The protocol of the present study and the procedures employed for sample collection were approved by the Ethics Committee of Chongqing Medical University and have therefore been performed in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments. Written informed consent was obtained from all participants. Twelve TBM subjects were diagnosed following the standard criteria (21) and enrolled in the Department of Neurology at the First Affiliated Hospital of Chongqing Medical University. Twelve healthy control (HC) subjects were enrolled after written informed consent was obtained in the medical examination center at the First Affiliated Hospital of Chongqing Medical University. The HC subjects were required to have neither active TB nor any neurological complaints.

Pooled CSF samples were generated by combining equal volumes of the 12 individual plasma samples from each group. Each sample (5 ml) was desalted and concentrated to 80 ml in a Vivaspin spin column with a molecular weight cut-off of 5 kDa (Vivascience AG, Hannover, Germany). Then, 120 μg of protein from each sample was dissolved in 30 μl of STD buffer (4% SDS, 100 mM DTT, 150 mM Tris-HCl pH 8.0), boiled in water for 5 min, cooled to room temperature, diluted with 200 μl of UA buffer (8 M Urea, 150 mM Tris-HCl pH 8.0), and transferred to 30 kDa ultrafiltration. The samples were centrifuged at 14,000 × g for 15 min and 200 μl of UA buffer was added. The samples were centrifuged again for 15 min under the same conditions. Then, 100 μl of 0.05 M iodoacetamide in UA buffer was added, and the samples were incubated for 20 min in the dark. After 10 min of centrifugation at the above conditions, the filters were washed three times with 100 μl of UA buffer. Then, 100 μl of DS buffer (50 mM triethylammoniumbicarbonate, pH 8.5) was added to the filters, and the samples were centrifuged for 10 min at the above conditions. This step was repeated twice. Subsequently, 2 μg of trypsin (Promega, Madison, WI, USA) in 40 μl of DS buffer was added to each filter, and the samples were incubated overnight at 37°C. The resulting peptides were collected by centrifugation. The filters were rinsed with 40 μl of 10X DS buffer again. iTRAQ™ labeling was performed according to the manufacturer’s instructions (Applied Biosystems, Foster City, CA, USA). As each sample consisted of a pool of CSF from 12 TBM subjects or 12 HC subjects, iTRAQ™ reagents 114 and 116 were applied to the HC sample and reagents 115 and 117 were applied to the TBM sample. The analytic processes, including plasma depletion, protein digestion, iTRAQ™ labeling, SCX fractionation and LC-MS/MS analysis, were repeated twice.

Prior to liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS) analysis, the peptides were purified from excess labeling reagent by strong cation exchange (SCX) chromatography. Briefly, the peptides were dried in a vacuum concentrator, dissolved in strong cation exchange buffer A (10 mM KH₂PO₄ in 25% of ACN, pH 3.0), and loaded onto a Polysulfoethyl 4.6 × 100 mm column (5 μm, 200 Å; PolyLC Inc., Columbia, MD, USA) at a flow rate of 1 ml/min. A suitable gradient elution was applied to separate peptides at a flow rate of 1 ml/min with elution buffer (10 mM KH₂PO₄, 500 mM KCl in 25% acetonitrile, pH 3.0). Eluted peptides were collected and desalted by an offline fraction collector. The resulting fractions were combined to 10 pools and desalted on C18 Cartridges [Empore™ SPE Cartridges C18 (standard density), bed I.D. 7 mm, volume 3 ml; Sigma, St. Louis, MO, USA]. Each final fraction was concentrated by a vacuum concentrator and resuspended in 40 μl of 0.1% (v/v) trifluoroacetic acid. All the samples were stored at −80°C until subsequent LC-MS/MS analysis.

For each fraction, 10 μl of solution was injected for nano LC-MS/MS analysis using an AB SCIEX TripleTOF 5600 MS (Toronto, Concord, ON, Canada) equipped with a splitless Eksigent nanoUltra 2D Plus nanoLC system and a cHiPLC Nanoflex microchip system (Eksigent, Dublin, CA, USA). The Nanoflex system uses replaceable microfluidic traps and columns packed with ChromXP C18 (3 μm, 120 Å) for online trapping, desalting and analytical separations. The sample was loaded, and trapping and desalting were performed at 2 μl/min for 10 min with 100% mobile phase A (2% acetonitrile/0.2% formic acid/98% water). For peptide elution, the gradient was initiated at 5% mobile phase B (98% acetonitrile/0.2% formic acid/2% water) and increased to 24% linearly in 70 min at a flow rate of 300 nl/min. The total gradient length was 120 min. MS data acquisition was performed in the information-dependent acquisition (IDA) mode. For TOF MS scans, the TripleTOF 5600 MS was operated with a resolving power of 30,000 FWHM. IDA survey scans were acquired in 250 msec with a mass range of m/z 350–1250. Specifically, 30×100 msec MS/MS were followed at a 120 cps (counts/sec) trigger with a precursor charge state of +2 to +5. Dynamic exclusion was set for 18 sec. Collision energies were calculated on-the-fly for all precursor ions using empirical equations based on mass and charge, and the Enhance iTRAQ™ function was turned on to improve the efficiency of the collision-induced dissociation.

All LC-MS/MS data were processed using ProteinPilot v4.2 and used to search the Uniprot_HUMAN.20140108_134997 database using the ‘thorough search’ option. The ‘iTRAQ™ 8-plex peptide labeled’ sample type and a ‘biological modification ID focus’ were selected for the analysis method. Trypsin was selected as the digestion enzyme, with cysteine alkylation by iodoacetamide as a modification. Data were normalized by ProteinPilot. The reported data were based on a false discovery rate (FDR) ≤1% confidence for protein identification. Differential proteins were analyzed by ProteinPilot for significant downregulation or upregulation. A significance threshold of 95% was selected as a criterion for each individual experiment. Peptides that matched multiple proteins were not included in the software quantification. The final ratios were calculated from the average of the ratios obtained in the two independent experiments in which the protein abundance was significant (P<0.05). In each experiment, bias correction for unequal mixing in the differently labeled samples was performed based on the assumption that most proteins do not change in expression. Thus, if the sample from each experimental condition was not combined in exactly equal amounts, this bias correction would fix the systematic error. The software identifies the median average protein ratio and corrects it to unity, and then applies this factor to all quantification results.

Functional clustering was conducted using DAVID Bioinformatics Resources v6.7 (http://david.abcc.ncifcrf.gov/home.jsp) (22). A functional annotation analysis was performed with the Gene Ontology tool (GOTERM_CC_ALL) using the UniProt accession numbers for all the differential proteins. Only those GO terms in each gene list that yielded a P-value of <0.05 using Fisher’s exact test were considered significantly enriched. A Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis was subsequently performed using the same software, in which a corrected P-value of <0.05 was deemed to indicate a statistically significant difference.

Western blotting was performed on the 12 individual TBM samples and 12 individual HC samples. CSF samples were separated by 10% SDS-PAGE, blocked in 5% skim milk in TBST, and transferred onto PVDF membranes. The membranes were incubated overnight at 4°C with an anti-NELL2 (diluted 1:1,000; Abcam, Cambridge, MA, USA) primary antibody. The membranes were washed and incubated with goat polyclonal to rabbit IgG horseradish peroxidase-coupled secondary antibody (diluted 1:10,000; Bio-Rad Laboratories, Hercules, CA, USA). After extensive washing, the protein bands detected by the antibodies were visualized by the enhanced chemiluminescence method and exposed to autoradiography film. After immunodetection, the membranes were stained with Ponceau S as an internal control for normalization. The western blot signals were densitometrically quantified in each sample with Quantity One (Bio-Rad Laboratories).

Statistical analysis was performed using the Statistical Package of Social Science (SPSS) for Windows v19.0. Data are expressed as the means ± SD. The Student’s t-test was applied to identify proteins with significant abundance differences between the TBM and HC subjects. Areas under the receiver operating characteristic (ROC) curve (AUC) were calculated to evaluate classifier performance. All the tests were two-tailed. The significance threshold was set at P<0.05.

Demographic and clinical data obtained from TBM and HC subjects are provided in Table I. The two groups of subjects were not differentiable by any key demographic characteristics, such as age, gender or body mass index (BMI).

Data from the two MS repetitions identified 572 unique proteins according to the previously described parameter set. Only the 81 differential proteins with a ±1.2-fold-change and P<0.05 were selected for further analysis (Table II). The threshold criteria were selected based on literature investigating the reproducibility of iTRAQ™ quantification (23).

Some differential proteins exhibited multiple functions and subcellular locations, which resulted in individual proteins being counted more than once. Considering that a given protein can be attributed to several functions and/or processes, we categorized differential proteins by their biological processes (Fig. 1A) and molecular functions (Fig. 1B) by pie chart. ‘Response to stimulus’, ‘biological adhesion’ and ‘biological regulation’ were the most over-represented biological processes. ‘Enzyme inhibitor activity’ was the most over-represented molecular function, distantly followed by ‘protein binding’ and ‘cell surface binding’.

The 81 differential proteins were analyzed for KEGG over-representation of pathways (‘proteomic phenotyping’) (24) to obtain insights into the functional differences between the CSF proteomes of TBM and HC subjects. The top three-ranking canonical KEGG pathways are listed in Table II. Inflammatory processes were the most statistically over-represented, including complement and coagulation cascades ranking first (P<0.001) and cell adhesion molecules (CAMs) ranking third (P<0.01).

The neuron-specific differential protein with the largest absolute fold-change value, neural epidermal growth factor-like like 2 (NELL2), was selected for western blot validation (Table II). In accordance with iTRAQ™, NELL2 decreased significantly in TBM compared to HC subjects (Fig. 2A). ROC curves were constructed using the western blotting data, in which the AUC, sensitivity, and specificity were 0.868, 83.3 and 75%, respectively (Fig. 2B).

KEGG analysis revealed over-representation in two inflammation-associated processes, namely complement and coagulation cascades and CAMs (Table III). As MTB infects and persists inside host macrophages, MTB infection is naturally accompanied by a host inflammatory response (25). In the initial stage, MTB activates several pro-inflammatory cytokines, including tumor necrosis factor-α (TNF-α) (26), interleukin 1 (IL-1) (27), and reactive oxygen/nitrogen intermediaries (ROI/RNI) that promote MTB termination (28). Simultaneously, MTB induces the release of anti-inflammatory molecules, such as transforming growth factor-β (TGF-β) (29) and IL-10 (30), to suppress the pro-inflammatory response and maintain MTB survival. In ~90% of MTB cases, the host inflammatory response results in granuloma formation, which is sufficient to contain the infection and prevent disease. However, in 5–10% of MTB cases, the pathogen evades or subverts the host’s granulomatous inflammatory response, producing either a latent or active infection (31). Therefore, the outcome of MTB infection depends on a complex set of host-pathogen interactions, and further investigation into the differential proteins associated with the host inflammatory response can offer insight into the pathogenesis and treatment of TBM. For example, drugs that intervene in the host inflammatory response, such as TGF-β blockers, have been suggested as a method to prevent the spread of MTB infection (32).

Two significantly altered proteins including neogenin and fructose-bisphosphate aldolase related to glycolysis/gluconeogenesis were identified. MTB prefers to utilize host fatty acids as an energy source. Thus, glycolytic/gluconeogenic proteins that are involved in fatty acid utilization would be required for the growth and persistence of MTB in vivo (33).

TBM is characterized by meningeal inflammation, CSF flow obstruction, fibrogelatinous exudates, and arterial vasculitis that traverse the exudates, resulting in cerebral infarction (34,35). TBM sometimes results in tuberculoma formation (36), which is most commonly observed in the cerebellum, cerebrum and pons (37). In the present study, by magnetic resonance imaging (MRI), the 12 TBM subjects excibited basal meningeal enhancement consistent with TBM, seven of whom excibited cerebral infarction shown by high DWI signal intensity. These findings suggest a strong possibility of brain neuronal cell damage via vasculitis and tuberculoma formation. Considering that CSF neuron-specific enolase (NSE) is the most commonly applied biomarker of nervous tissue damage and can act as an early diagnostic marker of TBM (38,39), we selected NELL2 for western blotting validation due to its neuronal specificity and high absolute fold-change value. NELL2, also known as neural epidermal growth factor-like like 2, NEL-like protein 2, Nel-related protein 2 (NRP2), and protein kinase C-binding protein, is a cytoplasmic enzyme that is expressed almost exclusively in the nervous system (40), predominantly in neurons with the highest density in the hippocampus and cerebral cortex (41). NELL2 promotes neuronal differentiation and survival during embryonic development (42) through mitogen-activated protein kinases (MAPK) (43), stimulates motor and sensory neuron differentiation in dorsal root ganglia (44), and promotes the differentiation of neural progenitor cells into neurons (45). NELL2 knock-out mice have demonstrated learning impairment in the Morris water maze task.

The present study revealed significantly decreased levels of NELL2 in TBM patients. As neuron-produced NELL2 supports neuronal survival and differentiation and neurons are compromised in TBM, it is reasonable to surmise that the reduced NELL2 levels observed in TBM patients result from brain neuronal cell damage or death. Additional study is required to determine whether there is a TBM disease stage in which NELL2 significantly increases to compensate for brain neuronal cell damage.

There are several limitations to the present study. First, eight of the 12 TBM patients also had pulmonary TB, and another TBM patient had vertebral TB. As 25% of miliary TB patients also have TBM, presumably by transmission across the blood-brain barrier (46), future studies should include TBM patient cohorts with and without pulmonary TB, which may provide additional novel biomarkers of TBM. Second, the World Health Organization (WHO) recommends that all newly diagnosed human immunodeficiency virus (HIV) and syphilis patients should be tested for TB, as individuals with HIV or syphilis and latent TB infection are at a significantly higher risk of developing active TB (47). However, since no subjects of the present study were co-infected with HIV or syphilis, future studies should include cohorts of HIV or syphilis co-infected patients to improve our understanding of MTB host interaction in these patient populations. Third, the influence of anti-TB agents on differential protein expression was not assessed and should be taken into consideration in future studies. Fourth, in order to improve specificity for TBM, the current findings require further validation in larger sample sets that include other intracranial infectious diseases.

In conclusion, an iTRAQ™-based quantitative proteomic approach was utilized in the present study to identify differential proteins in the CSF of TBM patients and healthy controls. A total of 111 differential proteins were identified, and NELL2 was validated by western blotting as a promising diagnostic biomarker for TBM. Further evaluation of these findings in larger studies including medicated and unmedicated cohorts with other intracranial infectious diseases is required for clinical translation.