Potential biomarkers for tuberculous meningitis diagnosis using metabolomics and proteomics: a systematic review

Abstract

Tuberculosis (TB) is the leading cause of death from a single infectious agent, with approximately 1.2 million deaths reported in 2023. While TB primarily affects the lungs, it can also spread to other organs, where it is classified as extrapulmonary TB. Tuberculous meningitis (TBM) is the most severe form of extrapulmonary TB, affecting 1–5% of TB cases. Delayed diagnosis contributes to its high mortality and severe neurological complications, with approximately 10% of affected individuals dying or suffering permanent neurological damage. When combined with adjunctive therapy, early detection and treatment can significantly improve survival outcomes. Currently, many studies have identified potential biomarkers of TBM; however, to date, there is no clear consensus on the markers altered in TBM. Hence, we conducted a systematic review aimed at identifying metabolites and proteins that are significantly altered in TBM when compared with healthy controls. Three databases — PubMed, Scopus, and Web of Science — were scanned by two independent reviewers for potential articles that met our inclusion and exclusion criteria. After quality assessment, 17 studies were included, comprising a total of 963 participants (healthy control, n = 576; TBM, n = 387). Metabolites and proteins identified as being significantly altered across studies included alanine, isoleucine, myo-inositol, valine, arachidonate 5-lipoxygenase (ALOX5), apolipoprotein B (APOB), and glial fibrillary acidic protein (GFAP), which were detected in serum, urine, brain tissue, and cerebrospinal fluid samples. These markers have potential diagnostic value for TBM. However, further validation is needed to determine their specificity to reliably distinguish TBM from other neurological infections, which could improve early diagnosis and patient outcomes in TBM.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12879-025-11740-6.

Introduction

Tuberculosis (TB) is a deadly infectious disease and the leading cause of mortality from a single infectious agent – Mycobacterium tuberculosis (M. tb). Although the primary site of infection is the lungs (resulting in pulmonary TB – the most common form of the disease), it can also spread to other parts of the body, where it is classified as extrapulmonary TB. The most severe form of extrapulmonary TB is tuberculous meningitis (TBM), which accounts for 1–5% of the global TB burden and results in mortality rates of 20–60%, particularly among children and individuals co-infected with HIV [1–3]. Early diagnosis of TBM and subsequent treatment are crucial for reducing mortality rates and improving patient outcomes [4].

The primary biofluid used for diagnosing TBM is cerebrospinal fluid (CSF). Diagnostic tools for TBM include microbial culture, nucleic acid amplification tests (NAATs), GeneXpert, and, more recently, Xpert Ultra [1]. Although these tools have proven somewhat effective in diagnosing TBM, they may be time-consuming, lack sufficient sensitivity or specificity, or both, and are additionally considered expensive and have limited availability in resource-limited regions [1, 5]. Furthermore, the decision to use a diagnostic tool depends on its availability within the local healthcare system [6]. Hence, novel, sensitive, specific, easily accessible, and cost-effective diagnostic tools for TBM are needed.

Advancements in technology and machine learning have facilitated the exploration and identification of potential biomarkers for TBM [7], broadening the range of sample types beyond CSF, to include other biological fluids such as serum, urine, and post-mortem brain tissue [8–11]. Metabolomics and proteomics, which involve the comprehensive analysis of metabolites and proteins, respectively, within tissues and biofluids, have been increasingly employed in the discovery of biomarkers. Recent studies propose that advancements in diagnostic accuracy may be achieved through the identification of host immune-based biomarkers that exhibit both disease specificity and therapeutic relevance [7, 12–14]. To identify biomarkers specific to TBM, patient samples should be compared with those from corresponding healthy individuals, carefully considering potential confounding factors that may influence the conclusions drawn. Metabolomics and proteomics have the capacity to identify new biomarkers of TBM. Although many studies have identified potential biomarkers of TBM, to date, there is no clear consensus on the markers most commonly altered in TBM cases when compared to healthy controls. Therefore, this systematic review aims to highlight metabolites and proteins identified in research studies that demonstrate potential as future TBM biomarkers, while assessing their alignment with established biomarker criteria. Additionally, this review provides recommendations based on the findings, focusing on better reporting practices and strategies to establish appropriate diagnostic benchmarks for TBM.

Methods

Study design

We aimed to summarise significant metabolites and proteins related to TBM identified in the literature. This review is descriptive, and no meta-analysis was carried out. This review has been approved by the North-West University ethics committee (number NWU-00073-25-A1) and has been registered on PROSPERO (number CRD420250626311). The study was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Fig. 1).

Fig. 1.

Preferred reporting items for systematic review and meta-analysis (PRISMA)

Eligibility criteria

The eligibility criteria were tissue and body fluids from children and adults with TBM. Only studies focused on metabolites and proteins with healthy control patients from these cases were included. Additionally, studies that utilised analytical platforms such as ¹H-NMR, liquid chromatography-tandem mass spectrometry (LC-MS), gas chromatography-mass spectrometry (GC-MS), and enzyme-linked immunosorbent assay (ELISA) were included to ensure uniformity in the reported results.

Studies that had patients with comorbidities such as HIV and control samples with neurological diseases or infections were excluded. Additionally, since this review focuses on human TBM, animal studies were excluded from consideration. Non-English studies were excluded because translating could lead to misinterpretation and misunderstandings by the authors and reviewers. To ensure that the metabolites and proteins significantly altered in the disease states were only found in TBM cases, only studies that had healthy controls were included, and since treatment could play a significant role in the metabolites and proteins present, we excluded studies that focused on the treatment of the disease rather than the disease.

Data sources

The databases used for this systematic review were PubMed, Web of Science, and Scopus. All published studies up to October 2024 were exported without restrictions on the publication date. The search terms applied were PubMed: (tuberculosis meningeal [MH] OR TBM OR TBM [Tw] OR tuberculous meningitis OR tuberculous meningitis [Tw]) and (metabolomics OR metabolomics [Tw] OR proteomics OR proteomics [Tw]); Web of Science: TS=(TUBERCULOUS meningitis or TUBERCULOSIS meningeal OR TBM) AND TS=(METABOLOMICS OR PROTEOMICS); Scopus: ALL (tuberculous meningitis OR TBM) AND ALL (metabolomics OR proteomics).

Data selection, cleaning, analysis and quality assessment

Two independent reviewers, ARI and SM, conducted data selection and review using Covidence, an artificial intelligence (AI) tool for systematic reviews, in accordance with the established inclusion and exclusion criteria. Conflicts were resolved based on discussions between the reviewers, and a consensus was reached based on these discussions. Data cleaning was performed using the web-based platform Open Refine. Inconsistencies, such as spelling, capitalisation, and punctuation, were standardised to avoid repeating the same word multiple times during data analysis. Power BI and Robvis were used for qualitative data visualisation.

Based on our study design, the JBI critical appraisal for case-control studies was used. This tool was amended based on the specific outcomes of our research. Questions used in the quality assessment were: (1) Were the groups comparable other than the presence of disease in cases or the absence of disease in controls? (2) Were cases and controls matched appropriately? (3) Were confounding factors identified and adjusted for in the stats? (4) Were there incomplete outcome data and selective reporting? (5) Is there a report on the current treatment? Based on these questions, we did a ranking of the included studies using the Likert scale [15]: 0 for no, 1 for unclear, and 2 for yes and 0 for high, 1 for unclear, and 2 for low in incomplete outcome data and other sources of bias (Questions 5 and 6). A score of 0–5 represents a low-quality paper, and a score of 6–10 represents a high-quality paper (Fig. 2).

Fig. 2.

Quality assessment and risk of bias assessment

Results

Study characteristics

A total of 847 studies retrieved from the databases were uploaded to Covidence, where title, abstract, and full-text screening were conducted. Fifty-seven duplicates were automatically removed, and ten were manually removed. Seven hundred eighty studies underwent abstract screening, and seven hundred and forty-eight of these were deemed irrelevant to our studies because they were either review articles, animal studies, employed a study design not suited to this study, or included participants with comorbidities such as HIV, which could complicate the metabolic or proteomics profile. Hence, 32 studies were selected for full-text screening. In addition, non-English studies (n = 1) were excluded. Fifteen studies were excluded from the analysis after full-text screening because they did not meet the inclusion criteria, leaving a final number of 17 studies that were extracted and quality-assessed.

The selection criteria included 17 studies, which were conducted across five countries: South Africa, China, India, Turkey, and Indonesia, with the majority of studies originating from South Africa. Only one study analysed human brain tissue, another serum, with the rest conducted using CSF (n = 12) and urine (n = 4) (more recently) (Fig. 3). The total participants included were 1,237 (including non-TBM patients), 576 controls, 387 TBM cases (patients who have been confirmed using bacterial isolation, polymerase chain reaction test for tuberculosis, acid fast bacilli AFB by Ziehl Neelsen stain where cases were diagnosed with the tuberculous meningitis disease) and 274 non-TBM (other forms of meningitis) cases, which included both children and adults, of ages ranging from 6 months to 65 years(additional information is available in Supplementary Table 1). The year of publishing ranged from June 2011 to August 2024. All the included studies were case-control studies, and the participants were recruited from those visiting clinics. A total of nine of the seventeen studies did not include any information on the treatment regimen of the TBM patients. A total of 17 unique proteins and 86 unique metabolites were identified across all studies. Four studies reported on the sensitivity and specificity of the identified proteins in TBM. In comparison, two other studies (diagnostic for definite and probable TBM) reported a 95% confidence interval of the identified metabolites in relation to TBM (Supplementary Table 1). The other studies only indicated the significance level of the metabolites and proteins compared to the healthy controls. Only one study included participants who had been confirmed with TBM and were treatment-naïve [16].

Fig. 3.

Total number of included studies by geographical location, categorized by study population (adults and children (A), and corresponding sample matrices by geographical location (B)

CSF = cerebrospinal fluid; iTRAQ = isobaric tags for relative and absolute quantification; LC-MS/MS = liquid chromatography-tandem mass spectrometry; ¹H-NMR = proton nuclear magnetic resonance spectroscopy; ELISA = enzyme-linked immunosorbent assay; GC-MS = gas chromatography-mass spectrometry.

The instruments used in the analysis of proteins included LC-MS, ELISA, iTRAQ coupled with LC-MS, and Q-TOF MS-MS. In the metabolomics analysis, ¹H-NMR (the most commonly used method), LC-MS, and GC-MS (the least commonly used method) were employed. The frequently occurring metabolites and proteins were alanine, isoleucine, myo-inositol, and valine (identified at least five times), and arachidonate 5-lipoxygenase (ALOX5) (identified at least three times) across all studies, respectively. These common metabolites and proteins were primarily identified in CSF and urine samples from China, Italy, and South Africa using ¹H-NMR, LC-MS, ELISA, and Q-TOF MS for analysis (Table 1).

Table 1.

Significant proteins and metabolites that were detected across all included TBM studies

Sample type	Protein(s) of significance	Metabolite(s) of significance	Equipment	Country	Reference
CSF		21 metabolites: glucose, 3-hydroxybutyrate, L-valine, myo-inositol, 1,3-dimethylurate, L-alanine, L-isoleucine, lactate, isobutyrate, 3-hydroxyisovalerate, creatinine, caprate, choline, acetamide, pyruvate, isovalerate, cyclohexane, glycine, 2-oxoglutarate, 2-hydroxyisovalerate, L-methionine	1H-NMR	China	[17]
CSF		20 metabolites: 2-hydroxybutyrate, acetate, alanine, choline, citrate, creatinine, isoleucine, lysine, myo-inositol, pyruvate and valine, carnitine, creatine, creatine phosphate, glutamate, glutamine, guanidinoacetate, proline, lactate, glucose	1H-NMR	South Africa	[18]
CSF		16 metabolites: alanine, valine, isoleucine, myo-inositol, creatinine, dimethyl sulfone, lysine, phenylalanine, tyrosine, acetate, pyruvate, citrate, dimethyl sulfone, acetate, glucose, leucine	1H-NMR	South Africa	[19]
CSF		17 metabolites: 1-methylhistidine, 2-aminobutyrate, alanine, asparagine, ethanolamine, histidine, isoleucine, lysine, methionine, norvaline, ornithine, phenylalanine, proline, sarcosine, threonine, tyrosine, valine	LC-MS/MS	Turkey	[20]
CSF		11 metabolites: lactate, glutamate, alanine, arginine, 2-hydroxyisobutyrate, formate, cis-aconitate, glucose, fructose, glutamine, myo-inositol	1H-NMR	India	[21]
Serum; CSF		139 metabolites with a focus on one metabolite: cerebral tryptophan	LC-MS	Indonesia	[9]
Urine		8 metabolites: 2-methylbutyrlglycine, 3-hydroxypropionate, 3-methylcrotonylglycine, 4-hydroxyhippurate, 5-hydroxyindoleacetate, 5-hydroxyhexanoate, isobutyrylglycine, phenylacetylglutamine	LC-MS/MS	South Africa	[22]
Urine		29 metabolites: quinolinate, tyrosine, leucine, 3-hydroxyisobutyrate, lysine, isoleucine, valine, glycine, o-cresol, 4-hydroxyphenylacetate, m-cresol, formate, arabinose, hippurate, methylamine, methylguanidine, myo-inositol, 3-hydroxyisovalerate, glucose, sucrose, mannose, pyruvate, acetone, acetate, acetoacetate, 1-methylnicotinamide, trigonelline, N-acetylglutamine, urea	1H-NMR	South Africa	[23]
Urine		5 metabolites: 1-methylnicotinamide, 3-hydroxyisovalerate, 5-aminolevulinate, N-acetylglutamine, methanol	1H-NMR	South Africa	[24]
Urine		4 metabolites: methylcitrate, 2-ketoglutarate, quinolinate, 4-hydroxyhippurate	GC-MS	South Africa	[25]
CSF	Neural epidermal growth factor-like like 2 (NELL2)		LC-MS/MS	China	[26]
CSF	Arachidonate 5-lipoxygenase (ALOX5) and glial fibrillary acidic protein (GFAP)		2D gel electrophoresis (DIGE) and Q-TOF MS/MS	India	[27]
CSF	Arachidonate 5-lipoxygenase (ALOX5)		ELISA	India	[28]
CSF	Apolipoprotein B (ApoB)		iTRAQ and LC-MS/MS	China	[29, 30]
CSF	9 proteins: apolipoprotein A-IV (APOA4), orosomucoid 1 (ORM1), cathelicidin antimicrobial peptide (CAMP), S100 calcium-binding protein A8 (S100A8), apolipoprotein B-100 (APOB), 1-antichymotrypsin (SERPINA3), prostaglandin D2 synthase (PTGDS), apolipoprotein E (APOE), calsyntenin-1 (CLSTN1)		iTRAQ coupled with 2D LC-MS	China	[16]
CSF	S100 calcium-binding protein B (S100B), arachidonate 5-lipoxygenase (ALOX5), glial fibrillary acidic protein (GFAP), defensin-alpha 1 (DEFA 1)		ELISA	Turkey	[30]
Brain tissue	Amphiphysin (AMPH), Neurofascin (NFASC), and ferritin light chain (FTL)		Quantitative protein expression profiling using iTRAQ labelling and LC-MS/MS	India	[8]

Population/cohort description

The sex distribution in the TBM group was 193 males and 155 females (three studies did not specify sex distribution, n = 39). No statistical differences were observed in the sex distribution across reported studies. However, the effects of the results as they relate to the different sexes were not stated. None of the studies examined children and adults in the same study; therefore, one can only infer the differences in metabolite and protein profiles between adults and children. All control and TBM cases were matched appropriately, except in cases where the controls exceeded the number of TBM cases, as in studies of other infectious diseases. The diagnosed TBM cases ranged from definite to probable disease. The definite TBM cases were microbiologically tested, and the presence of M. tb proved that the patients were infected. Although TBM can be categorised into stages (1, 2, and 3), very few studies describe the TBM stage of their patients.

Common TBM differentially altered metabolites and proteins detected

This review identified several proteins and metabolites that are frequently reported as being significantly altered in TBM cases. The majority of the studies included in this review aimed to compare the metabolites or proteins differentially altered in TBM cases with those in controls and to infer the metabolites or proteins involved in disease severity. The most commonly identified proteins were arachidonate 5-lipoxygenase (ALOX5), apolipoprotein B (APOB), and glial fibrillary acidic protein (GFAP). Among the metabolites, alanine, isoleucine, myo-inositol, valine, glucose, lysine, creatinine, lactate, pyruvate, tyrosine, 1-methylnicotinamide, 3-hydroxyisovalerate, 4-hydroxyhippurate, acetate, choline, citrate, glutamate, glutamine, glycine, leucine, methionine, N-acetylglutamine, phenylalanine, proline, and quinolinate were most frequently observed (Supplementary Table 2). Some studies have validated these differentially altered metabolites and proteins, but still require validation in a larger cohort (Tables 1 and 2).

Table 2.

Information on the TBM category, treatment received, and the aim of the study

TBM category	Treatment received	Aim of study	Validated	Reference
Patients diagnosed with TBM	No information on treatment	To identify differential proteins in CSF obtained from live TBM patients and healthy controls	No	[26]
Patients diagnosed with TBM	No information on treatment	To identify molecular markers and sensitive methods in early-stage TBM infection	No	[27]
Confirmed TBM	No information on treatment	To identify differentially altered proteins from the brain tissue of TBM cases as compared to controls	No	[8]
Confirmed TBM	Yes	Study directed towards the host’s metabolic response to TBM	No	[19]
Confirmed TBM	No information on treatment	To evaluate ALOX5 as a diagnostic tool in TBM	No	[28]
Patients diagnosed with TBM	No information on treatment	To identify differentially altered proteins from the CSF of TBM patients as compared to patients with viral meningitis and healthy controls	No	[29]
Definite and probable TBM	No information on treatment	A systematic study of CSF metabolic profiling among adult patients with TBM, VM, and BM, as well as meningitis-negative controls, using an NMR-based platform and stringent statistical analyses	No	[17]
Definite TBM	Yes	To compare cerebrospinal fluid (CSF) and serum metabolomes of patients with tuberculous meningitis with those of controls without tuberculous meningitis, and assess the link between metabolite concentrations and mortality	Yes	[9]
Definite TBM	No information on treatment	To better characterise the CSF metabolic profile in a South African TBM paediatric cohort	No	[18]
Patients diagnosed with TBM	No information on treatment	To detect amino acid changes, if any, by comparing the levels of amino acids in CSF samples of patients with aseptic, bacterial, and TBM and the control group using the liquid chromatography-tandem mass spectrometry method (LC-MS/MS)	No	[20]
Confirmed TBM	No	To investigate whether ALOX5, S100B, DEFA1, and GFAP could be used as biomarkers to distinguish between different types of infectious meningitis	No	[30]
Confirmed TBM	Yes	To identify urinary metabolites associated with host and/or microbial changes associated with dysbiosis in TBM cases during treatment	No	[22]
Definite TBM	No	To identify differential proteins in patients with TBM	No	[16]
Definite TBM	Yes	To identify metabolites in urine that characterise the metabolic profile of severe TBM in paediatric cases, using 1H-NMR metabolomics	No	[23]
Confirmed TBM	Yes	To identify biomarkers of TBM with high precision and specificity using noninvasively collected urine	No	[24]
Confirmed TBM	Yes	To report the potential of 1H-NMR-based metabolomics in CSF for discrimination of the definite and probable cases of TBM from control subjects and evaluate the correlation of metabolomics with clinical and radiological findings and outcomes	No	[21]
Confirmed TBM	Yes	To exploit metabolomics as an approach to identify metabolites as potential diagnostic predictors for children with TBM through non-invasive means	No	[25]

These proteins and metabolites have the potential to serve as biomarkers for TBM, provided they are validated to confirm their relevance in disease diagnosis and management.

Sample-specific differences in significantly altered proteins and metabolite determination

The distribution of these significantly altered proteins and metabolites was strongly influenced by the type of biological sample used for analysis.

• CSF: ALOX5 was the most frequently identified protein in CSF across different analytical pipelines. Among metabolites, alanine was the most commonly detected.

• Urine: The most frequently detected metabolites were 4-hydroxyhippurate and quinolinate. Notably, no studies included protein analysis of TBM urine samples. This is likely because normally functioning kidneys should remove proteins from urine, making it redundant to attempt to measure proteins in urine.

• Brain tissue: Only one study analysed proteins in TBM brain tissue. The proteins detected in significant quantities in the brain tissue were amphiphysin (AMPH), neurofascin (NFASC), and ferritin light chain (FTL). This study cannot be compared with others, as it was the only one to use human brain tissue, making it difficult to draw firm conclusions.

Age-related differences in TBM significantly altered proteins and metabolites

Although this review included adult and paediatric TBM cases, we did not observe distinct differences in the protein and metabolite profiles between these groups. This could be due to the limited number of paediatric studies, as most protein analyses were conducted on adult samples (six studies), and only one proteomics study focused on paediatric TBM. The role of commonly found metabolites and proteins was linked to inflammation.

Metabolites and their diagnostic potential

Metabolites tend to be predominantly elevated in disease states compared to healthy controls. Among the commonly identified metabolites, three were reported with high sensitivity and specificity for TBM from urine samples:

• 1-Methylnicotinamide: 100% sensitivity, 97.3% specificity. This metabolite has been identified in the urine of patients with pellagra, which is associated with a deficiency in vitamin B₃. It can be produced through the tryptophan pathway, utilising vitamins B₂ and B₆. The deficiency of these vitamins impairs the activity of kynurenine hydroxylase and kynureninase, which in turn reduces flux through the tryptophan metabolic pathway [31]. Vitamin B₃ deficiency also causes diarrhoea in patients and is associated with Pellagra.

• 3-Hydroxyisovalerate: 75% sensitivity, 97.3% specificity. This metabolite has been observed in higher amounts in the urine of individuals deficient in 3-methylcrotonyl-CoA (MCC). This deficiency notably occurs in leucine metabolism disorder and may cause vomiting, seizures, coma, and involuntary movements [32].

• N-Acetylglutamine: 100% sensitivity, 97.3% specificity. This metabolite is notably found in patients with rare genetic disorders with disrupted urea and is associated with N-acetylglutamate synthase (NAGS) deficiency [33].

These findings were reported in a single study [24]. Validation is required to confirm their diagnostic value.

Discussion

Role of key proteins and metabolites in TBM pathogenesis

Proteins associated with inflammation and neurological damage

Only one study analysed proteins in TBM brain tissue. Its findings provided insights into the potential role of these biomarkers in disease progression. Proteins such as Ferritin light chain (FTL), which is detected in TBM brain tissue, play a role in oxygen regulation in lipids and are upregulated in response to cell damage during pathogenic conditions, a process known as ferroptosis. Arachidonate 5-lipoxygenase (ALOX 5), although detected in CSF rather than brain tissue, ALOX5 is known to generate polyunsaturated fatty acids and oxygenated lipids that trigger cell death signalling [34, 35]. Although ALOX5 has been proposed as a TBM-specific biomarker [27], its presence requires further validation. ALOX5 also suggests a dysfunctional lipid metabolism, a relatively underexplored aspect of the metabolome in TBM. Amphiphysin is detected in TBM brain tissue. This protein is associated with neurological deficits and has been linked to stiff-person syndrome [36–38]. Its role in TBM is unclear, but it may contribute to the neck stiffness observed in severe cases. Neurofascin, also detected in TBM brain tissue, is essential for maintaining neuronal conduction and has been implicated in neurological disorders [39]. Its presence in TBM could explain the lack of coordination observed in severe cases. Glial fibrillary acidic protein (GFAP) and S100B are detected in the CSF; these proteins are well-established markers of neurological injury [40]. In one study, GFAP was found to be 100% specific to TBM [30], suggesting its potential as a highly selective biomarker. Although many of these proteins are not exclusive to TBM, their co-occurrence in TBM patient samples may be an essential diagnostic signature.

The commonly altered metabolites are alanine, isoleucine, myo-inositol, valine, glucose, lysine, creatinine, lactate, pyruvate, tyrosine, 1-methylnicotinamide, 3-hydroxyisovalerate, 4-hydroxyhippurate, acetate, choline, citrate, glutamate, glutamine, glycine, leucine, methionine, N-acetylglutamine, phenylalanine, proline, and quinolinate (Supplementary Table 2). Many of these metabolites, such as the amino acids, are involved in contributing to nutrient availability in the pathway involving host cell interactions [41], particularly, the branched chain amino acids (valine, leucine, and isoleucine) play important roles in the production of neurotransmitters, lack of which can lead to increased levels of tryptophan in the brain, which could lead to fatigue [42, 43]. Also, amino acids such as proline and isoleucine have been found to be involved in drug sensitivity or resistance in tuberculosis patients [44]. Quinolinate, a known neurotoxin, is a downstream metabolite of the kynurenine pathway and has been known to be associated with diseases such as Alzheimer’s and dementia [45]. Alanine, which has been identified for its neuroprotective role [46, 47], was mostly depleted in all reported TBM cases, which may be responsible for the rapid degeneration of the TBM patient’s mental health. These commonly altered amino acids could provide important insights into TBM.

Tuberculostearic acid (TBSA) has been historically recognized as a potential diagnostic marker for TBM, first reported in 1983 [48]. However, key early studies on TBSA were excluded from this review due to methodological limitations, such as the absence of control groups or a lack of distinction between TBM and non-TBM cases. Although these studies did not meet our inclusion criteria, the historical significance of TBSA warrants its mention in the context of TBM biomarker development.

Recommendations

Our review found that selecting a single biomarker across all sample types may prove challenging due to the differences observed in metabolites and proteins among sample types. Perhaps, rather than focusing on a single biomarker, the occurrence of multiple biomarkers per sample type could be a more accurate indicator of disease severity. Additionally, rather than focusing on a single sample type, using various samples that are less invasive, such as urine (which has recently been utilised) or serum, could be beneficial in the rapid diagnosis of TBM disease.

When considering a disease such as TBM, the disease stage is also essential, as different stages will reflect a distinct set of metabolites or proteins that are most likely to be present. This information would also help to appropriately profile the sample type, thereby providing insights that could arise from the metabolite and protein profiles as pointers to the TBM disease.

The Lancet consensus and Thwaites’ scoring system have been used in conjunction with validated clinical diagnostic markers, achieving varying degrees of success. We propose that once the proteins and metabolites associated with TBM have been validated in larger cohort studies, they could be incorporated into future scoring systems to enhance TBM diagnosis.

We also found that many proteins and metabolites that have been significantly altered have yet to be validated. Therefore, we propose that more validation studies be conducted. Lastly, the mode of reporting results obtained in TBM studies should be standardised. For studies characterising TBM, results that measure accurate concentrations should be considered.

Limitations

Although the search strategy incorporated both conventional keywords and MeSH terms relevant to tuberculous meningitis and molecular biomarkers, some relevant studies—particularly older ones—may have been missed due to indexing or terminology inconsistencies. For example, a notable study on tuberculostearic acid published in 1983 was not retrieved by the search engines [48]. This highlights the challenge of capturing all relevant literature using keyword-based methods alone. Given that systematic reviews are highly labour-intensive and require defined, reproducible criteria to ensure methodological transparency, our approach represents a necessary balance between inclusivity and feasibility.

This review has several other limitations. Firstly, no protein analysis was performed on the TBM urine samples. Additionally, all urine studies were conducted in the same region. Additionally, only one study on brain tissue proteins was conducted. This restricts conclusions about their diagnostic value in these sample types. Thus, there is a need for further studies on these sample types. The majority of the protein analyses were performed in CSF or serum, which may not capture the full range of TBM biomarkers.

Secondly, there is a lack of validation for the identified differentially altered proteins. While proteins ALOX5, APOB, and GFAP, as well as various metabolites, were frequently detected, their specificity to TBM has not yet been confirmed. However, APOB has been closely associated with cardiovascular diseases. Therefore, they cannot be categorised as biomarkers unless validation studies have been conducted. Additionally, only one study provided sensitivity and specificity data for the key metabolites, including 1-methylnicotinamide and 3-hydroxyisovalerate, making them candidate biomarkers. However, further validation studies are necessary to identify additional metabolites and proteins that could serve as biomarkers.

Thirdly, there is an imbalance in paediatric vs. adult studies. Most protein-focused studies were conducted in adults, while only one study examined proteins in paediatric TBM. This makes it difficult to determine if metabolic or proteomic profiles differ between children and adults with TBM.

Fourthly, sample collection and the analytical methods used vary, as differences in sample processing, analytical pipelines, and study designs across included studies could introduce bias or inconsistencies in biomarker detection. On the other hand, however, this does allow for a broader scope of novel metabolites to be detected. Considering this, standardised biomarker validation protocols are needed to ensure reproducibility when using multiple instruments.

Lastly, a meta-analysis on this topic could not be conducted due to the heterogeneity in the reported results, as some studies did not include the concentrations of the metabolites or proteins. In contrast, others reported on specificity and sensitivity (Supplementary Table 1).

Conclusion

Our systematic review highlights a set of potential protein and metabolite biomarkers for TBM. However, their specificity remains challenging because they have not been tested against other neurological infections. Research in areas such as urine and tissue could expand our knowledge of the diagnostic ability of these sample types for TBM. Further validation studies are required to determine whether a biomarker panel, rather than a single marker, would provide the most accurate diagnostic tool for TBM. Well-validated studies may lead to the identification of putative biosignatures that can be used for diagnosing TBM, assessing disease severity, or monitoring treatment response, thereby providing data that can enhance our understanding of the pathophysiology of TBM.

Authors’ contributions

ARI wrote the original manuscript and generated the images. All co-authors (MW, DTL, AAW, NNC, MTvF, MvdK, SM) provided feedback, comments, and/or edits on various drafts of the manuscript. All authors (ARI, MW, DTL, AAW, NNC, MTvF, MvdK, SM) approved of the final version of this manuscript.

Funding

This research was funded by the Amsterdam Institute for Infection and Immunity.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

This systematic review has been approved by the North-West University ethics committee (number NWU-00073-25-A1) and has been registered on PROSPERO (number CRD420250626311). Consent to participate is not applicable.

Consent for publication

All authors consent to publication.

Competing interests

The authors declare no competing interests.