Inadequate diagnostics: the case to move beyond the bacilli for detection of meningitis due to Mycobacterium tuberculosis

Abstract

Tuberculosis (TB) meningitis is extremely difficult to diagnose due to its pauci-bacillary disease nature and new techniques are needed. Improved test sensitivity would allow for greater clinician confidence in diagnostic testing and has the potential to improve patient outcomes. Traditional microbiologic and molecular tests for TB meningitis focus on detection of TB bacilli and are inadequate. Smear microscopy is rapid but only ~10–15  % sensitive. Culture has 50–60  % sensitivity but is slow. Xpert MTB/Rif Ultra is a rapid, automated PCR-based assay with ~70  % sensitivity versus clinical case definition. Thus, even the best current testing may miss up to 30  % of cases. Clinicians are often left to treat empirically with prolonged regimens with significant side effects or risk a missed case that would result in death. Rather than relying strictly on microbiologic or molecular testing to diagnose TB meningitis, we propose that testing of CSF for biomarkers of host response may have an adjunctive role to play in improving the diagnosis of TB meningitis.

Introduction

Tuberculosis (TB) meningitis is infamously difficult to diagnose, contributing towards unacceptably high mortality [1]. TB meningitis is generally thought of as causing a lymphocytic pleocytosis with elevated total protein and low glucose in cerebrospinal fluid (CSF); however, CSF cell count and chemistry profile are neither sensitive nor specific enough to use as the primary method of diagnosis [2]. In fact, CSF in up to one-third of TB meningitis patients can show a neutrophil predominance [3]. Further, although clinical criteria have been developed to assist in diagnosis, performance variability may be seen based on local human immunodeficiency virus (HIV) and tuberculosis prevalence and due to atypical clinical features such as CSF neutrophilia in individual cases. Furthermore, the Thwaites criteria were developed to distinguish TB meningitis from bacterial meningitis, but not meningitis from other causes [4, 5].

Detection of M. tuberculosis bacilli from CSF is limited by TBM being pauci-bacillary as compared to pulmonary TB. For example, Xpert MTB/RIF quantifies bacillary burden with semi-quantitative categories based on PCR threshold cycle values. Of 302 subjects with pulmonary TB by sputum Xpert MTB/RIF, 73 (24 %) had a very low burden, 75 (25 %) low, 114 (38 %) medium and 40 (13 %) high [6]. In comparison, we reviewed our data from 22 individual cases of TB meningitis that were included in prior studies [7, 8]. Among those 22 cases identified by having a positive CSF Xpert MTB/RIF, 14 cases (64 %) had a very low burden, 7 (32 %) a low burden, and 1 (5 %) a medium burden in CSF [7, 8].

Thus, our current diagnostic tests, focused on detecting TB bacilli, are inherently disadvantaged as there are simply less bacilli present in lumbar CSF. Moreover, it is possible that in some cases there are no bacilli in the lumbar CSF due to significant inflammation in the brain and poor CSF circulation from the ventricular to the lumbar CSF [9]. The fact that few bacilli are present in the accessible, lumbar CSF means that diagnostic tests must have a very low limit of detection in order to be effective. Additionally, patients with low bacillary burden often have significant inflammation, although this can be variable. For instance, among the same 22 cases, 15 had available results for total protein of which the nine cases with a very low burden by Xpert had a median total protein of 94 mg dl⁻¹ [interquartile range (IQR) 19–300 mg dl⁻¹)], while the five cases with low burden had a median total protein of 50 mg dl⁻¹ (26–152 IQR mg dl⁻¹). The one case with a medium burden had a total protein of 20 mg dl⁻¹ [8]. Of the 22 subjects note above, 16 had results available for CSF white cell count. Of those 16 cases, 12 had very low burden by Xpert and had a median of 73 white cells μl⁻¹ (IQR <5–295 cells µl⁻¹), while the five cases with a low burden had a median of 22 white cells μl⁻¹ (IQR 8–1262 cells µl⁻¹), and the one case with a medium burden <5 white cells μl⁻¹ [7, 8]. Thus as inflammation increased in the CSF, the detectable burden of bacilli decreased.

Current state of bacillary diagnosis of TB meningitis

Visualization of CSF smear for acid-fast bacilli (AFB) has ~10–15 % sensitivity in routine practice although the test is rapid and cheap [2, 8]. Mycobacterial culture has greater sensitivity, ~60–70 % compared to any positive CSF TB test [8] or clinical research criteria [10], but takes at least 2 weeks (and in many cases up to 6 weeks) to provide results [2]. In our recent experience in Uganda, 7 of 10 cultures became positive after patients had died with the median time to culture positivity being 20 days (range 13 to 48). This time frame does not allow for clinically actionable results. Rather, clinicians must often decide whether or not to initiate empiric therapy prior to culture results being available. Xpert MTB/Rif (Cepheid, Sunnyvale, CA, USA) is a PCR-based single cartridge technology that allows for rapid results in 2 h. Sensitivity for TB meningitis is generally 50–60 % although values from 30–70 % have been reported when compared to any positive CSF TB test or clinical research criteria [8, 10, 11]. One study found that Xpert sensitivity improved to ~70 % with centrifugation of larger volumes of CSF (~6 ml) against a composite case definition of a positive PCR test or culture although this finding has not been universal [8, 12].

Although Xpert and culture have similar sensitivities, they may show discordant results [8]. Among 18 microbiologically proven cases of TB meningitis in one study, Xpert and culture had equivalent sensitivities of ~70 % [8]. However, only 39 % (7/18) of cases were positive by both culture and Xpert with 4 cases only positive by Xpert and 5 cases only positive by culture [8].

A re-engineered version of Xpert, termed MTB/Rif Ultra, has been produced (Cepheid, Sunnyvale, CA, USA). A recent prospective study found 95 % (21/22) sensitivity for Xpert Ultra when compared to culture and/or PCR (Xpert or Xpert Ultra) positivity whereas standard Xpert and culture each were found to have 45 % (10/22) sensitivity [7]. However, when compared with consensus case definitions [13], Xpert Ultra had only 70 % (16/23) sensitivity for definite or probable TB meningitis [7]. Though Xpert and Xpert Ultra provide rapid results and sensitivity similar to that of culture, their negative predictive value is not sufficient for this to be a reliable test to ‘rule-out’ TB meningitis [1]. Other nucleic acid amplification tests have been tested but have not been widely adopted [14]. Other microbe-based targets, such as lipoarabinomannan antigen (LAM), have been evaluated though sensitivity has been poor [1]. Clearly, we have not yet found adequate approaches for diagnosing TB meningitis via bacillary detection.

Immunologic findings in TB meningitis

Mycobacterium tuberculosis is an intracellular pathogen that elicits a Th1 response (IFN-γ, TNF-α) and pro-inflammatory cytokines such as IL 1α and IL1-β to control the infection [15, 16]. In TB meningitis, significant meningeal and intracerebral inflammation reflected in the CSF profile may contribute to poor outcomes [17, 18]. Frequently, total protein in the CSF ranges from ~200–300 mg dl⁻¹ and white blood cell (WBC) count from ~300–400 cells µl⁻¹ [2]. This inflammation can be variable, particularly in HIV-infected persons where CSF pleocytosis may be completely absent [19]. Fewer CSF white cells have been associated with mortality in some, but not all settings [18, 20].

Harm to the host broadly appears to result from a number of scenarios. First, absent or minimal immune response can allow for proliferation of bacilli. Second, a dysfunctional immune response (e.g. Th2-IL4, IL-13) may generate inflammation without controlling the organism, but this inflammation may activate coagulation cascades leading to cerebrovascular events. Lastly, over-exuberant Th1 inflammation can also lead to immunopathology, such as can occur in paradoxical immune reconstitution inflammatory syndrome (IRIS) events. Each situation potentially leads to poor outcomes – particularly among HIV-infected subjects [21–23].

An example of the mixed/dysfunctional immune response in TB meningitis was shown in a mixed HIV-status population (total n=15, HIV infected n=2), where higher levels of IL-10, TNF-α and IFN-γ were found compared to control subjects with non-infectious neurologic disease including cerebrovascular disease, hydrocephalus, encephalomyelitis and epilepsy [24]. Neutrophils, in particular, may play a role in inflammation-related pathology in TB meningitis. A recent Indonesian study found that among subjects with CSF cultures positive for M tuberculosis , higher neutrophil count (per 10⁹ CSF neutrophil/L increase) was associated death, hazard ratio 1.35, (95 % CI 1.03–1.78, P=0.029) [17]. Further, neutrophils generate inflammation but may not adequately clear the pathogen as evidenced by Thwaites’ finding that higher CSF neutrophil proportion was associated with greater CSF M. tuberculosis culture positivity [25]. Neutrophils, drawn by chemokines, create inflammation but do not kill TB bacilli, an example of a dysfunctional response. Interestingly, the proportion of neutrophils among CSF cell counts among HIV-infected subjects with a CD4⁺ T-cell count <150 (median 25 %) were higher than those with CD4⁺ T-cell count >150 (median 10 %, P=0.021) or those without HIV infection (median 5 %, P<0.0001) in a recent study in Vietnam [18].

Overall, we hypothesize that those lacking an appropriate Th1 immune response have less control of the pathogen, greater bacillary burden, and thus, are easier to detect by microbiologic and molecular diagnostics; whereas those with appropriate or exuberant immune responses have greater pathogen control, lesser bacillary burden, and thus, are more difficult to detect via pathogen-focused techniques (Fig. 1). Both scenarios may result in clinical disease, but those with a better immune response and lower burden of disease will likely be more difficult to diagnosis, and paradoxically may have worse outcomes by lack of prompt diagnosis.

Fig. 1.

Diagnostic framework for TB meningitis. The figure displays a modified damage response framework. Right to left shows increased immune response and thus decreased organism burden (requiring more sensitive, bacilli-based detection methods for diagnosis). At the extremes of immune response, poor clinical outcomes occur.

Using immunologic techniques to diagnose TB meningitis

There has been interest in immunology-based tests for diagnosis of TB meningitis (Table 1). Adenosine deaminase (ADA) is produced by lymphocytes, is important for T-cell proliferation and differentiation, and is used as a TB diagnostic test in other body fluids [26]. However, a large meta-analysis found wide-ranging sensitivity in CSF with a median 79 % sensitivity (95 % CI, 75–83 %) [27]. Further, in HIV-infected persons, a high number of false positives occur [28]. Cost, variable performance, and laboratory infrastructure requirements have limited ADA use.

Table 1.

Summary of recent meta-analyses of immunologic CSF assays studied for diagnosis of TB meningitis

Diagnostic test	Sensitivity	Specificity	Limitations
Adenosine deaminase [27]	79 % (75–83 %)	91 % (89–93 %)	Cost, variable performance, lab infrastructure
IGRA [29]	77 % (69–84 %)	88 % (74–95 %)	Cost, lab infrastructure, reliance on T-cell function, frequent indeterminate results
Anti-M37Ra [30]	91 % (71–98 %)	99 % (89–100 %)	Study heterogeneity, lack of commercial assays, lab infrastructure
Anti-M37Rv [30]	84 % (71–92 %)	98 % (96–99 %)	Study heterogeneity, lack of commercial assays, lab infrastructure
Anti-antigen 5 [30]	84 % (71–92 %)	95 % (91–98 %)	Study heterogeneity, lack of commercial assays, lab infrastructure

Values represent median (95 % CI).

IGRA, interferon gamma release assay.

CSF IFN-γ levels have been considered as a diagnostic tool as well [31–33]. One representative study of 20 subjects with TBM compared levels of CSF IFN-γ with 59 subjects with meningitis of other etiologies and 49 subjects without meningitis – a cutoff of 6.4 IU ml⁻¹ yielded a sensitivity of 70 % and a specificity of 94 %. The majority of subjects with TBM were HIV-infected [32]. However, the high numbers of false positives (7/59, 12 % with other causes of meningitis in this study) make use of IFN-γ levels alone problematic.

Interferon gamma release assays (IGRAs) measure ex vivo T-cell immune responses to M. tuberculosis antigens and are commonly used to detect latent tuberculosis. A 2016 meta-analysis of six studies measuring CSF IGRA use found 77 % sensitivity and 88 % specificity for TB meningitis [29]. IGRA implementation has been limited by cost, reliance on T-cell function (potentially problematic in HIV-infection), laboratory infrastructure requirements, and frequent indeterminate results, including in HIV-infected subjects [2, 34, 35]. Variability in cut-points used has hampered interpretation of findings across different studies. The inadequate specificity compared to microbiologically proven TB meningitis of IGRAs may reflect cases that truly are TB meningitis being mislabelled as ‘not TB meningitis’ due to the inability of traditional microbiological testing to detect cases of TB meningitis with low bacillary load yet significant inflammation or may reflect a lack of IFN-γ specificity to M. tuberculosis .

CSF antibodies have also been considered for TB meningitis diagnosis. 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 [30]. Antibody-based assays and studies thereof are limited by heterogeneity across research studies, lack of a single gold-standard in studies, and lack of commercial assays. Additionally, the World Health Organization has advised against use of these antibody testing in blood for diagnosis of pulmonary or extrapulmonary tuberculosis [36]. Thus, current clinical use is minimal [30, 36].

Combining immunology, molecular diagnostics and microbiology

Given that molecular and microbiology-based diagnostics have thus far proved inadequate, a new conceptual approach may be warranted. Though understanding of the human immunologic response to TB meningitis is incomplete, a number of methods that use an assessment of immunologic response to diagnose TB meningitis (e.g. antibodies, ADA and IGRA) have been studied. None have been adopted for clinical use due to various limitations although Vita and colleagues published a case report demonstrating the use of immunologic markers as adjunctive tools in diagnosing TB meningitis [37].

Such innovative combinations of microbiologic and immunologic diagnostics may be the future for diagnosis of TB meningitis. While some cases are easily detected by microbiologic methods, many are not (Fig. 2), and while molecular methods such as Xpert Ultra are promising, molecular methods will almost certainly not be able to be used as stand-alone tests. From the data available, it appears that up to one-third of TBM cases are not detected by the best, currently available microbiologic/molecular diagnostic tests. This may be even higher outside of research studies (e.g. due to suboptimal CSF volumes).

Fig. 2.

Venn diagram describing current TB meningitis diagnosis as related to bacillary burden and immune response. The figure shows increasing bacillary burden from right to left, reflective of an absent or dysfunctional immune response. From left to right, the response is more likely to be functional (Th1), thus, increasing levels of inflammation and lower organism burdens are seen. The dark circle on the left represents cases of TB meningitis able to be detected using microbiologic/molecular methods – higher bacillary burdens. The white circle represents cases that are more difficult to detect currently, with less bacillary burden and more inflammation. These cases may potentially be detectable by immunology-based diagnostic tests.

Harnessing assays that evaluate the immune response in combination with available microbiologic/molecular technologies may allow us to make sure those who need treatment for TB meningitis get it rapidly, while sparing those who do not from prolonged unnecessary antimicrobial therapy that has potential for side effects. Possible targets include Th1 cytokines such as IFN-γ, CSF IGRA and/or novel antibodies. In addition, discovery techniques such as mass spectrometry and RNAseq or microarray may lead to the identification of novel biomarkers of metabolic profiles, secreted lipid, protein products or RNA transcripts [38]. It is likely that more than one target might be required in addition to current techniques, thus, modelling and/or machine learning may play a role in defining the most promising diagnostic combination.

Of course, for any new diagnostic paradigm to be truly accessible for most patients who are afflicted with TB meningitis – the technology must also be affordable, easily operationalized, and ideally, require minimal laboratory infrastructure. Thus, if any of the above targets were to be considered in combination with currently available technologies effort would be required to streamline laboratory techniques for scale-up. One successful example of this is the collaborative efforts between the Foundation for Innovative NEW Diagnostics (FIND) and Cepheid in the development and widespread implementation of Xpert for pulmonary TB diagnosis.

Conceptual applicability outside of TB meningitis

TB meningitis is an example of an infectious condition with a variety of microbiologic and immunologic phenotypic presentations. While some cases are easily detected by microbiologic methods, others show significant inflammation but bacilli are not detected by current techniques ( Fig. 2). It is likely that those cases that are easily detectable have a high bacillary load and a dysfunctional or absent immune response whereas those with a low bacillary load have a properly directed, robust Th1 immune response, which results in fewer bacilli present for detection.

Yet, this concept is not unique to TB. Other infections have heterogeneous disease presentations and may benefit from a diagnostic approach incorporating immunologic diagnostics as well. For instance, Histoplasma capsulatum is known to cause progressive disseminated histoplasmosis, typically in persons with profound immune suppression [39]. In other patients, H. capsulatum causes mediastinal fibrosis, likely related to an exuberant Th2 inflammatory response to H. capsulatum antigens present with few or no viable yeasts [39]. While progressive disseminated histoplasmosis is readily detected microbiologically, antibody response is often the only evidence of recent histoplasmosis infection in mediastinal fibrosis [39]. Similarly, invasive pulmonary aspergillosis is typically more amenable to microbiological diagnosis while allergic bronchopulmonary aspergillosis involves detection of an altered immune response (total and specific IgE antibodies as part of criteria for diagnosis) [40].

Conclusions

TB meningitis is notoriously difficult to diagnosis due to the pauci-bacillary CSF associated with the disease. We propose that the diagnostic approach for TB meningitis may need to be altered to include immunologic methods of detection. Rather than relying strictly on microbiologic or molecular diagnoses, we believe that immunologic methods may be complementary and additive, allowing for more sensitive detection of TB meningitis and thus potentially improving outcomes. Further research studies are required to address whether such approaches would be beneficial. We promote both research investigating the potential of novel technologies such as transcriptional analysis and coupling those and/or other tools such as IGRA or IFN-γ measurement with pathogen-based molecular and/or microbiologic techniques.