Knowledge gaps and research priorities in tuberculous meningitis

Tuberculous meningitis (TBM) is the most severe and disabling form of tuberculosis (TB), accounting for around 1-5% of the global TB caseload, with mortality of approximately 20% in children and up to 60% in persons co-infected with human immunodeficiency virus even in those treated. Relatively few centres of excellence in TBM research exist and the field would therefore benefit from greater co-ordination, advocacy, collaboration and early data sharing. To this end, in 2009, 2015 and 2019 we convened the TBM International Research Consortium, bringing together approximately 50 researchers from five continents. The most recent meeting took place on 1 st and 2 nd March 2019 in Lucknow, India. During the meeting, researchers and clinicians presented updates in their areas of expertise, and additionally presented on the knowledge gaps and research priorities in that field. Discussion during the meeting was followed by the development, by a core writing group, of a synthesis of knowledge gaps and research priorities within seven domains, namely epidemiology, pathogenesis, diagnosis, antimicrobial therapy, host-directed therapy, critical care and implementation science. These were circulated to the whole consortium for written input and feedback. Further cycles of discussion between the writing group took place to arrive at a consensus series of priorities. This article summarises the consensus reached by the consortium concerning the unmet needs and priorities for future research for this neglected and often fatal disease.


Introduction
Tuberculous meningitis (TBM) is caused when Mycobacterium tuberculosis enters the cerebrospinal fluid (CSF), leading to inflammation of the meninges, and subsequent clinical features of meningitis 1 . Cerebral ischaemia, hydrocephalus, raised intracranial pressure, and cranial nerve dysfunction commonly occur. TBM is the most severe form of tuberculosis (TB) and is universally fatal if untreated. Even with treatment, mortality is high, and some degree of morbidity is near universal. Our current understanding of the epidemiology, natural history and pathophysiology of TBM is limited and it will only be through an improved appreciation of these areas that appropriate clinical and public health interventions will be possible. Current diagnostic tools have inadequate test performance, meaning that most individuals with TBM are either not diagnosed or diagnosed late, once significant cerebral damage has occurred. Current antimicrobial treatment is based mainly on the treatment of pulmonary TB with little consideration to mode of drug action or CSF penetration. Given that most damage caused by TBM is due to host inflammation, it is anticipated that host-directed therapies should play an important role in modulating the damaging immune-medicated effects. However, other than corticosteroids, no agent has been demonstrated to be effective in reducing mortality and corticosteroids have not been shown to improve long-term morbidity 2 . The critical care management of TBM is also poorly understood, with highly divergent practice in the management of raised intracranial pressure, hydrocephalus, blood pressure, oxygenation and hyponatraemia 3 . Understanding the optimal management of these areas would almost certainly improve the long-term outcome of diagnosed patients. Finally, the management of TBM within health systems is sub-optimal with services and tools rarely provided at the locations where patients present. Public sensitisation is infrequently undertaken, and clinical training is often limited.
Globally, there are relatively few researchers of TBM and only a small number of centres of excellence. The field, therefore, would benefit from greater co-ordination, collaboration and data sharing. In 2009, the TBM International Research Consortium was formed and met for the first time in Cape Town, South Africa. Subsequent meetings took place in 2015 in Dalat, Vietnam, and in 2019 in Lucknow, India. The Consortium brings together approximately 50 experts from five continents. At the Lucknow meeting, the Consortium felt that it would be useful to develop a document detailing research priorities in TBM. The practice of establishing knowledge gaps and priorities for research helps researchers to focus efforts on areas that are perceived to be most likely to lead to clinical benefit. It also allows funders to target their funding. This practice can also be beneficial for advocacy through a clearer message and co-ordinated agenda. In this article we describe the process of developing a TBM research agenda and outline the priorities that were developed.

Identifying knowledge gaps and research priorities for TBM
The 3 rd TBM International Research Consortium took place in Lucknow on the 1 st and 2 nd March 2019. During the meeting, research updates were presented by leading researchers and clinicians in each respective field. Each had been requested to additionally present their thoughts and suggestions for knowledge gaps and research priorities in that field. Following each presentation, debate around these gaps and priorities took place. After the meeting, these priorities were assimilated into tables of research gaps by a core writing group, with gaps grouped into the themes of epidemiology, pathogenesis, diagnosis, antimicrobial therapy, host-directed therapy, critical care and care cascades. For each knowledge gap, study designs were proposed that might address them. These tables were circulated to the whole consortium who were asked to provide input and make suggestions as to which areas were the highest priority. Through further cycles of discussion within the writing group, the tables of knowledge gaps were updated ( Table 1-Table 7) and a consolidated list of top priorities for TBM research was developed, which are presented in Table 8. This article describes the knowledge gaps and research priorities for TBM.

Epidemiology
Accurate estimates of the incidence of TBM are lacking, in part as many cases die prior to diagnosis. Adult autopsy studies demonstrate that a high proportion of those with human immunodeficiency virus (HIV), in a high TB-burden setting, die from undiagnosed disseminated TB 4 . Given that M. tuberculosis is now the most common cause of bacterial meningitis in some high TB burden settings 5 , following wide implementation of routine vaccinations to meningococcus, pneumococcus and haemophilus, it is likely that this situation is replicated in children.
The World Health Organization (WHO) estimates that 10 million individuals developed TB in 2017 6 . Of the 6.4 million cases that were notified to WHO that year, 14% were classified as extrapulmonary 6 . Given that a patient with any pulmonary involvement is classified by WHO as a pulmonary case (such as an individual with both pulmonary TB and TBM) 7 , the proportion of patients who have extrapulmonary involvement is higher than this. To be notified, an individual must have been diagnosed and so it is likely that TBM is under-represented in cases notified to WHO.
Data on the proportion of all TB cases that have extrapulmonary involvement is limited and likely varies with multiple factors including the age and gender characteristics of the population, the HIV prevalence, where the cohort is recruited from (hospital vs. community), and the force of TB infection. The proportion of all TB patients that have TBM is even less unclear. A population-based estimate in a low TB-prevalence setting suggested that around 1% of TB cases were TBM 8 , while a paediatric hospital-based cohort of confirmed TB in a high TB-burden setting suggested that over 10% of TB cases were TBM 9 . If 1% of the overall TB burden is TBM then 100,000 individuals develop TBM each year. This figure could be significantly higher. TBM is universally fatal if untreated 10 but estimates of mortality in treated TBM are poor. A fifth of children die, with half of survivors experiencing neurodisability 11 . In some cohorts of HIV-infected adults, mortality is as high as 60% 12,13 . To improve estimates, it would be informative to review the extensive pre-chemotherapy literature to both quantify the proportion of meningitis deaths that are What is the cost of TBM?
• Collection of costing data in different settings to determine the cost of diagnosis and treatment of TBM, as well as the cost of caring for an individual who is disabled by TBM from a health system and societal perspective • Health economic evaluations to determine the health system and family costs of TBM diagnosis, treatment, mortality and disability in terms of life years, DALYs and QALYs lost due to TBM

Individual risk factors for developing TBM
Which adults and children are most likely to develop TBM following TB exposure and infection?
• Systematic reviews of household contact studies to identify risk factors in household contacts of developing both TB and TBM • Cohort studies of individuals who have been exposed to TB and followed to determine who progresses to TBM and evaluates risk factors for progression • Case control studies of people who had been exposed to TB who developed TBM, compared to TB-exposed individuals who developed pulmonary TB or no TB. • Cross-sectional study comparing TBM cases with pulmonary TB cases, to evaluate which patients with TB are at increased risk of TBM  A. In vitro models including isolated cell populations, blood brain barrier models and cerebral organoids.
B. In vivo models including zebrafish, mice, rabbits and non-human primates.
C. Correlates of protection studies in humans.
D. Observational, diagnostic and randomised intervention studies in humans. Is metagenomic sequencing a feasible and accurate means to detect M. tuberculosis in CSF?

Host-response based
Can single or multiplexed combinations of hostderived molecules adequately rule out TBM when applied to CSF, serum of peripheral blood?
• Capitalise on interest in the development of point of care tests for all forms of tuberculosis and specifically apply to TBM • Adequately designed and powered STARD compliant multicentre diagnostic evaluations that encompass important subgroups including children, HIV co-infected persons and those with drug resistant disease.
• Biobanking of well-characterised samples to aid discovery and evaluation of novel diagnostic modalities

Clinical algorithms
Do clinical algorithms have useful diagnostic accuracy when evaluated against microbiological and composite endpoints?
• Adequately designed and powered STARD compliant multicentre diagnostic evaluations that encompass important subgroups including children, HIV co-infected persons and those with drug-resistant disease. Do clinical algorithms positively or negatively influence patient outcome?

Diagnostic Strategies
How can clinical, pathogen and host-response diagnostic approaches be combined into a diagnostic strategy?  What is the optimal duration and regimen for tuberculomas and spinal TB? D

Study design
What is the best way to find optimal treatment regimens for TBM? B, D TBM, tuberculous meningitis; HIV, human immunodeficiency virus; TB, tuberculosis.
A. In vitro models including isolated cell populations, blood brain barrier models and cerebral organoids.
B. In vivo models including zebrafish, mice, rabbits and non-human primates.
C. Correlates of protection studies in humans.

D. Observational, diagnostic and randomised intervention studies in humans.
caused by M. tuberculosis as well as to define what proportion of all TB cases are TBM and how this varies by age. It would also be useful to carry out contemporary studies to document the proportion of TB cases that are TBM in large cohorts of routine data or research cohorts, and document outcomes in those treated. Mathematical modelling could be used to estimate the incidence of TBM, as well as mortality.
Improved estimates of incidence and mortality would be vital for appropriate advocacy and resource planning. It would also be important to quantify morbidity in TBM survivors and the cost impact of TBM to society. TBM commonly causes profound neurodisability in survivors, requiring ongoing medical care and extensive community/family support. The degree of long-term disability in survivors has been poorly quantified, especially the impact on education, employment and quality of life. Better estimates of morbidity and cost to society could be developed through collection of appropriate qualitative, neurodevelopmental and costing data, combined with modelling studies.
Beyond burden estimates, it would be useful to better understand why some individuals develop TBM and characteristics that influence clinical outcome. These could be elucidated through systematic review of published studies, as well as larger, prospective clinical cohorts. Knowledge gaps for TBM epidemiology are shown in Table 1.

Pathogenesis
The pathogenesis of TBM has recently been reviewed in depth 14 .
The classic paradigm is that meningitis arises as a consequence of bloodborne dissemination, with a single breach of the blood brain barrier being sufficient to form a central nervous system (CNS) focus 15 , although this has been challenged 16 . Whether bacilli predominantly transit the endothelium directly or within infected cells is incompletely understood, with a recent study suggesting both are possible 17 . This work was performed using Mycobacterium marinum in zebrafish, an ostensibly attractive system due to genetic tractability, high-throughput and the transparency of larvae, which allows in vivo imaging. Determinants of Zika virus host tropism by deep mutational scanning in cerebral organoids have recently been described 18 ; this may also be a tractable system to study TBM. Attempts to develop murine models of TBM to study pathogenesis in vivo have been described 19 , but have not been widely adopted, as mice incompletely replicate pathological features of TB, in particular caseous necrosis. The most commonly used animal model therefore has been the rabbit, which does recapitulate some of the pathological features of TBM 20 . However, high dose intracisternal infection to rapidly produce neurological features also results in dissemination to other organs, which is the reverse of what is envisaged to occur in humans. Dose de-escalation studies may have a role in reproducing the early indolent phase of infection. The original description of the Cynomolgus model of TB, now increasingly adapted as most representative animal model of human TB, described the occurrence of meningitis 21 , but this finding does not appear to have been exploited and may be worth revisiting.
Bacille Calmette Guérin (BCG) vaccine has shown consistently high efficacy against childhood TBM 22 . Negative association between positivity in immune tests of M. tuberculosis sensitisation and prior history of vaccination 23,24 encouraged the recent re-evaluation of BCG revaccination in adolescents with similar results 25 . The inference is that BCG may have partial efficacy to prevent infection but it is unclear whether this is sufficient to explain protection against TBM. Further, it has also been shown that BCG-exposed haematopoietic stem cells generate

Question Potential study designs Corticosteroids
What is the best drug, dose, route of administration and duration?
• Comparative randomised trials against the current standard of dexamethasone in adults, and prednisolone in children What determines corticosteroid treatment effect and how do they improve outcomes?
• Unbiased characterisation of the cerebral pathophysiology of TBM, using serial brain imaging and CSF analysis of inflammatory and metabolic pathways, in those treated with or without corticosteroids (ideally linked to a randomised comparison) Do corticosteroids prevent or reduce the severity of cerebral immune reconstitution inflammatory syndrome in patients coinfected with HIV and starting anti-retroviral therapy • Randomised placebo-controlled trial of corticosteroids given at the initiation of antiretroviral therapy

Thalidomide
What is the role of thalidomide in the management of TBM?

Anti-TNF biological agents (e.g. infliximab)
Do they have a role in the management of the inflammatory complications of TBM?
• In vivo models, including rabbits and mice  dose (1000 mg) adjunctive aspirin therapy was associated with a distinct shift in the pattern of arachidonic acid metabolites 27 , corroborating prior work by the same group that a polymorphism in the gene encoding leukotriene A4 hydrolase (LTA4H) influences both intracerebral inflammation and survival from TBM 28 , and also associates with corticosteroid responsiveness 29 . Second, metabolomic profiling and quantitative trait analysis demonstrated increased survival associated with decreased CSF tryptophan levels, partially under genetic influence 30 . Third, a recent RNA sequence-based study of ventricular and lumbar CSF in children also indicated significant enrichment associated with glutamate-and GABA-mediated neuronal excitotoxicity and cerebral damage 31 . The finding that metabolic derangement may contribute to ongoing brain damage was also suggested by a study of markers of brain injury (non-specific enolase, glial fibrillary acidic protein and S100B), which remained elevated despite a reduction of cytokine mediators in children whose outcome was poor 32 . Taken together, these studies illuminate the power of unbiased 'omics' approaches using materials from the site of disease, which have great potential to better understand pathogenesis and thus illuminate new adjunctive treatments in humans.
The 2019 TBM International Research Consortium meeting in Lucknow also collated information on a significant expansion of ongoing and intended clinical trials activity in TBM which is long overdue. These trials provide an unrivalled opportunity to collect materials for ex vivo analysis as above and to understand interventions. They also present opportunity, where facilities exist, to non-invasively track metabolic disturbance and drug interventions by dynamic imaging (e.g. combined positron emission and computed tomographic [PET/CT], advanced magnetic resonance image [MRI] sequences, or magnetic resonance spectroscopy [MRS]). An interesting example of such an approach in the context of drug intervention was recently provided by the use of 11 C-rifampicin PET/CT in rabbits and humans to investigate rifampicin biodistribution in TBM 34 . Cancer imaging is more advanced in this respect, for example with the availability of a Caspase 3 tracer to detect apoptosis during treatment for cerebral malignancy 35 . Caspase 1 and 5 mediated inflammation via inflammasome activation are implicated in TBM 31,36 and the prospect to determine the anatomical distribution and potential treatment-mediated reduction of such inflammation is of interest. Knowledge gaps for TBM pathogenesis are shown in Table 2.

Diagnosis
Lack of diagnostic suspicion and thus delay is a major contributor to poor outcome in TBM 37 . Modalities in use include pathogen-based or host-response-based tests; and clinical algorithms, which may be supported by radiographic means where available.
Microbiological tests (microscopy, culture and nucleic amplification tests) mostly have good specificity but suboptimal sensitivity and, in the case of culture, render results too late to influence immediate management. The GeneXpert Ultra appears to provide an advance in sensitivity over the first generation of this test and, importantly, where positive, gives early indication of rifampicin resistance 38,39 , which presents an especially difficult management challenge. Poor sensitivity of around 22-24%, but high potentially useful 'rule-in' specificity, has recently been reported for lipoarabinomannan (LAM) detection in CSF or urine by means of a lateral flow assay 40 . A second generation of this test with greater sensitivity has also recently reported and would appear to warrant prospective evaluation 41 .
Host-response-based tests include the CSF determination of free, or antigen-specific release of, interferon-γ, adenosine deaminase, and anti-TB antibodies 1,42 . Diagnostic accuracy tends towards suboptimal, especially in specificity. However, multiplexed pointof-care assays to be performed either on CSF or perhaps serum with cut-offs set at useful triage levels of sensitivity may be of utility and could be evaluated further 43 . Lumbar puncture is invasive and may fail or be impracticable in primary care and thus, blood or other body fluid-based tests applicable at the point of care may have utility. There is, for example, great interest in the application of blood-based transcriptomic assays to the diagnosis of TB and expansion of such evaluations to TBM studies may be of interest 44 .
A substantial number of clinical algorithms have been proposed for TBM: one widely cited and in use was originally intended not for clinical diagnostic use but to help standardise research case definitions to aid comparability of studies 45 . Prospective evaluations of such definitions against microbiological and composite clinical endpoints could be incorporated into prospective diagnostic evaluations.
A general deficiency in TBM research is that high-quality adequately-powered STARD compliant diagnostic evaluations are few 46 . Multi-centre designs would be efficient and also potentially increase generalisability by the inclusion of important groups such as children and HIV co-infected participants, in whom diagnostic accuracy may differ. Coupled to the standardised biobanking of samples for future discovery and evaluation of novel diagnostic, this would seem an obvious priority. An interesting issue is whether such studies should be randomised to one test or another to best-reflect the likely clinical application that only one test would be performed. Concentration techniques applied to CSF consistently improve microbiological yield: in clinical practice large volumes of CSF tend not to be drawn for TB tests. However, the efficient alternative is to divide the same CSF sample for different diagnostic tests. In the context of a prospective well-designed research evaluation, a greater volume than may be considered routine (15-20 ml in adults) is regarded as safe and feasible. Knowledge gaps for the diagnosis of TBM are shown in Table 3.

Antimicrobial therapy
One intuitive way to improve outcome of TBM is by intensifying antimicrobial treatment 47 . Transporters, like P-glycoprotein (P-gp), actively pump substances out of the brain and into the blood or CSF, to protect the CNS from the free entry of potentially neurotoxic substances 48 . These pumps also lower penetration of some first-line TB drugs, especially rifampicin and ethambutol. Also, it is largely unknown how the anatomical distribution, quantity and metabolic status of M. tuberculosis bacilli in TBM

Pathogenesis
• To identify the causes brain injury using pathway analysis in order to determine which pathways should be targeted • To determine the pattern of brain injury using advanced and dynamic cerebral imaging

Diagnosis
• To evaluate strategies that use currently-available tools (such as Xpert Ultra) in rigorous multicentre studies to determine sensitivity and specificity across relevant sub-groups and in different contexts • To evaluate host-based biomarkers of CSF that do not rely on mycobacterial identification for the diagnosis of TBM • To evaluate non-CSF tests for the diagnosis of TBM

Anti-Microbial Drug Therapy
• To determine if higher doses of rifampicin improve outcome in adults and children with TBM • To evaluate if the inclusion of newer drugs, such as linezolid, improve outcome • To evaluate novel mechanisms of drug delivery to the sites of disease

Host directed therapy
• To identify which patients will benefit most from corticosteroids and which will not • To determine how to manage steroid-resistant paradoxical reactions • To determine if aspirin improves outcome

Critical Care
• To evaluate the optimal method for managing raised intracranial pressure • To quantify the target systemic blood pressure for minimal brain injury evolution in TBM • To identify the optimal method of treating hyponatraemia • To determine the minimum package of supportive care that can improve outcome in TBM that can be delivered in a low-resource setting affect the killing effect of specific anti-tuberculous drugs, and if antimicrobial killing may induce (damaging) host inflammatory responses in TBM.

Cascade of Care
There are different strategies to optimize TBM treatment. Prolonging treatment is unlikely to help; in a meta-analysis, the risk of relapse was extremely low (~0.8%), and virtually identical in patients who received <6 months versus >6 months of therapy 49 . A more promising strategy is to increase drug exposures in the CNS by increasing doses of poorly penetrating drugs. In a phase II clinical trial in Indonesia, rifampicin administered intravenously (600 mg, or 13 mg/kg iv) for two weeks resulted in a three-fold higher exposure to rifampicin in plasma and CSF than 450mg (~10 mg/kg) given orally, and a 50% lower mortality 50 . In contrast, a large trial involving 817 TBM patients in Vietnam did not show a survival benefit of an intensified regimen with 15 (rather than 10) mg/kg of oral rifampicin plus levofloxacin versus standard of care for eight weeks 51 , except for patients with isoniazid-mono-resistant TBM 52 . The lack of an effect in that study might be explained by the relatively modest dose increase of rifampicin, as modelling of >1100 pharmacokinetic measurements and survival data from 133 TBM patients in Indonesia suggest that rifampicin exposure strongly predicts survival and that the optimal oral dose is likely to be 35 mg/kg or higher 53 . These findings were corroborated in a study that examined rifampicin cerebral distribution in a rabbit model of TBM 34 . This dose is used in a phase 3 randomised controlled trial (RCT) to be conducted in Uganda, South Africa and Indonesia (ISRCTN: 15668391), and other RCTs that are planned.
Although isoniazid has excellent CSF penetration 54 , its dose may also be important. Isoniazid is metabolized by the genetically polymorphic N-acetyltransferase 2 (NAT2), and rapid acetylators have lower isoniazid exposure in plasma and CSF 55 , and might benefit from higher doses.
Fluoroquinolones have also been examined, but no effect was seen, both in Indonesia 50 and Vietnam 51 . Other second-line or new drugs that may provide meaningful benefit include cycloserine or ethionamide (though these drugs carry high risk of toxicity) 56 , linezolid, and possibly the nitroimidazoles (e.g. delamanid or pretomanid) 57 . It is hoped that new compounds in the pipeline will have characteristics associated with high probability of achieving effective concentrations in the brain and CSF.
Besides higher drug doses or new drugs, modification of drug delivery (e.g. by liposome or nanoparticle-mediated permeation), or inhibition of efflux pumps might also increase drug exposure and effect 48,58,59 . Knowledge gaps for antimicrobial drug therapy in TBM are shown in Table 4.

Host directed therapies
The importance of intra-cerebral inflammation to the complications and outcome of TBM has been appreciated ever since it became a treatable disease in the late 1940s. Streptomycin, the first anti-tuberculosis drug to be trialled in humans, converted TBM from a universally fatal disease to one in which around 30-40% survived 60 . Contemporaneous with this breakthrough was the notion that outcomes might be further improved if inflammation was controlled whilst antibiotics killed the bacteria.
Corticosteroids were the first host directed therapy to be employed in TBM treatment, with trials published in the early 1950s suggesting they caused faster resolution of symptoms and CSF inflammatory indices 61 . However, it took 50 more years to acquire the definitive data from randomised controlled trials that showed adjunctive corticosteroids reduced deaths from TBM 2 .
Corticosteroids for TBM remain the only host directed therapy for any form of tuberculosis with high-quality randomised controlled trial evidence of benefit. There are, however, many reasons to be dissatisfied with them as therapeutic agents. First, they reduce deaths, but do not reduce disability. Second, their benefit appears heterogenous; some patients respond rapidly, with marked improvements in clinical and radiological parameters, whilst others do not respond at all. Third, the mechanism by which they reduce deaths is unknown; it does not, for example, appear to associate with a measurable anti-inflammatory effect 61 . Finally, corticosteroids have a panoply of well-recognised side effects, that can limit their use.
The limitations of corticosteroids as host directed therapy have stimulated various avenues of investigation, all of which represent current research priorities. There is much interest in trying to understand the underlying mechanisms of their therapeutic heterogeneity and how they reduce the risk of death in some people with TBM but not others. The objective of these investigations is to identify the relevant molecules or pathways, which may be targeted by more effective and less toxic drugs.
Clinical necessity in those with corticosteroid-refractory disease has driven the use and study of other available antiinflammatory drugs. Their selection has mostly been predicated on the hypothesis that tumour necrosis factor (TNF) is a critical cytokine in TBM pathogenesis and its inhibition may improve clinical outcomes. To this end, thalidomide and the anti-TNF biological drugs have all been studied, but only in individual reports or small case series 1 . Adequately powered clinical trials have yet to be instigated with these agents.
There is growing interest in the use of aspirin as a host directed therapy for TBM. Brain infarcts are an important cause of disability following TBM and are not reduced by corticosteroids. Thus, the hypothesis that aspirin might prevent infarcts is attractive, but to date has only been subject to three relatively small phase 2 clinical trials 27,63,64 . The most recent trial compared low dose (81mg/day) with higher dose (1000mg/day) aspirin for the first 60 days treatment of 120 adults, complementing clinical response assessments with serial brain imaging and the measurement of lipid inflammatory mediators in the CSF 27 . The results suggested higher dose aspirin may reduce infarcts and deaths, possibly through the up-regulation of lipid mediators that stimulate inflammation resolution. Large phase 3 trials of adjunctive aspirin are now a priority.
Aside from research driven by the benefit and limitations of corticosteroids, and the exploration of other already available anti-inflammatory agents, the priority is to better understand the pathophysiology of TBM. This approach may expose novel pathways and molecules amenable to drugs and could lead to innovative new host directed therapies. It will likely require coordinated investigations that link cellular and animal disease models with careful clinical studies characterising molecular determinants of outcome. The application of 'omics' technologies and analyses may assist in the search for new drug targets. Knowledge gaps for host directed therapy in TBM are shown in Table 5.

Critical care
Critical care, or acute supportive care, is important because neurological disability and mortality commonly occurs early after presentation. TBM pathology initiates secondary mechanisms such as cerebral ischemia, increased intracranial pressure (ICP), excitotoxicity, and other metabolic derangements 30,31 , all of which may interact in a negative downward spiral. Several factors limit our current ability to detect and respond to such derangements, including lack of laboratory and bedside tools to study the disease processes, limited capacity for advanced brain imaging, and few neuromonitoring strategies, compounded by the general lack of resources at sites where TBM is most common. Lack of clinicians trained in critical care and surgical care at these sites further limits options for clinical intervention and research. Late diagnosis may lead to consequences of cerebral pathology too advanced for therapeutic intervention to make a significant difference. Key aspects of initial clinical treatment of TBM are not standardised and practice at different centres is widely divergent. For example, even though hydrocephalus is common in TBM and is the leading cause of increased ICP, there is no consensus on optimal care, including the evaluation of increased ICP, the role of the level of CSF block (communicating versus non-communicating hydrocephalus), indications for medical therapy, criteria for surgical intervention, or the choice of surgical technique 65 .
Although there are small single centre randomised trials of surgical technique, selection criteria vary, and the studies have major limitations. Similarly, hyponatremia is common and associates with poor outcome. Hyponatraemia may be caused by cerebral salt wasting, or the syndrome of inappropriate ADH secretion, or both, and distinguishing between them is difficult. This is important because optimal treatment depends on the diagnosis, and inappropriate care may lead to harm. Protocols for safe and effective amelioration of hyponatremia are needed 3 . Supportive systemic care, including haemodynamic and respiratory support, are important in a general sense, but whether more aggressive interventions, such as blood pressure augmentation and optimization of systemic and cerebral oxygenation, can limit cerebral ischemia remains unknown.
Advanced monitoring of the brain to detect derangements in perfusion and metabolism has been increasingly employed for other acute neurological conditions, but there is little experience of these in TBM and the costs are substantial 66,67 . Non-invasive means to detect increased ICP or impaired brain perfusion have been used in several conditions with variable success and remain under investigation [68][69][70][71] . Finally, early identification of patients at high risk of neurological deterioration, based on clinical criteria, radiological features, and/or biomarkers, could help direct more aggressive care before permanent infarction occurs. Knowledge gaps for the critical care of individuals with TBM are shown in Table 6.

The cascade of care for tuberculous meningitis
Mortality of TBM varies between settings and this may be due to specific variation in the availability and quality of health care services, both prior to, during and after hospitalization. It was recently estimated that 50% of TB deaths result from poorquality care 72 . For TBM it might be even worse, as it requires advanced diagnostics and specialized care, which are often either absent or suboptimal in low-resource settings.
Care for TB patients comprises a cascade of essential steps, with each step unable individually to guarantee a good outcome. It starts with the number of TB patients, followed by the number of patients that accesses TB services for testing, the number diagnosed, the number started on treatment, and finally the number successfully completing treatment 73 . For TBM, the cascade should probably be adjusted to include additional steps such as 'discharged alive' or 'retained to care after discharge' and 'those completing treatment without significant disability' 33 . Some studies have also measured the time between steps 74 . This will be important for TBM, which can be rapidly progressive if no diagnosis is made or treatment started.
Patient pathway studies, which asses the alignment of health systems' infrastructure (e.g. diagnostic, referral and treatment capacity) with patients' care-seeking behavior, could help identify factors that account for losses or delays across a cascade of care for TBM 75 . Many of the possible gaps in diagnosis and treatment of TBM are related to health systems factors, such as the availability of the right facilities or workforce, health information, guidelines, drugs, financing, and organization of the healthy system. A health needs assessment framework measures indicators of performance across system parameters and quantifies the gaps in care against an 'ideal' system, and then considers, using pre-determined criteria, different options to fill each gap. Such public health frameworks have previously been used to identify gaps between current and ideal practice for management of child-case TB contacts 76 , and interventions based on this framework are now tested in a multi-country cluster-randomized clinical trial 77 . Establishing the cascade of care for TBM, conducting a patient pathway analysis, and further study of health systems factors could thus help identify priority areas for further action to improve care and outcomes for TBM patients. Some of the knowledge gaps are presented in Table 7, and the theoretical framework is presented in more detail elsewhere in this supplement 33 .

Conclusions
TBM is the most severe form of TB and contributes disproportionately to TB mortality and morbidity. In order to improve the care of patients with this devastating condition we need to have a better understanding of the disease epidemiology and pathogenesis, as well as improved diagnostic tests and treatment strategies. These then need to be incorporated into the health systems in which care is delivered. Many questions remain unanswered and much research is required. However, the process of defining these questions and deciding on areas of priority allows researchers to focus and coordinate their efforts and allows funders a clearer idea of how and where to spend their money.

Data availability
No data are associated with this article.
Finally, although perhaps less pressing than the acute issues surrounding timely diagnosis and initiation of effective antimicrobial therapy, there continues to be uncertainty regarding the appropriate duration of therapy for tuberculous meningitis, with regards to the risk of relapse. This is a particular challenge with rifampin-resistant disease. For example, what is the role of serial imaging (where available) to guide clinical decision-making for cessation of treatment. Although these questions appear to be the focus of a separate review stemming from the Consortium, it could be worthwhile to highlight this question as well in the broader overview.
In summary, Dr. Seddon have developed an essential framework for the tuberculous meningitis et al. research community, which should be of great value to investigators, public health authorities, and funding agencies.

Does the article adequately reference differing views and opinions? Yes
Are all factual statements correct, and are statements and arguments made adequately supported by citations? Yes

Is the Open Letter written in accessible language? Yes
Where applicable, are recommendations and next steps explained clearly for others to follow? Yes No competing interests were disclosed.

Competing Interests:
Reviewer Expertise: Clinical epidemiology and population pharmacokinetics.
I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard.