionized calcium binding adapter molecule (Iba-1) to approximate microglial activation with flurodeoxyglucose-positron emission tomography (FDG-PET) and demonstrated the presence of activated microglia and macrophages localized to TB lesions 37 . In humans, case reports and a prospective study have advocated the use of FDG-PET as a diagnostic tool, as it has been effective in detecting extra-cranial evidence supportive of a TBM diagnosis 38-40 . The role of FDG-PET in unravelling time course of inflammation in TBM remains to be seen, although it has played a role in understanding Alzheimer's, a disease in which, similar to TBM, inflammation plays a key pathogenic role 41,42 .
Pathogen factors: bacillary load, pathogen strain and virulence factors TBM patients generally have low bacterial loads in CSF which causes difficulties in both diagnosis and ability to study bacterial load evolution-related pathophysiology. The time-to-positivity of a culture and cycle threshold (Ct) of nucleic acid amplification tests such as GeneXpert MTB/Rif (Xpert) can provide an indication of likely bacterial burden 43 . Over 50% of diagnosed cases are microbiologically undetectable and defined as 'probable' or 'possible' TBM which obviously limits this approach 44 . Marais et al. showed that in patients where M.tb was cultured from CSF taken before and after two weeks of antituberculosis treatment, there was a 9.3-fold increased risk of subsequently developing TBM-IRIS, although the sample size is small with 15 TBM-IRIS patients compared with 6 non-TBM-IRIS patients 45 . Thuong et al. found that among 692 Vietnamese adults with TBM, pre-treatment CSF M.tb load (by Xpert Ct) was correlated with increased CSF neutrophil counts, increased cytokine production, and new neurological events after treatment initiation, but not death 43 .
In addition, epidemiological trends of M.tb lineage from TBM (n=73) and pulmonary TB (n=220) patients in Thailand showed that the Indo-Oceanic lineage is more frequently found in TBM patients (41% versus 13% in PTB) 46 . This association did not hold true in Indonesia, though specific genetic variations were identified which were associated with TB phenotype, including one (Rv0218) whose encoded protein may play a role in host-pathogen interaction 47 .

Host-pathogen interactions
It is estimated that the global burden of latent TB infection (LTBI) is approximately 23.0% (95% CI 20.4%-26.4%), amounting to approximately 1.7 billion people 48 . Innate immune responses are critical to control TB infection yet also contribute to tissue damage. This delicate balance is illustrated in the damage response framework which provides a theory of microbial pathogenesis that incorporates the contributions of both host and microbe to host damage that stems from host-microbe interaction 49,50 . This framework likely applies to TBM based on evidence of both failed immunity and excessive inflammation being linked to increased TBM pathology, see Figure 3 23,51,52 . Both the microbe and the host contribute to host damage and where an individual patient's immune response lies on the continuum of the damage response framework parabola determines the nature of the disease process 16,43,53 . Evidence from recent studies shows LTA4H genotype, CSF cytokines and CSF immune cells such as neutrophils are determinants of inflammatory state, which impacts both bacterial growth and host damage and thus leads to different outcomes. The current one-size-fits-all approach to TBM treatment fails to recognize divergent pathologies and may explain the poor outcomes in certain populations. Being able to identify where on the parabola an individual lies and tailoring therapy to achieve the optimal milieu is an approach that warrants further investigation. LTA4H genotype is an example of using host genotype to predict inflammatory response and to tailor treatment by host directed therapy. Omics technology are now being used to identify additional host genetic markers and treatment targets in TBM. On the left side of the parabola, shaded in blue, the immune system fails to limit mycobacterial growth and invasion which results in host damage. On the right side, shaded in red, the immune response is excessive and the resultant inflammation and host-damage. The proportion of the parabola lying below the black line represents disease latency, which is not associated with clinically evident host damage. On the blue side therapeutic interventions could be targeted at stimulating an immune response, whilst on the red side therapeutic interventions could aim to dampen immune response.

TBM diagnostics
The field of TBM diagnostics has evolved rapidly in recent years with both complex and low-tech assays being explored in a variety of populations. Whilst progress has been made, no single assay can be used as a rule out test. The characteristics of CSF tests studied to date are summarized in Table 1.

Host-based diagnostic biomarkers
Traditional diagnostic techniques for TBM include CSF smear microscopy for acid fast bacilli (rapid and cheap but insensitive in most settings, 10-15%) and CSF culture (improved sensitivity of 50-60% but results in 2-6 weeks with a biosafety lab level three requirement) 3 . Given the limitations of traditional, Interferon gamma release assays (IGRAs) are commonly used to infer LTBI. A 2016 meta-analysis of six studies performing CSF IGRA's found a pooled sensitivity and specificity of 77% (95% CI 69%-84%) and 88% (95% CI 74%-95%), respectively, for TB meningitis, though reference standards varied by study 57 . Limitations of IGRA include high cost, the need for advanced lab infrastructure, frequent "indeterminate" results, and false positives associated with other causes of meningitis. Additional host biomarkers including delta-like ligand 1, vitamin D binding protein, and fetuin have been evaluated in CSF though none were found to have satisfactory performance 58 . Numerous CSF antibodies to M.tb in CSF has also been evaluated. Huang and colleagues found pooled sensitivities of 91% (95% CI 71-98%) for anti-M37Ra across five studies, 84% (95% CI 71-92%) for anti-antigen-5 across eight studies, and 84% (95% CI 71-92%) across 12 studies for anti-M37Rv, again using a variety of reference standards (making the pooled estimates somewhat flawed) 59 . Use of blood antibody assays are discouraged for the diagnosis of TB, and their utility in CSF is limited by heterogeneity and the lack of a uniform reference standard across research studies as well as a lack of commercial assays.
Though on the surface, many of these markers look to have promise, uptake has been limited. Heterogeneity in study design and widely variable study performance has limited the consensus regarding the utilities for most host-based tests. Further, many of these tests require sophisticated laboratory infrastructure, are costly, and in some cases are not commercially available. None of these CSF tests are routinely used and as far as the authors are aware, none are actively being studied further.
In the field of pulmonary tuberculosis, host RNA transcriptomic signatures have been leveraged to predict incipient and active tuberculosis with moderate short-term (<3 month) sensitivity (41-81%) 61

Biomarkers in children
The often-dismal outcome of TBM is contributed to by delayed diagnosis and/or initiation of treatment, especially in high burden settings 4 . Currently available diagnostic test performance is especially poor in young children with TBM. Thus, diagnosis of childhood TBM is mostly based on a combination of clinical findings, CSF analysis and radiological findings 63 . Even so, there are often multiple missed opportunities prior to a diagnosis of childhood TBM 64 . Since it can be challenging to identify bacilli in paediatric extrapulmonary TB, the use of host or pathogen biomarkers to aid diagnosis is being explored. Host biomarker-based tests have shown promise in extrapulmonary TB outside of the CNS and therefore have potential applications in TBM 65 . Recent technological advances have made it possible to screen for many biomarkers in as little as 3 μl of sample using the Luminex multiplex cytokine beaded arrays, albeit in research context currently, rather than routine clinical practice. Limitations of the studies assessing CSF host biomarkers in childhood TBM include relatively small sample size, and therefore few children with confirmed TBM, confirmed meningitis due to other pathogens, and confirmed HIV co-infection. External validation is a necessity in order to generalize the clinical usefulness of the prediction model in an independent group of patients. the potential of CSF-based biosignatures, a further limitation is the invasive nature of CSF collection, and blood or urine-based inflammatory biosignatures require exploration. In a study evaluating serum biomarkers, the combination of CRP, IFN-γ, IP-10, CFH, Apo-A1 and SAA showed moderate diagnostic accuracy for clinically-defined TBM, including both 'definite' and 'probable' TBM (AUC of 0.75, sensitivity of 69.6% and specificity of 62.5%). A three-biomarker combination of adipsin, Aβ42 and IL-10 showed improved accuracy (AUC of 0.84, sensitivity of 82.6% and specificity of 75.0%). Cut-off values for CRP, IFN-γ, IP-10, CFH, Apo-A1, SAA, adipsin, Aβ42 and IL-10 were >80721.0 ng/ml, <61.5 pg/ml, <57.2 pg/ml, >350185.0 ng/ml, >287512.0 ng/ml, >59894.0 ng/ml, <2393.0 ng/ml, <278.4 pg/ml and <7.0 pg/ml, respectively. Although sample size was small, these biomarkers warrant further exploration 67 .

Pathogen-based diagnostics
The absence of a perfect gold standard for use in TBM diagnostic studies means that the results must be interpreted with an awareness of the pros and cons of the reference standard used.
The 2010 uniform TBM case definition which defines cases as 'definite', 'probable', 'possible' or 'not TBM' is the most standardised tool to use when defining a case definition 63 . This case definition was derived by expert consensus rather than being data-driven and, although designed to be applicable to any age, HIV infection status or geographical setting, may perform better in some contexts than others. In HIV-negative populations a reference standard of 'definite, probable or possible' is often used, however in PLWHIV including 'possible' in the reference standard can be imprecise due to the wide variety of infectious and non-infectious aetiologies that can fall into this category. We do strongly advocate the use of the case definition to standardise results, allow for greater comparison between studies and meta-analysis of data; use of other standards must be interpreted with a degree of caution.

Nucleic-acid amplification tests
To address the limitations of conventional microscopy and culture techniques, NAATs have emerged as important tools for rapid and accurate diagnosis of TBM 44 . A recent meta-analysis evaluating NAATs in TBM reported heterogeneity in results with a pooled sensitivity of 82% against culture and 68% against a clinical reference standard 68 . This variability, especially in in-house NAATs, is subject to difference in volume of sample, method of extraction, choice of targets used, presence of inhibitors in the sample and lack of optimal reference standard. Traditional NAATs require expensive equipment, stringent operational conditions and technical expertise limiting their use in routine clinical practice in lower-resource, high endemic settings.
To circumvent these challenges, loop mediated isothermal amplification (LAMP) assays were developed and can be conveniently carried out under isothermal conditions in an ordinary laboratory water bath or heating block within one hour. Though LAMP has outperformed PCR in an Indian study on TBM 69 , the assay is still in its infancy and needs further validation. Another method to potentially reduce the overall cost of NAAT would be to utilize magnetic bead assay technology, thus obviating the need of gel electrophoresis system or expensive dyes.
Xpert is a rapid (90 min run-time) fully-automated cartridgebased real-time PCR assay that detects the presence of M.tb complex DNA, as well as rpoB gene mutations responsible for rifampicin resistance. The pooled sensitivity and specificity of Xpert against culture in 33 studies on TBM, was 71.1% and 98%, respectively 70 . Xpert has been shown to significantly increase microbiological confirmation of TBM in Uganda over a 6.5-year period but its impact on clinical outcomes in unknown 71 . Individual studies have also found inferior performance for Xpert compared to multiplex PCR 72 or Amplicor assay 73 in diagnosing TBM although these results have not been confirmed. The next generation, GeneXpert MTB/Rif Ultra (Ultra) has an 8-fold lower limit of detection than Xpert (16 CFU/ml versus 113 CFU/ml) attributable to a larger chamber allowing double the volume of sample to reach the PCR reaction and two additional DNA probes (IS1081 and IS6110) 74  Another commercial NAAT, the MTBDRplus assay, has been evaluated only in few cases of TBM and needs further validation 80 . Accurate and rapid detection of drug resistance is another challenge, rifampicin resistance detection by Xpert has imperfect sensitivity (93%) and where detected and ideally requires confirmation by sequencing or culture 72,81 . Ultra uses melt curve analysis to improve detection of rifampicin resistance but both are about 95% sensitive 82,83 . Ultra will not be able to adequately define rifampin resistance in samples with a low quantity of bacilli (trace category positive) 75 . In summary NAATs, are a major diagnostic advance but they cannot yet fully replace culture methods. Ultra is too insensitive to rule out TBM, and like Xpert, should be considered as the first test and not the last in TBM diagnosis 84 . Ultra is an important step in the right direction but the result should be considered in the context of the clinical probability of TBM 85 .

CRISPR-MTB and metagenomic next generation sequencing.
Clustered regularly interspaced palindromic repeat (CRISPR) associated proteins (Cas) have the ability to cleave DNA at specific sites and are being used widely in gene-editing and more recently in infectious disease diagnostics. When combined with DNA amplification, the CRISPR system can detect nucleic acid molecules at extremely low abundance. There is one recent report of utilizing the CRISPR system for detection of M.tb (CRISPR-MTB). The study included 26 CSF specimens and found CRISPR-MTB to have a sensitivity of 73% compared to 54% for Xpert and 23% for culture against a reference standard of 'clinical TBM' as determined by the physician based on clinical presentation imaging and response to TB therapy. The specificity of the test was 98% when tested against 63 non-TB cases. CRISPR-MTB is isothermal and can be performed in under 2 hours using only 500 μl of CSF. CRISPR-MTB remains to be tested against Ultra and requires a higher level of laboratory expertise, resources, and time than the Xpert platform but may be an advance in TB diagnostics if these findings can be confirmed in other settings with more standardized reference standards 86 .
Metagenomic next generation sequencing (mNGS) is a rapidly developing technology that has proved useful in determining aetiologies for CNS infections that have evaded detection by conventional techniques. Further, mNGS, as opposed to organism-specific molecular tests has the ability to detect any low abundance infection with a single test 87 . A recent small study applied mNGS to stored CSF samples from 23 TBM cases and found a sensitivity of 67% (8/12) against a reference standard of definite TBM, higher than AFB stain (33%, 4/12), PCR (25%, 3/12) and culture (8%, 1/12) 88 . Paucibacillary conditions such as TBM where the bacillary load may fall below the LOD of commercial NAATs, or where mutations exist around specific PCR primer binding sites may find particular use for mNGS. Targeted enrichment of low abundance genes with Finding Low Abundance Sequences by Hybridization (FLASH), a novel CRISPR-Cas9 technology can increase DNA read abundance by up to 10 5 -fold before sequencing occurs 89 . Combining FLASH and mNGS technologies could improve detection of TB DNA and associated antimicrobial resistance mutations mutations 90 . A first pilot of FLASH technology in TB demonstrated up to a 100-fold increase in TB read abundance, detection of 6/6 cases of TBM positive with Ultra and detection of an additional case of TBM that had been missed by Xpert, Ultra and MGIT culture 91 . Here again, large studies need to be performed to better understand this technology's performance and the cost, laboratory infrastructure, and degree of expertise will need to be improved upon to permit widespread usage.

Pathogen-based biomarkers.
A urine lateral flow assay (LFA) that detects M.tb lipoarabinomannan (TB-LAM), a 17 kDa glycolipid found in the outer cell wall of MTB, has recently been recommended by the World Health Organization for the diagnosis of HIV-associated TB in HIV-positive inpatients (Alere Determine TB-LAM, Abbott, Chicago, USA). The unique characteristic of the test is that its sensitivity increases as CD4 T-cell count falls, with a sensitivity of 56% in those with CD4 <100 cells/ml 60 . Yet, in CSF, despite some initial optimism related to an autopsy-based study in Uganda, the Alere TB-LAM has shown poor sensitivity on lumbar CSF in Uganda 92,93 , along with a larger Zambian study which examined culture positive TBM in Zambia (TB-LAM sensitivity 22% (23/105)) 92,94,95 . The Alere TB-LAM is also limited by its susceptibility to individual reader interpretation of the darkness of the test line compared to the reference card ( Figure 4). A novel LAM assay (Fujifilm SILVAMP TB-LAM, Fujifilm, Japan) is able to detect concentration of LAM at approximately 30-fold lower than Alere TB-LAM due to design differences, including a silver amplification step and gives a result in one hour 96

Clinical prediction rules
Work is underway to develop a more accurate multivariable clinical prediction rule derived from large international cohorts using individual patient data 98 . The hope is that a data-driven scoring system will be developed for use in a range of clinical settings by using common, readily available clinical or laboratory parameters to aide in clinical decision making.

Discussion
In the last five years, the pathogenesis of TBM has been better elucidated, in part thanks to detailed immunological studies on clinical samples preemptively stored during clinical trials. These advances highlight the importance of collecting and storing samples appropriately for future research to maximize scientific outputs, as highlighted in the paper on sampling strategies in this collection. HIV infection is a major predictor of mortality in TBM and advanced HIV infection (CD4 T cell count <150 cell/μl) appears to drive a dysregulated, hyperinflammatory phenotype with very poor outcomes. In HIV-endemic sub-Saharan African settings around 90% of all adult TBM occurs amongst HIV-positive individuals 71,95 , often with either untreated advanced HIV or having recently initiated ART -both driving a hyperinflammatory response. In Vietnamese adults with TBM, LTA4H genotype is a strong predictor of mortality though this finding was not duplicated in Indonesia.
Recent insights have shown that neutrophils play a significant role in the immunopathogenesis of TBM, and that both a paucity and an excess of inflammation can be equally damaging in TBM. It has become increasingly clear that a 'one-size-fits-all' approach is too simplistic in TBM treatment, as in other infections such as pulmonary TB and sepsis 99,100 . The damage-response framework may provide a useful structure for understanding host-pathogen interactions in TBM, illustrating how immune response could be exploited for therapeutic purposes. Additional anti-inflammatory therapy with aspirin 17,101 or more targeted immunotherapy could have a role in persons with an excessive inflammatory response; whilst individuals with an inadequate response might do better without corticosteroid treatment or might even benefit from immunomodulating therapy to boost their immune response 102 . Future trials of novel specific hostdirected therapies are needed and must include immune markers to allow for post-hoc identification of subgroups benefitting from the initiated therapy. Because of the lack of correlation between blood and CSF compartments we advocate inclusion of both blood and CSF markers when studying adjuvant therapies.
The field of TBM diagnosis is rapidly evolving with GeneXpert MTB/Rif Ultra being the most promising test to date for diagnosis of TBM. Ultra is rapid and has potential to confirm more cases of TBM at lower bacillary loads, though whether this will improve outcomes remains to be determined. Most importantly, Ultra does not appear to have adequate predictive value to 'rule-out' TBM and so it cannot meet the potential of an ideal TBM diagnostic test to avoid long, toxic TBM therapy in persons without TBM. Novel sequencing technologies hold potential to provide increased understanding of pathogen genomics and behavior and further illuminate host response, which may in turn lead to novel diagnostic and therapeutic targets. Sequencing technologies are increasingly available in TB endemic settings but will need further improvements in affordability and speed in addition to more data on accuracy to unlock their potential as diagnostic tools for TBM. It is now a realistic hope that a test (or set of tests) will one day be available that will be able to confirm or rule out TBM, provide M.tb resistance information, and direct clinicians to targeted, adjunctive host-directed therapy within hours.
Utilising host biomarkers for TBM diagnostics are an exciting development. In the section on 'Biomarkers in Children', the authors describe the results of two studies (by the same group) using different combinations of host CSF biomarkers among children [15,64]. The authors should state what type of host markers (i.e. a mixture of cytokine and soluble mediators) were measured in the studies (using ELISA and Luminex platform). In the follow-up study [64], a relatively small number of patients (n=47) were tested in un-blinded study design. The STARD diagram shows a series of markers (n=69) were measured before the samples were then split into two clinical groups (TBM [definite/probable] n=23 or non-TBM n=24) and later the accuracy of the best combination of markers reported. Given the small size and unblinded design, the results of these studies (particularly the latter) could be summarized further and technical study limitations stated (beyond using CSF).
In the section on TBM diagnostics-many different types of host biomarkers are now being measured and their diagnostic potential beginning to be assessed in TB, including host transcripts. Many thanks again for taking the time to improve the manuscript through these comments.
There were multiple minor grammatical errors that should be corrected: P 3. Second to last sentence in the Introduction -"that" should be removed.
couple of lines about CSF FujiLAM following a recent publication relating to this assay in TBM.
In response to minor comments: 1. We have edited the text to reflect the fact that there may be more than one route to TBM disease. 2. We have removed this text about the points relating to the reference standard in the interest of brevity.
3. In the Ugandan study MGIT culture and Xpert were used alongside Ultra, this detail has been added to the paper. In the CRISPR study "clinical TB" was as determined by the physician based on clinical presentation imaging and response to TB therapy. This has been added to the paper.
We have rectified the multiple typos spotted.
Many thanks for your suggestions which have improved the manuscript.

Is the review written in accessible language? Yes
Are the conclusions drawn appropriate in the context of the current research literature? Yes authors have listed a series of findings from separate immunological studies but do not introduce the concept fully to the reader and do not draw the findings together at the end of each section to provide an overall interpretation or explanation for the reader. Some further explanation around these concepts would significantly improve the review.
explanation for the association between low levels and better survival.

TBM pathogenesis -HIV co-infection and IRIS
In the first paragraph (line 9, copied below), a comparison in neutrophil counts and mortality is presented between patients with a CD4 count >150, CD4 <150 and HIV negative.
"PLWHIV with CD4+ T-cell counts <150 cells/μL showed higher median CSF neutrophil percentage (25%), cytokine concentrations and 9-month mortality (44%) than those with a CD4+ T-cell count ≥150 cells/μL (neutrophils 10%; P=.021, mortality 13%) and patients without HIV infection (neutrophils 5%; P<.0001, mortality 18%). These findings, amongst others, suggest a role for neutrophils in the immunopathogenesis of HIV-associated TBM." However, there are only two p values included and it is not clear whether they refer to neutrophils or mortality. Could the authors please alter this to make it a little clearer (and if appropriate add the two additional p-values).
○ TBM pathogenesis -Pathogenesis of TBM in childhood I don't think this section in its current form particularly adds to the review and there is no new data. I would suggest this section is either expanded (to further discuss the differences between adult and paediatric patients with TBM and the pathological processes underlying these differences) or dropped. If the authors do decide to drop it they could consider amalgamating the statements about Donald and Schoeman's TBM and milary TB findings into the earlier section: "Dissemination to the central nervous system".

TBM pathogenesis -Brain injury markers
This paragraph would also benefit from further explanation so that the reader can appreciate the importance of the findings. Further detail about the brain injury markers S100B and NSE would be helpful (particularly whether they are used in other neurological conditions). Further text is also required to explain what the CSF transcriptomic signature means.

TBM pathogenesis -Neuroimaging in pathogenesis studies
Is there a comma missing here?: "anterior cingulate gyrus parahippocampal gyrus and globus pallidus".

○
The novel imaging studies reporting structural damage in cortical and subcortical areas are interesting. Perhaps the authors could expand slightly to suggest the potential implications ○ of this?

TBM pathogenesis -Host pathogen interactions.
This section discusses the damage response framework concept and applies it to TBM. I am not sure why the first sentence includes data on global latent TB prevalence -it doesn't seem directly relevant to this section.

○
It may improve the review if the authors expanded this section slightly and used it as a summary to draw together all the immunological findings discussed above. Particularly to draw attention to the importance of neutrophils.

TBM diagnostics
This section provides a good summary of host-based and pathogen-based diagnostics and reads well.

○
With regard to the host-based biomarkers, some of the pooled sensitivities appear promising but have drawbacks. It would be useful if the authors could provide a concluding comment at the end of the section discussing whether there is any future in these modalities (e.g. are any of those mentioned being taken forward into further trials? Are there any plans to introduce any of these tests? We have not added further information on tryptophan in the interest of word limit as there is already a full paragraph on this topic ○ The section on pathogenesis of TBM in childhood has been removed as suggested. Donald and Schoeman's TBM and miliary TB findings are now included in the "Dissemination to the central nervous system" section ○ We have added a summary statement on TBM biomarkers explaining that despite extensive research on this topic there are no currently biomarkers with adequate performance to be commercially viable in TBM diagnostics.

○
Typos and grammatical errors have been corrected. Many thanks! ○ We are very grateful for Dr Scriven's expertise and hope he is satisfied with the updated article.
Competing Interests: No competing interests were disclosed.