Abstract

   Despite anti-tuberculous treatment (ATT), central nervous system
   tuberculosis (CNS-TB) still causes permanent neurological deficits and
   death. To identify prognostic factors, we profiled a prospective cohort
   of pediatric HIV-negative tuberculous meningitis (TBM) and non-TBM
   patients. We found significantly increased cerebrospinal fluid (CSF)
   matrix metalloproteinases (MMPs) and neutrophil extracellular traps
   (NETs) in TBM patients with neuroradiological abnormalities and poor
   outcomes. To dissect mechanisms, we used our existing CNS-TB murine
   model, which shows neutrophil-rich necrotizing pyogranulomas with MMP-9
   and NETs colocalizing, as observed in human CNS-TB pathology. Spatial
   transcriptomic analysis of both human and murine CNS-TB demonstrates a
   highly-inflamed and neutrophil-rich microenvironment of inflammatory
   immune responses, extracellular matrix degradation and angiogenesis
   within CNS-TB granulomas. Murine CNS-TB treated with ATT and MMP
   inhibitors SB-3CT or doxycycline show significantly suppressed NETs
   with improved survival. MMP inhibition arms show attenuated
   inflammation and well-formed blood vessels within granulomas.
   Adjunctive doxycycline is highly promising to improve CNS-TB outcomes
   and survival.

Graphical abstract

   [84]graphic file with name 12974_2025_3548_Figa_HTML.jpg

Supplementary Information

   The online version contains supplementary material available at
   10.1186/s12974-025-03548-7.

   Keywords: Central nervous system tuberculosis, Matrix
   metalloproteinases, Neutrophil extracellular trap, Doxycycline,
   Host-directed therapy

Introduction

   Central nervous system tuberculosis (CNS-TB) is a devastating, often
   fatal disease that disproportionately afflicts children and
   immunocompromised individuals [[85]1]. Immunopathology in CNS-TB,
   characterized by extensive brain inflammation and tissue destruction,
   is driven by a matrix-degrading phenotype, which results from an
   increase in matrix metalloproteinases (MMPs) relative to their
   tissue-specific inhibitors (tissue inhibitor of metalloproteinases,
   TIMPs) [[86]2, [87]3]. MMPs degrade extracellular matrix (ECM) and are
   key to blood-brain barrier (BBB) breakdown and brain tissue damage
   [[88]4]. In particular, MMP-9, a gelatinase responsible for the
   degradation of type IV collagen, the major ECM component of the BBB,
   was found to be increased in the cerebrospinal fluid (CSF) of adult
   CNS-TB patients, which was associated with neurological deficit and
   death [[89]2]. MMPs are implicated in driving CNS-TB immunopathology.

   Ischemic strokes occur in up to 45% of CNS-TB patients [[90]5], but the
   mechanism of thrombosis leading to strokes in CNS-TB remains
   uncharacterized. Neutrophil extracellular traps (NETs), which are a
   meshwork of chromatin fibers studded with antimicrobial peptides and
   proteases, may be implicated in the crosstalk between inflammation,
   infection and thrombosis [[91]4, [92]6]. During the release of NETs,
   protein arginine deiminase 4 (PAD4) is activated, which converts
   arginine to citrulline in core histones (CitH3). Citrullination of
   histones allows chromatin to undergo decondensation and be dispersed in
   the form of NETs [[93]7]. Upon Mycobacterium tuberculosis (M. tb)
   infection, neutrophils produce NETs to trap mycobacteria but are unable
   to kill the pathogen [[94]8]. Instead, NETs can cause local host tissue
   damage or thrombosis if present in the circulation. The presence of
   NETs in respiratory samples and necrotic lung lesions of pulmonary TB
   patients implicates NETs in pulmonary TB pathogenesis [[95]9, [96]10].
   NET formation is also a hallmark of caseating granulomas in pulmonary
   TB [[97]11]. However, the pathological role of NETs in CNS-TB remains
   to be explored.

   To better understand the immunopathogenesis in CNS-TB, we first studied
   human clinical samples and then developed a murine CNS-TB model that
   exhibited the brain pathology and neurological symptoms similar to
   those observed in humans [[98]12]. Concurrently, we examined the role
   of MMPs and NETs in human CNS-TB immunopathology in driving CNS-TB
   disease severity. CSF MMPs and NETs in a cohort of human pediatric
   tuberculous meningitis (TBM) patients were analyzed together with
   neuroimaging and compared to clinical outcomes of severe neurological
   deficit and death. Spatial transcriptomic profiling elucidated the
   differential gene expression and pathways associated with human and
   murine CNS-TB immunopathology. Critically, we demonstrate the efficacy
   of adjunctive MMP inhibition together with anti-TB treatment (ATT) as
   host-directed therapy in the murine CNS-TB model.

Results

Up-regulated human CSF MMPs and NETs strongly associate with
neuroradiological abnormalities and poor clinical outcomes in human TBM

   CSF from 72 human pediatric patients with TBM and non-TBM causes of CNS
   diseases (Supplemental Fig. S1, Supplemental Table S2, S3) were
   analyzed for expression of MMPs and their functional activity, NETs
   with neuroimaging and clinical outcomes (Fig. [99]1A). None of the TBM
   patients had HIV nor were immunosuppressed. Consistent with previous
   reports [[100]2, [101]13–[102]17], MMP-9 is higher in TBM by 11.3-fold
   (P = 0.0007) (Fig. [103]1B). MMP-2 and Extracellular MMP inducer
   (EMMPRIN), an activator of MMPs, are 3.5-fold (P = 0.0369) and 7.3-fold
   (P = 0.0182) higher in TBM respectively (Fig. [104]1B). Since net
   proteolysis is affected by relative local MMPs and TIMPs
   concentrations, we also determined the concentrations of TIMP-2 and
   − 4, which inhibit MMP-2 and − 9 respectively [[105]18, [106]19]. CSF
   TIMP-2 (P = 0.0047) and TIMP-4 (P = 0.0338) concentrations in TBM
   patients are significantly down-regulated relative to non-TBM patients
   (Fig. [107]1C). This disproportionate increase in CSF gelatinases
   relative to TIMPs extends the previous finding in adult TBM patients
   [[108]2], and demonstrates the presence of a matrix-degrading phenotype
   in both pediatric and adult CNS-TB. CSF MMP-2 and − 9 of TBM patients
   are demonstrated to be functionally active and the addition of
   neutralizing antibodies significantly reduces the functional gelatinase
   activity (MMP-2, P = 0.0078; MMP-9, P = 0.0007), confirming that the
   ECM degradation in the CSF is in part driven by MMP-2 and − 9
   (Fig. [109]1D).

Fig. 1.

   [110]Fig. 1
   [111]Open in a new tab

   Up-regulated MMP and NET concentration in human pediatric TBM patients
   are strongly associated with neuroradiological abnormalities and poor
   clinical outcomes. A CSF samples from a pediatric cohort of TBM
   patients and non-TBM patients were analyzed for the expression of MMPs,
   NETs, and functional gelatinase activity. The CT brain images of TBM
   patients were assessed for radiological abnormalities together with
   clinical outcomes. Illustration created using Biorender. B CSF
   concentrations of MMP-9, MMP-2 and EMMPRIN are upregulated in human
   pediatric TBM. C CSF concentrations of TIMP-2 and TIMP-4 are
   downregulated in TBM. In B and C, bars represent median ± IQR with
   analysis by Mann-Whitney test. Non-TBM, n = 52; TBM, n = 20. D MMP-2
   and − 9 neutralization significantly suppress CSF gelatinase activity
   in TBM patients. Analysis by paired t-test. MMP-2, n = 4; MMP-9, n = 6.
   E Representative CT brain imaging of human pediatric TBM patients. Left
   panel shows normal CT brain in a pediatric TBM patient. Right panel
   shows widespread leptomeningeal enhancement (arrow) and dilated
   ventricles from hydrocephalus (H) in a pediatric TBM patient.
   F Increased CSF MMPs significantly associate with leptomeningeal
   enhancement, ventricular dilatation and poor clinical outcomes in human
   TBM. Leptomeningeal enhancement, n = 14; ventricular dilatation,
   n = 14; clinical outcomes n = 17. Bars represent median ± interquartile
   range (IQR), dots represent individual values. Analysis by Mann-Whitney
   test. G NETs marker extracellular DNA (n = 52 non-TBM, 20 TBM) and
   citH3 (n = 6 per group) are increased in human TBM. Bars represent
   median ± IQR. Analysis by Mann-Whitney test. H CSF CitH3 concentration
   strongly associates with CSF extracellular DNA concentration and CSF
   MMP-9 concentrations. TBM patients form a cluster with high NET and
   MMP-9 concentrations (circled). Analysis by Pearson correlation
   coefficient. I High CSF extracellular DNA significantly associate with
   leptomeningeal enhancement, ventricular dilation and poor clinical
   outcomes in human TBM. Bars represent median ± IQR, dots represent
   individual values. Statistical analysis was conducted using
   Mann-Whitney test. * P < 0.05; ** P < 0.01, *** P < 0.001

   Next, we determined the neuroradiological and clinical significance of
   CSF MMPs in these TBM patients. Contrast enhancement, a marker of BBB
   disruption, and ventricular dilatation (hydrocephalus) are established
   poor prognostic factors [[112]20] which are observed in TBM patients
   (Fig. [113]1E). Four MMPs are found to be significantly up-regulated in
   TBM patients with leptomeningeal contrast enhancement (collagenase
   MMP-1, P = 0.0027; stromelysin MMP-3, P = 0.0213; matrilysin MMP-7,
   P = 0.008; stromelysin MMP-10, P = 0.0007) and ventricular dilatation
   (MMP-1, P = 0.0007; MMP-3, P = 0.001; MMP-7, P = 0.0007; MMP-10,
   P = 0.0013) (Fig. [114]1F). In patients with poor clinical outcomes of
   severe neurological deficit or death, CSF MMP-7 (P = 0.0185) and MMP-10
   (P = 0.0243) are significantly higher (Fig. [115]1F). These findings
   highlight that increased CSF MMP concentrations are associated with
   neuroradiological abnormalities and poor clinical outcomes of severe
   neurological deficit and death in pediatric TBM patients.

   As NETs are implicated in TB pathogenesis, we next determined NETs in
   the human cohort and found extracellular DNA to be 4.9-fold higher in
   TBM (P = 0.0006) (Fig. [116]1G). Since extracellular DNA concentration
   may include bacterial DNA, we further measured citrullinated histone 3
   (CitH3), a NETs marker, and verified a 6-fold higher of CSF CitH3 in
   TBM (P = 0.0152) (Fig. [117]1G). The significant positive association
   between CSF CitH3 and extracellular DNA concentrations (r = 0.67,
   P = 0.016) indicate that extracellular DNA is a good surrogate marker
   for NETs (Fig. [118]1H). MMP-9, which degrades type IV collagen in the
   BBB and is established to be associated with NETs [[119]8, [120]21,
   [121]22], also demonstrates a significant and strong positive
   correlation with CitH3 (r = 0.85, P = 0.0005), with TBM patients
   forming a cluster with higher NETs and MMP-9 (Fig. [122]1H).

   Next, we determined whether elevated CSF NETs are a poor clinical
   prognostic factor. Independent of etiology, we observe no significant
   correlation of CSF NETs with GCS score (r^s = −0.31, P = 0.104) and BCS
   score (r^s = 0.022, P = 0.889) (Supplemental Fig. S2). Interestingly,
   we consistently found elevated CSF extracellular DNA in TBM patients
   with leptomeningeal enhancement (P = 0.0013), ventricular dilatation
   (P = 0.0193) and poor clinical outcomes (P = 0.002) (Fig. [123]1I).
   These findings indicate that raised CSF NETs is a marker of poor
   clinical prognosis in human pediatric TBM.

   We further re-analyzed the human cohort by including only the definite
   and probable TBM patients (M. tuberculosis isolated elsewhere with
   clinical meningitis) in TBM group (TBM group n = 11 and non-TBM group
   n = 52) and the same trend is observed; the elevated MMP-9 and NETs,
   decreased TIMP-2/−4 in the CSF of TBM patients, and the significant
   association of raised MMPs and NETs with leptomeningeal enhancement,
   ventricular dilation and poor clinical outcomes in TBM patients
   (Supplemental Fig. S3). Comparing definite and probable TBM (n = 11) to
   the microbiologically-confirmed non-TBM group (n = 25), we still found
   increased MMP-2, −9, EMMPRIN, NETs, and decreased TIMP-2, −4
   expressions in the CSF (Supplemental Fig. S4), substantiating our
   previous findings.

MMP-2 and − 9 are upregulated in M. tb-infected C57BL/6 Nos2^-/- mice with
increased extracellular matrix degradation and NETs

   Our previous work demonstrated that compared with C3HeB/FeJ mice,
   C57BL/6 Nos2^−/− mice is a better murine model of CNS-TB, as they
   developed more necrotizing TB granulomas in the brain, exhibited worse
   neurological signs and had a higher expression of pro-inflammatory
   mediators [[124]12, [125]23]. Since CNS-TB in humans is uniformly fatal
   without TB treatment, instead of the CNS-TB C3HeB/FeJ mice which failed
   to show mortality, we selected the C57BL/6 Nos2^−/− mice CNS-TB model
   which displayed higher mortality to test CNS-TB therapeutics.

   In this study, we analyzed the MMPs expression between the two murine
   models, a comparison which has not been investigated previously. Mice
   were infected with 10^5 CFU M. tb intracerebroventricularly, and the
   brain tissues and homogenates were analyzed eight weeks after the
   infection. We found significant upregulation of MMP-2 (3.3-fold,
   P = 0.0075), MMP-9 (40-fold, P = 0.011), MMP-3 (21-fold, P = 0.013) and
   MMP-12 (49-fold, P = 0.016) concentrations in Nos2^−/− mice compared to
   C3HeB/FeJ mice (Supplemental Fig. [126]S5). We replicated the study in
   C57BL/6 Nos2^−/− mice with a larger sample size (n = 6 each) and found
   significantly increased MMP-2 (2.8-fold, P = 0.0022), MMP-9
   (154.2-fold, P = 0.0022), MMP-3 (347.3-fold, P = 0.0022) and MMP-12
   (122.3-fold, P = 0.0022) upon M. tb infection (Fig. [127]2A). In M.
   tb-infected mice, while TIMP-4 concentration remains unchanged, TIMP-2
   is significantly increased (P = 0.0411) (Fig. [128]2B). Overall, these
   findings support the use of C57BL/6 Nos2^−/− mice as an inflammatory
   murine CNS-TB model with a destructive matrix-degrading phenotype and
   profile.

Fig. 2.

   [129]Fig. 2
   [130]Open in a new tab

   Functional gelatinase activity is increased with H37Rv infection and is
   associated with MMP-2 and − 9 concentrations in murine CNS-TB. Six- to
   eight-week-old male C57BL6 Nos2^−/− were infected with 10^5 CFU M. tb
   H37Rv via intracerebroventricular (i.c.vent). route, control mice
   received saline (n = 6). Eight weeks post-infection, the brain tissues
   and homogenates were used for downstream assay. Brain concentrations of
   (A) gelatinase MMP-2, −9, stromelysin MMP-3, and elastase MMP-12, and
   (B) TIMP-2 and − 4 in saline control versus H37Rv i.c.vent.-infected
   Nos2^−/− mice. MMP and TIMP concentrations are normalized to total
   protein concentration. Bars represent median ± IQR, and analysis was
   conducted using Mann-Whitney test. C Functional gelatinase activity in
   the brain homogenates of H37Rv i.c.vent.-infected Nos2^−/− mice is
   significantly increased. Bars represent mean ± standard deviation (SD).
   Analysis by unpaired t-test. D Brain gelatinase activity is associated
   with MMP-9 concentration, but shows no significant association with
   MMP-2 in M. tb-infected Nos2^−/− mice, analyzed by Spearman’s
   correlation coefficient. E Brain concentration of NETs marker CitH3 is
   significantly up-regulated in H37Rv i.c.vent.-infected Nos2^−/− mice
   relative to saline control. Bars represent median ± IQR. Analysis by
   Mann-Whitney test. F Brain MMP-9 is associated with NETs marker CitH3
   concentrations in M. tb-infected Nos2^−/− mice. Analysis by Spearman’s
   correlation coefficient. * P < 0.05; ** P < 0.01. G Representative
   immunohistochemistry (IHC) stains of granulomas in human CNS-TB (n = 6)
   and murine CNS-TB (n = 6) show expression of MMP-9 and CitH3 at the
   necrotic area. Top panel, scale bar = 100 μm; Bottom panel, scale bar
   = 200 μm

   Next, we sought to determine the net proteolytic activity in the murine
   brain by determining the functional gelatin degradation. Brain
   gelatinase activity is significantly increased in M. tb-infected mice
   by 5.5-fold (P = 0.0033) (Fig. [131]2C). Gelatinase activity is
   significantly and strongly associated with MMP-9 concentration (r^s =
   0.89, P = 0.033), but not with MMP-2 (r^s = 0.37, P = 0.4972) in M.
   tb-infected Nos2^−/− mice (Fig. [132]2D). This indicates that M. tb-
   infection of murine CNS-TB brain result in the development of a tissue
   destructive microenvironment.

   We also measured NETs, which are elevated in the CSF of human CNS-TB,
   in murine brain homogenates. CitH3 is significantly upregulated in M.
   tb-infected mice by 1.9-fold (P = 0.0022) (Fig. [133]2E). Similar to
   human TBM, MMP-9 demonstrates a strong association with CitH3 (r^s =
   0.94, P = 0.0167) in M. tb-infected Nos2^−/− mice (Fig. [134]2F).
   Immunohistochemical staining confirmed high expression of MMP-9 and
   CitH3 in the necrotic region of tuberculomas in both human and murine
   CNS-TB (Fig. [135]2G, Supplemental Fig. S6 and S7). Together, the
   upregulated MMPs and NETs during infection, which is similar to the
   human CNS-TB cohort, supports the use of this murine CNS-TB model to
   further evaluate immunopathology and therapeutics.

Spatial transcriptomics analysis reveals human and murine CNS-TB granuloma
microenvironments are highly inflamed with elevated immune, cytokine and
defense responses

   To further determine the underlying drivers of immunopathology within
   the tissue microenvironment, we performed laser capture microdissection
   RNA sequencing (LCM RNA-seq) to profile the transcriptome of granulomas
   and inflammatory regions in human and murine CNS-TB (Fig. [136]3A).
   Since obtaining healthy human brain tissues are unethical, we used
   adjacent normal brain tissue of human CNS-TB samples and further
   validated with human demyelination tissues [[137]24] (Supplemental Fig.
   S8A). Normal brain tissues from healthy mice were used as control.
   Quality checks of the data indicate no significant difference in total
   genes between groups (Supplemental Fig. S8B, S9A and Supplemental File
   S1). Principal component analysis (PCA) of all expressed genes show
   clustering of CNS-TB granuloma and inflammatory groups, differentiating
   them from adjacent normal and demyelination groups (Fig. [138]3B).
   Similarly, hierarchical clustering performed using the top 500 variable
   genes expressed across the human samples distinguish the gene
   expression of human CNS-TB granulomas from CNS-TB inflammatory,
   adjacent normal tissues and demyelination tissues (Fig. [139]3C and
   Supplemental File S2). In human CNS-TB granulomas, gene clusters show
   enrichment of biological processes related to epithelial mesenchymal
   transition, TNFα signaling via NFκB, angiogenesis, apoptosis and IFN-ɣ
   response (Supplemental Fig. S8C). To determine disease-specific genes
   and pathways in CNS-TB, we identify the differential expressed genes
   (DEGs) (false discovery rate (FDR) < 0.05) in human CNS-TB granulomas
   or CNS-TB inflammatory tissues compared to control adjacent normal
   tissues and human demyelination tissues. Processes enriched in human
   CNS-TB granulomas include ECM organization, vasculature development,
   angiogenesis, coagulation and complement (Fig. [140]3D, Supplemental
   Fig. S8D, E and Supplemental File S3). Compared to demyelination
   tissues, CNS-TB granulomas show up-regulated cytokine responses,
   cell-substrate adhesion, ECM organization, inflammatory, immune
   responses, angiogenesis and TNFα signaling via NFκB (Supplemental Fig.
   S8F). Both CNS-TB granulomas and inflammatory tissues show
   up-regulation in angiogenesis and inflammatory responses compared to
   demyelination tissues (Supplemental Fig. S8G), but do not show the same
   enriched pathways relative to adjacent normal tissues (Supplemental
   Fig. S8H), indicating that there are indeed processes uniquely enriched
   in CNS-TB granulomas. Comparing human CNS-TB granulomas to adjacent
   normal tissues, CNS-related processes are reduced including synaptic
   signaling, glutamate receptor signaling and chemical synaptic
   transmission (Fig. [141]3D and Supplemental Fig. S8I).

Fig. 3.

   [142]Fig. 3
   [143]Open in a new tab

   Spatial transcriptomic profiling of human and murine CNS-TB shows
   enrichment of inflammatory and immune genes and pathways. A Target
   regions of formalin-fixed paraffin-embedded (FFPE) CNS tissue samples
   from human CNS-TB (n = 6), control human demyelination tissues (n = 4),
   murine CNS-TB (n = 4) and healthy murine controls (n = 4) were
   microdissected for LCM RNA-Seq. Illustration created using Biorender.
   B PCA plot of gene expression libraries demonstrates clear separation
   of human CNS-TB granulomas and CNS-TB inflammatory from demyelination
   and adjacent normal tissues. C Hierarchical clustering of top 500
   variable genes expressed in human CNS-TB. Red frame highlights gene
   cluster that is distinctly expressed in human CNS-TB granulomas
   compared to other tissues. D Volcano plot of up-regulated (log
   fold-change > 2, FDR < 0.05) and down-regulated (log fold-change < −2,
   FDR < 0.05) genes in human CNS-TB granulomas compared to adjacent
   normal tissues. E PCA plot of gene expression libraries showing
   distinct clustering of murine CNS-TB granulomas and inflammatory
   tissues from adjacent normal tissues and normal tissue from healthy
   murine controls. F Hierarchical clustering of top 500 variable genes
   expressed in murine samples. Red and blue frames highlight gene cluster
   that are distinctly up-regulated and down-regulated, respectively, in
   murine CNS-TB granulomas relative to other tissue types. G Volcano plot
   of up-regulated (log fold-change > 2, FDR < 0.05) and down-regulated
   (log fold-change < 2, FDR < 0.05) genes in murine CNS-TB granulomas
   compared to normal tissues from healthy murine controls

   We interrogated the gene expression of murine CNS-TB in a similar
   fashion. PCA and hierarchical clustering of the top 500 variable genes
   expressed across murine samples show distinct gene expression clusters
   separating murine CNS-TB granulomas and inflammatory tissues from
   adjacent normal tissues and normal tissue from healthy mice
   (Fig. [144]3E - F and Supplemental File S4). There is consistent
   downregulation of genes associated with neuron differentiation, neuron
   development and synapse organization (Supplemental Fig. S9B), and
   enrichment of genes related to IFN-ɣ response, leukocyte chemotaxis and
   inflammatory response in CNS-TB granulomas and inflammatory tissues
   (Supplemental Fig. S9C). The up-regulated DEGs in murine CNS-TB
   granulomas compared to normal tissue from healthy mice involve
   processes in defense, immune, inflammatory and cytokine response,
   neutrophil degranulation, antigen presentation and Toll-Like Receptor
   (TLR) cascades (Fig. [145]3G, Supplemental File S5 and Supplemental
   Fig. S9D, E). Conversely, genes involved in trans-synaptic signaling,
   synapse and cell junction organization, neuron development and
   differentiation are decreased in murine CNS-TB granuloma tissues
   (Supplemental Fig. S9F). DEG analysis shows the gene expression in
   CNS-TB granulomas or inflammatory tissues are similarly associated with
   inflammatory, immune response, neutrophil degranulation relative to
   normal tissues from both healthy mice or adjacent normal tissues
   (Supplemental Fig. S9G- I).

   We observe five DEGs that are consistently up-regulated in human and
   murine CNS-TB granulomas compared to their respective controls
   (Supplemental Fig. 8J and Supplemental Fig. S9J). The genes are GBP5,
   which encodes guanylate binding protein 5, the activator of NLRP3
   inflammasome [[146]25]; IL7R, which encodes interleukin 7 receptor
   which regulates development and homeostasis of immune cells, including
   T cells, B cells and NK cells [[147]26]; CXCL9, which encodes
   IFN-γ-induced chemokine which recruits T cells and NK cells [[148]27];
   BIRC3, which encodes Baculoviral IAP Repeat Containing 3 protein which
   suppresses TNF-α stimulated cell apoptosis signaling [[149]28]; COL1A1,
   which encodes pro-alpha1 chains of type I collagen [[150]29]. Human
   MMP-2, −1, and murine Mmp-9, −3, −8 and − 12 are also up-regulated in
   CNS-TB granulomas, corresponding to the increased protein expression in
   our study (Supplemental Fig. 8K and Supplemental Fig. S9K).

Neutrophil genes and associated transcriptional factors are predominantly
present in human and murine CNS-TB granulomas

   We sought to investigate the immune cell types in both human and murine
   TB granulomas. CIBERSORTx analysis was performed to infer the fractions
   of immune cell types in each sample [[151]30]. Compared to human
   demyelination tissues, we observe significantly higher fractions of
   neutrophils (P = 0.0316), macrophages M1 (P = 0.0181), activated mast
   cells (P = 0.0027) and activated CD4 memory T cells (P = 0.0002) in
   human CNS-TB granulomas (Fig. [152]4A and Supplemental Fig. S10A). The
   major immune cell composition in the murine CNS-TB granulomas comprises
   of neutrophils (P < 0.0001), dendritic cells (P = 0.0013) and NK cells
   (P < 0.0001), with higher percentage compared to normal tissues from
   healthy murine controls (Fig. [153]4B). IHC was performed to validate
   the presence of the key immune cells populations, including
   neutrophils, macrophages and CD4^+ T cells in human CNS-TB granulomas,
   and neutrophils and NK cells in murine CNS-TB granulomas (Supplemental
   Fig. S11A-B). We further examined the presence of neutrophil
   transcripts in human and murine CNS-TB granulomas based on published
   neutrophil gene markers [[154]31–[155]34]. Neutrophil genes are
   significantly up-regulated in human CNS-TB granulomas (P = 0.0212) and
   murine CNS-TB granulomas (P = 0.0015) (Fig. [156]4C, D and Supplemental
   File S6). Marker genes associated with nearly mature neutrophils
   (Lgals3, Mcemp1, Mmp8, Mmp9, Retnlg) and mature circulating neutrophils
   (Btg1, Ccl6, Csf3r, Dusp1, Fth1, Fxyd5, H2-D1, Il1b, Msrb1, Srgn,
   Wfdc17) are significantly up-regulated in murine CNS-TB granulomas
   (FDR < 0.05) (Fig. [157]4D and Supplemental File S6). These gene
   signatures indicate the presence of neutrophils in granulomas and may
   play key roles in CNS-TB immunopathology. We further analyzed the gene
   sets associated with neutrophil functions. Both human and murine CNS-TB
   exhibited upregulated gene sets related to neutrophil chemotaxis,
   indicating neutrophil infiltration into the CNS-TB granulomas
   (Supplemental Fig. S10B-C). We also observe no significant difference
   in neutrophil differentiation genes across both datasets, suggesting
   that neutrophils may have differentiated before entering the
   granulomas. Additionally, gene sets related to neutrophil activation
   and NETosis are found to be elevated in the murine CNS-TB granulomas
   compared to normal tissues from healthy mice.

Fig. 4.

   [158]Fig. 4
   [159]Open in a new tab

   Key immune cells, neutrophil marker genes and transcription factors
   driving immunopathology in human and murine CNS-TB granulomas.
   A Deconvolution analysis by CIBERSORTx LM22 signature reveals that
   relative to human demyelination tissue, the in silico cell composition
   in human TB granulomas has higher fractions of neutrophils, macrophages
   M1, activated mast cells and activated CD4 memory T cells. Bars
   represent mean ± SD, and analysis was conducted using unpaired t test
   with Welch’s correction. * P < 0.05; ** P < 0.01; *** P < 0.001; ****
   P < 0.0001. B In silico deconvolution analysis using ImmuCC signature
   matrix indicates that relative to healthy murine brain tissues, murine
   CNS-TB granulomas consist mainly of neutrophils, dendritic cells and NK
   cells. Bars represent mean ± SD, and analysis was conducted using
   unpaired t-test with Welch’s correction. C-D Heatmap shows higher
   expression of neutrophil genes in (C) human CNS-TB granuloma samples
   compared to demyelination samples and (D) murine CNS-TB granuloma
   samples compared to control normal samples. Genes in bold font are
   significantly upregulated with FDR < 0.05. Enrichment of neutrophil
   gene sets is analyzed by gene set variation analysis (GSVA). Box
   represents 25th and 75th percentile, line represents median and
   whiskers denote the extremes. Analysis using Mann-Whitney test. *
   P < 0.05; ** P < 0.01. E-F Key transcriptional factors, which regulate
   neutrophil functions, correlate significantly with expression of their
   neutrophil-associated target genes in human and murine CNS-TB
   granulomas. Analysis by Spearman’s correlation coefficient

   Given the enrichment of neutrophils and cytokine responses in human and
   murine CNS-TB, we evaluated the expression of key transcription factors
   (TFs) known to be associated with neutrophil functions. In human CNS-TB
   granulomas, RUNX1 and KLF6, which modulate neutrophil maturation and
   migration [[160]35], correlate significantly with neutrophil-associated
   target genes including CCND2 (r^s = 0.4923, P = 0.0323), MYL9 (r^s =
   0.5119, P = 0.0251) and DAPK2 (r^s = 0.7739, P = 0.0001)
   (Fig. [161]4E). Similarly, in murine CNS-TB granulomas, Runx1 and Klf6
   associate significantly with their respective targets Cxcr2 (r^s =
   0.4871, P = 0.0404) and Cd9 (r^s = 0.515, P = 0.0287) (Fig. [162]4F).
   In addition, Cebpb, which regulates emergency granulopoiesis and
   enhances de novo neutrophils synthesis during inflammation [[163]36],
   significantly correlates with neutrophil-associated genes Cdkn1a (r^s =
   0.5823, P = 0.0112), Il1b (r^s = 0.6677, P = 0.0025), Ptgs2 (r^s =
   0.7155, P = 0.008) and Csf1r (r^s = 0.6285, P = 0.0052) in murine
   CNS-TB granulomas (Fig. [164]4F). These transcriptional activities
   implicate neutrophil-driven inflammation in human and murine CNS-TB.
   Together, our spatial transcriptomic analyses show similarities between
   the human and murine CNS-TB transcriptome.

Adjunctive MMP inhibitors SB-3CT and doxycycline with standard ATT therapy
significantly reduce brain immunopathology and improve murine CNS-TB survival

   As both our human cohort and our CNS-TB murine model consistently
   demonstrate elevated MMP-2 and − 9 expression, we next evaluated if
   adjunctive MMP inhibitor therapy with SB-3CT or broad-spectrum MMP
   inhibitor doxycycline and standard ATT therapy can suppress
   immunopathology in our CNS-TB murine model. Steroids were not added to
   determine if MMP inhibition itself suppresses inflammatory responses.
   The ATT treatment regime was initiated 3 weeks post-M. tb infection for
   a period of 8 weeks (Fig. [165]5A) when brain granulomas and tissue
   necrosis were observed in M. tb-infected mice. This occurred when the
   mice showed loss of balance and limb weakness. Hydrocephalus and
   infarcts, which are usually observed in humans in late stage CNS-TB,
   were not observed during murine brain harvesting. Limb weakness could
   be due to spinal cord involvement, which was not examined in the mice.
   There was no mortality during the first 3-weeks of pre-treatment
   incubation. Compared to ATT treatment alone, both ATT with SB-3CT
   (P = 0.0135) and ATT with doxycycline (P = 0.0328) significantly
   improve murine survival (Fig. [166]5B). The high mortality rate in our
   CNS-TB murine model receiving ATT may be attributed to the delayed
   initiation of treatment but we acknowledge no existing animal CNS-TB
   studies have shown that earlier TB treatment improves outcomes. Early
   treatment compared to delayed treatment in CNS-TB improves outcomes in
   humans [[167]37]. Brain histology shows vehicle control (infected) mice
   and ATT alone-treated mice develop pyogranulomas with liquefactive
   necrosis in the brain, while groups that received ATT with adjunctive
   MMP inhibitors have non-necrotizing granulomas (Fig. [168]5C).
   Furthermore, mice that received adjunctive SB-3CT or doxycycline
   demonstrate an overall reduced histopathological lesion burden
   (Fig. [169]5D and Supplemental Table S4). In groups that received ATT
   with MMP inhibitors, there is a significant reduction of gliosis
   (ATT + SB-3CT, P = 0.0038; ATT + Doxycycline, P = 0.0079),
   pyogranulomas (ATT + SB-3CT, P = 0.0026; ATT + Doxycycline,
   P = 0.0134), and bacilli (ATT + SB-3CT, P = 0.0102; ATT + Doxycycline,
   P = 0.0102) (Fig. [170]5D). CFU enumeration of brain homogenates
   relative to ATT only group demonstrates suppressed M. tb growth in
   ATT + SB-3CT and ATT + Doxycycline groups but do not reach statistical
   significance (Supplemental Fig. S12). Intracerebroventricular infection
   of mice introduced M. tb into CSF and subsequently spread through
   meningeal space, hence granulomas were found not only at the infection
   site, but also in meningeal and parenchymal tissues. Compared to
   ATT-only group, the total number of granulomas in ATT + SB-3CT
   (P = 0.0005) and ATT + Doxycycline groups (P = 0.0473) are
   significantly decreased (Fig. [171]5E). Ziehl-Neelsen (ZN) staining
   confirms the presence of acid-fast bacilli (AFB) in the murine CNS-TB
   granulomas (Fig. [172]5F). Notably, a smaller number of AFB positive
   granulomas is observed in the group that received ATT + SB-3CT relative
   to VC (P = 0.0036) or ATT alone (P = 0.0155), and a decreasing trend is
   observed in the ATT + Doxycycline group (Fig. [173]5G).

Fig. 5.

   [174]Fig. 5
   [175]Open in a new tab

   Adjunctive MMP inhibition with SB-3CT or doxycycline with standard ATT
   significantly improves murine CNS-TB survival and reduces
   immunopathology with decreased expression of NETs and inflammatory
   mediators. A Six- to eight-week-old male C57BL/6 Nos2^−/− mice were
   infected with 10^5 CFU M. tb H37Rv via the i.c.vent. route and treated
   with ATT, ATT + SB-3CT or ATT + Doxycycline from 3 weeks
   post-infection. (n = 7 to 8 per group). After eight weeks of treatment,
   the brain homogenates were collected for downstream analyses.
   Illustration created using Biorender. B Kaplan–Meier curve shows a
   significant improvement in survival time of M. tb-infected mice treated
   with MMP inhibitor SB-3CT or doxycycline plus ATT. Log rank test was
   performed. C Representative hematoxylin–eosin stains of M. tb-infected
   mouse brain pyogranulomas in VC (infected) and ATT alone groups (top
   panel), and non-necrotizing granulomas in ATT + SB-3CT and
   ATT + Doxycycline groups (bottom panel). Scale bar = 100 μm.
   D Histopathological severity scoring of brain lesions is higher in
   vehicle control (M. tb-infected) and ATT-alone mice with lower severity
   scores in adjunctive MMP inhibitor-treated mice. Bars represent
   median ± IQR. Analysis was conducted using Kruskal-Wallis test with
   Dunn’s multiple comparisons test. E Granuloma numbers are less in M.
   tb-infected mice treated with ATT with adjunctive MMP inhibitors. Bars
   represent median ± IQR, analysis was conducted using Kruskal-Wallis
   test with Dunn’s multiple comparisons test. F Ziehl–Neelsen staining
   shows the presence of AFB in the granuloma of murine CNS-TB of
   different groups. Scale bar = 10 μm. G Smaller number of AFB-containing
   granulomas are found in mice on ATT and adjunctive MMP inhibitor SB-3CT
   or doxycycline. Bars represent median ± IQR, analysis was conducted
   using Kruskal-Wallis test with Dunn’s multiple comparisons test.
   H CitH3 expression is significantly up-regulated in brain homogenates
   of mice with liquefactive necrosis. Bars represent median ± IQR.
   Analysis by Mann-Whitney test. I CitH3 expression is suppressed in
   CNS-TB mice treated with adjunctive MMP inhibitors SB-3CT or
   doxycycline. Bars represent median ± IQR. Analysis by Mann-Whitney
   test. J Decreased CitH3 in the brain homogenates of murine CNS-TB is
   strongly associated with longer survival days. Analysis was conducted
   using Spearman correlation coefficient. K Brain homogenate MMP-9 is
   strongly associated with CitH3, analyzed by Spearman correlation
   coefficient. L Heatmap shows suppressed median expression of
   inflammatory mediators in adjunctive MMP inhibitor groups.
   M Inflammatory mediators TNF-α, IFN-ɣ, CXCL2, CXCL10, CCL2, CCL3, CCL7,
   CCL11, and CCL19 are inversely correlated with longer survival days
   (all strong associations with r^s < −0.5), analyzed by Spearman
   correlation coefficient. * P < 0.05; ** P < 0.01

   We sought to evaluate the role of NETs in our murine model and found
   that mice with liquefactive necrosis of brain granulomas have
   significantly up-regulated CitH3 (P = 0.0268) (Fig. [176]5H). After
   treatment with MMP inhibitors, CitH3 concentration is significantly
   decreased in mice treated with ATT + SB-3CT (P = 0.0379) and
   ATT + Doxycycline (P = 0.0207) (Fig. [177]5I), indicating that MMP
   inhibition reduces NETs formation. The reduced CitH3 significantly
   associates with longer survival (r^s = −0.695, P < 0.0001)
   (Fig. [178]5J). We then investigated if neutrophil mediators,
   specifically MMP-9 and NETs, correlate to tissue destruction in CNS-TB,
   and found a significant association between MMP-9 and CitH3
   concentrations (r^s = 0.704, P < 0.0001) (Fig. [179]5K).

   Next, we analyzed the brain cytokines and chemokines that govern the
   recruitment of diverse immune cells that drive inflammation. TNF-α, a
   cytokine that is key to CNS-TB pathogenesis by increasing BBB
   permeability [[180]38], and other proinflammatory cytokines IFN-ɣ,
   IL-1α, IL-1β, IL-6, IL-17 A, IL-27 are downregulated in groups that
   received ATT with adjunctive SB-3CT or doxycycline (Fig. [181]5L).
   Compared to vehicle control and ATT groups, neutrophil chemoattractant
   CXCL1 and CXCL2, monocytes and macrophages chemoattractant CCL2, CCL3,
   CCL7, CCL12, eosinophil chemoattractant CCL11, and Th1-associated
   chemokines CXCL10, CCL4, CCL5 are also down-regulated in ATT + SB-3CT
   and ATT + Doxycycline groups (Fig. [182]5L). We further demonstrate
   that the following suppressed cytokines and chemokines are strongly
   associated with longer survival; TNF-α (r^s = −0.620, P = 0.0002),
   IFN-ɣ (r^s = −0.660, P < 0.0001), CXCL2 (r^s = −0.668, P < 0.0001),
   CXCL10 ((r^s = −0.779, P < 0.0001), CCL2 (r^s = −0.679, P < 0.0001),
   CCL3 (r^s = −0.711, P < 0.0001), CCL7 (r^s = −0.812, P < 0.0001), CCL11
   (r^s = −0.604, P = 0.0003) and CCL19 (r^s = −0.680, P < 0.0001).
   Specifically, mice that received adjunctive MMP inhibitors form a
   cluster with lower concentrations of these inflammatory mediators
   relative to those without (Fig. [183]5M). These observations indicate
   that MMP inhibition suppresses pro-inflammatory cytokines and
   chemokines which strongly associate with improved survival outcomes.

Adjunctive MMP inhibition suppresses the matrix-degrading phenotype and
normalizes blood vasculature in murine CNS-TB

   To determine if adjunctive MMP inhibition suppresses the
   matrix-degrading phenotype in CNS-TB, we analyzed the brain MMPs and
   TIMPs of the different murine treatment groups. Relative to vehicle
   control and ATT-treated mice, mice that received adjunctive MMP
   inhibitors and ATT have decreased gelatinases MMP-2 and − 9,
   collagenase MMP-8, stromelysin MMP-3, −10, and elastase MMP-12
   (Fig. [184]6A). The suppressed brain MMP-3 (r^s = −0.793, P < 0.0001),
   MMP-8 (r^s = −0.662, P < 0.0001), MMP-9 (r^s = −0.699, P < 0.0001),
   MMP-10 (r^s = −0.594, P = 0.0004) and MMP-12 (r^s = −0.695, P < 0.0001)
   are associated with longer survival (Fig. [185]6B). Additionally,
   higher MMP-3 (r^s = 0.457, P = 0.01), MMP-9 (r^s = 0.438, P = 0.014)
   and MMP-12 (r^s = 0.386, P = 0.032) are significantly associated with
   worse histopathological scores, whereas treatment groups that received
   adjunctive MMP inhibitors have lower histopathological scores
   (Fig. [186]6C). We find decreased TIMP-1 concentration and increased
   TIMP-2 and − 4 concentrations in MMP inhibitors groups compared to
   vehicle control (Supplemental Fig. S13A). Next, we analyzed the
   functional gelatinase activity which would better reflect the overall
   proteolytic activity in the brain. Gelatinase activity in the ATT
   (P = 0.0026), ATT + SB-3CT (P = 0.0005) and ATT + Doxycycline groups
   (P = 0.0009) are significantly suppressed (Fig. [187]6D), and
   gelatinase activity is strongly associated with MMP-9 concentration
   (r^s = 0.902, P < 0.0001) (Fig. [188]6E). As expected, the decreased
   gelatinase activity indicating a more intact blood-brain-barrier is
   strongly associated with longer survival (r = −0.650, P < 0.0001)
   (Fig. [189]6F).

Fig. 6.

   [190]Fig. 6
   [191]Open in a new tab

   Adjunctive doxycycline with ATT suppresses functional gelatinase
   activity and normalizes blood vasculature in murine CNS-TB granulomas.
   A Heatmap showing median expression of brain MMP-2, −9, −8, −3, −10,
   and − 12 downregulated in the MMP inhibitor groups of murine CNS-TB.
   MMP concentrations are normalized to total protein concentration.
   B Suppressed murine brain homogenate MMP-3, −8, −9, 10, and − 12 are
   strongly associated with longer survival, analyzed by Spearman
   correlation coefficient. C Up-regulated MMP-3, −9, and − 12 are
   significantly associated with worse histopathological scores, analyzed
   by Spearman correlation coefficient. D Functional gelatinase activity
   is suppressed in the brain homogenates of murine CNS-TB which received
   ATT, ATT + SB-3CT or ATT + Doxycycline. Bars represent mean ± SD,
   analysis was conducted using one-way analysis of variance (ANOVA) with
   Dunn’s multiple comparisons test. E Brain homogenate MMP-9 of murine
   CNS-TB is strongly associated with functional gelatinase activity,
   analyzed by Spearman correlation coefficient. F The functional
   gelatinase activity of murine brain homogenates is strongly associated
   with longer survival, analyzed by Pearson correlation coefficient. G In
   silico deconvolution analysis using ImmuCC signature matrix was
   conducted to infer immune cell fractions in murine CNS-TB of different
   treatment groups. Bars represent mean ± SD, and analysis was conducted
   using two-way ANOVA with Dunn’s multiple comparisons test. H Enrichment
   of Gene sets using the transcriptome data from murine CNS-TB treated
   with different regimes was analyzed by GSVA. Box represents 25th and
   75th percentile, line represents median and whiskers denote the
   extremes. Analysis using unpaired t-tests. * P < 0.05; ** P < 0.01, ***
   P < 0.001. I Blood vessels were double-stained for endothelial cells
   with CD31and alpha-smooth muscle actin (α-SMA) for pericytes in the
   human CNS-TB brain. Inset shows the granuloma rim and center. Vessels
   with wide lumen are distributed at the rim and mostly poorly-formed
   microvessels are found at the center. Gray outline indicates granuloma
   rim. Orange arrow points to the blood vessel. Scale bar = 200 μm.
   J Compressed vessels and poorly formed microvessels are found in the
   granuloma periphery of M. tb-infected mice treated with VC or ATT
   alone. Conversely, vessels with open lumen are present in the granuloma
   center of M. tb-infected mice received adjunctive SB-3CT or
   doxycycline. Yellow “c” indicates granuloma center. Scale bar = 200 μm.
   K CNS-TB mice treated with ATT + Doxycycline have significantly
   increased total area of blood vessels with pericyte coverage compared
   to mice that received vehicle control or ATT alone. In each group, four
   to six tissue sections with the presence of granulomas are assessed in
   a blinded fashion. Bars represent median ± IQR, analysis was conducted
   using Kruskal-Wallis test with Dunn’s multiple comparisons test. L VC
   and ATT alone groups have more vessels located around the periphery,
   whereas the MMP inhibitor groups have more vessels distributed towards
   the center of the granuloma. Each point represents the distance of the
   blood vessel from the granuloma rim. The distance towards the granuloma
   center is calculated as a negative value. In each group, four to six
   tissue sections with seven to thirty-eight blood vessels within the
   maximum range of 300 μm away from the granuloma rim are quantified.
   Bars represent median ± IQR, analysis was conducted using
   Kruskal-Wallis test with Dunn’s multiple comparisons test. M Top panel
   shows representative matrix-assisted laser desorption ionization mass
   spectrometry imaging (MALDI-MSI) images of murine CNS-TB brain tissues.
   Rifampicin is evenly distributed in the ATT alone group, in contrast to
   ATT + SB-3CT and ATT + Doxycycline groups which show higher
   concentrations within the granuloma. Bottom panel shows respective ZN
   stains. Red circled area indicates granulomas region with AFB stained.
   Scale bar = 200 μm

   We performed LCM RNA-Seq to profile the transcriptomic changes in the
   brain tissues of murine CNS-TB treated with ATT, ATT + SB-3CT and
   ATT + Doxycycline. Low number of DEGs (FDR < 0.05) are identified
   (Supplemental Fig. S13B). The gene expressions in murine CNS-TB that
   received different treatment regimes are heterogeneous (Supplemental
   Fig. S13C and Supplemental File S7). Based on the immune cells
   fractions inference from CIBERSORTx, the neutrophil fraction is
   significantly suppressed in the two groups of murine CNS-TB treated
   with adjunctive MMP inhibitors (ATT + SB-3CT, P = 0.0003;
   ATT + Doxycycline, P = 0.0005), while basophils are upregulated in
   murine CNS-TB on ATT with doxycycline (P = 0.0358) (Fig. [192]6G). Gene
   Set Variation Analysis (GSVA) analysis indicates that gene sets
   associated with tight junction organization (P = 0.0138), blood vessel
   morphogenesis (P = 0.0032), vascular development (P = 0.0436),
   complement (P = 0.022), VEGF (P = 0.0007), Notch (P = 0.0008) and KRAS
   (P = 0.0067) signaling are increased in murine CNS-TB on ATT with
   SB-3CT. Compared to ATT alone, ATT + Doxycycline shows increased
   expression in gene sets related to angiogenesis in wound healing
   (P = 0.0042) and hypoxia (P = 0.0139) (Fig. [193]6H and Supplemental
   File S8).

   As we found MMP inhibition reduced AFBs in granulomas, reduced tissue
   destruction and enhanced gene sets associated with blood vasculature
   development, we sought to evaluate if the vasculature and drug
   distribution are also improved. In human CNS-TB brain tissue,
   well-formed vessels (CD31^+αSMA^+) with wide lumen area are found at
   the granuloma rim, whereas poorly-formed microvessels are located at
   the granuloma center (Fig. [194]6I). Similarly, murine CNS-TB in the
   vehicle control or ATT alone groups demonstrate blood vessels with
   compressed lumen along the granuloma periphery (Fig. [195]6J). In
   contrast, murine CNS-TB in the SB-3CT or doxycycline groups show blood
   vessels with pericytes (CD31^+αSMA^+) in both granuloma center and rim,
   with larger luminal area in ATT + Doxycycline group (Fig. [196]6J).
   Quantitatively, the ATT + Doxycycline group has significantly more
   well-formed pericyte-covered blood vessels compared to vehicle control
   (P = 0.0051) and ATT alone group (P = 0.0153) (Fig. [197]6K). Compared
   to ATT group, the well-formed pericyte-covered vessels are located
   significantly nearer to the granuloma center in ATT + SB-3CT
   (P = 0.0094) and ATT + Doxycycline (P = 0.0033) groups (Fig. [198]6L).

   We also measured the drug distribution and concentration in the murine
   CNS-TB brains via MALDI-MSI and LC-MS/MS respectively. Comparable
   concentrations of ATT drugs Rifampicin, Isoniazid, Ethambutol and
   Pyrazinamide are detected in the brain homogenates for all the treated
   murine CNS-TB, indicating successful BBB penetration of ATT drugs in
   all the TB treatment regimes (Supplemental Fig. S13D). Published
   literature has similarly showed that Rifampicin, Isoniazid, Ethambutol
   and Pyrazinamide were able to cross the blood-brain-barrier and not
   detected in the human brain homogenates but CSF [[199]39–[200]41].
   Doxycycline is also detected in the brain homogenates of mice that
   received ATT + Doxycycline (Supplemental Fig. S13E, F). Although we are
   unable to detect SB-3CT in the brain homogenates of M. tb-infected mice
   received ATT + SB-3CT, previous studies have proven the ability of
   SB-3CT to cross the BBB in mice [[201]42, [202]43]. We further
   determined the distribution of rifampicin in the brain tissues using
   MALDI-MSI. Despite similar drug concentrations found in all treatment
   groups of murine CNS-TB (Supplemental Fig. S13D), rifampicin is
   homogenously distributed throughout the brain tissue in ATT-only group,
   indicating there was global capillary leak throughout the brain tissue,
   while with adjunctive MMP inhibition with more well-formed capillaries,
   we observe more localized drug delivery (Fig. [203]6M). This may also
   result in more efficient M. tb killing thus explaining the lower AFB
   granulomas in the MMP inhibitor groups. Collectively, MMP inhibition in
   murine CNS-TB shows suppressed MMPs which reduce tissue destruction and
   pathological lesions, thus decreasing both inflammatory neutrophil
   infiltration and NETs release. The less inflamed and non-necrotic
   tissues found in MMP inhibitor groups, with improved vasculature found
   in doxycycline-treated mice, enhance drug penetration into the
   granulomas to improve survival.

Discussion

   The current standard-of-care for human CNS-TB is ATT with steroids but
   this does not improve neurological outcomes for which new therapeutic
   strategies are urgently needed [[204]44]. Here, we describe MMPs and
   NETs in human and murine CNS-TB, demonstrating their pathological role
   in driving CNS-TB immunopathology. Crucially, we found significant
   association of MMPs and NETs with neurological abnormalities and poor
   clinical outcomes. Spatial transcriptomic profiles of human and murine
   CNS-TB indicate a highly-inflamed and neutrophil-rich microenvironment,
   together with suppressed neuronal functions. In our pre-clinical drug
   trial, we demonstrate novel findings of ATT with doxycycline to improve
   murine CNS-TB survival and reduce immunopathology. We also uncover the
   mechanism of action of doxycycline and another MMP-inhibitor SB3CT,
   where blood vessel structure to the granulomatous regions of the brain
   is improved, potentially improving ATT drug delivery to tuberculous
   granulomas. Most importantly, our results are highly translatable for
   clinical use, as doxycycline is widely available and cheap which are
   key in TB endemic regions where resources are limited.

   CSF and brain tissues comprise of distinct immune cell composition. The
   CSF of TBM patients showed a predominance of αβT cells, highly variable
   proportions of NK cells and neutrophils [[205]45], in contrast to the
   human CNS-TB brain tissues which demonstrated abundant presence of
   macrophages, microglia, neutrophils and T-cells [[206]46]. Considering
   this may result from different biological processes from different body
   compartments, we validated the outcome comprehensively through CSF
   immunological profiling, brain radiology and CNS biopsy transcriptomic
   analysis. In the human cohort, we found TBM patients to have
   significantly up-regulated CSF gelatinase MMP-2 and − 9 and
   down-regulated TIMP-2 and − 4, driving an ECM-degrading phenotype.
   Several human studies have previously demonstrated an upregulation of
   MMP-9 in TBM patients [[207]2, [208]13–[209]17], which reflected the
   ability of M. tb to induce the secretion of human host enzymes to break
   down the basement membrane of the BBB [[210]47]. The finding of
   increased CSF MMP-2 expression was similar in another pediatric TBM
   cohort [[211]48], but in contrast to previous findings in adult TBM
   [[212]2]. We also found significantly higher expressions of MMP-1, −3,
   −7 and − 10 in TBM patients who presented with neuroradiological
   abnormality and poor clinical outcomes, demonstrating that multiple
   MMPs contribute to the pathogenesis in CNS-TB. MMPs were reported to
   have pathological roles in neurological illnesses such as stroke,
   multiple sclerosis and vascular cognitive impairment [[213]49].
   Furthermore, we describe novel findings of significantly increased CSF
   NETs in TBM patients which were associated with poorer neurological
   outcomes and death. Other non-TB studies have reported that NETs
   exacerbate neurological deficits in traumatic brain injury [[214]50]
   and poorer clinical outcomes in stroke [[215]51]. NETs inhibitor may be
   explored as a potential adjunctive therapy to improve CNS-TB outcomes
   in future studies.

   The critical formation of necrotizing granulomas is key in
   understanding TB immunopathology. Several TBM murine models, including
   intracerebral M. tb H37Rv infection of C57BL/6 mice [[216]52],
   intracranial M. tb H37Rv infection of Swiss albino mice [[217]53], and
   intravenous M. tb C3 infection of BALB/c mice [[218]54], did not
   develop the hallmark necrotizing TB granulomas. We previously
   demonstrated that intracerebroventricular M. tb H37Rv infection of
   C57BL/6 Nos2-/- resulted in neutrophilic necrotic granulomas, meningeal
   inflammation, neurological signs of seizures and limb weakness and
   severe pathology recapitulating human CNS-TB [[219]12].
   Intraventricular M. tb H37Rv infection of Kramnik mice also formed
   necrotizing parenchymal lesions in the brain tissue and did not mirror
   the severity of human disease where fatality is uniform without
   treatment [[220]55]. The potential for Nos2-deficient C3HeB/FeJ mice to
   exhibit a disease phenotype similar to that of C57BL/6 Nos2^⁻/⁻ mice,
   along with the investigation of cell-specific Nos2 depletion, warrants
   further exploration. Our murine model has inherent limitations in
   recapitulating clinical and radiological features of human CNS-TB given
   differences between species, though can be used to study specific
   pathological mechanisms common to both. In humans, the immunophenotype
   arise from complex immunological and physiological factors which are
   not fully recapitulated in the murine model. Comprehensive
   characterization using advanced neuroimaging and detailed
   histopathological studies is necessary to precisely define the
   translational relevance of our murine model to human CNS-TB.

   In this study, we now show that C57BL/6 Nos2^−/− exhibit upregulation
   of MMPs and shows a rich neutrophilic inflammation, similar to CNS-TB
   patients with poor prognosis which have CSF which is predominantly
   neutrophilic [[221]45, [222]56–[223]58]. On this scientific basis, the
   Nos2^−/− mice may serve as a representative model of human CNS-TB to
   investigate specific pathologic mechanisms, which disproportionately
   afflicts immunosuppressed individuals such as patients living with HIV
   and patients on systemic immunosuppressants [[224]59–[225]62]. In
   keeping with our findings in human CNS-TB, our pre-clinical CNS-TB
   murine model show high expression of NETs and MMP-9 that colocalized at
   necrotic regions, implicating the role of NETs and MMP-9 in brain
   tissue destruction in this devastating infection.

   Existing transcriptomic data on CNS-TB are limited to whole blood and
   CSF samples from TBM pediatric patients and brain tissues from M.
   bovis-infected mice [[226]63, [227]64]. In this study, we present novel
   spatial transcriptomic analysis of the granulomas from human and murine
   CNS-TB. Our main findings reveal a significant enrichment of pathways
   associated with inflammation, immune responses, ECM organization and
   angiogenesis, with elevated neutrophil-related genes and TFs. Gene
   signatures from human TB lung tissues and whole blood have similarly
   shown that inflammation in TB is associated with interferon signaling
   pathways [[228]65, [229]66]. We now extend findings of dominant IFN-ɣ
   and Type I IFN signaling which are neutrophil-driven in TB lungs
   [[230]67], to human and murine CNS-TB granulomas. Neutrophil
   qualitative states may be crucial determinants of disease [[231]68],
   thus we suggest further characterizing which neutrophil subsets are
   associated with IFN transcriptional program in CNS-TB using single-cell
   RNA sequencing. Neutrophils release NETs, which is associated with
   necrosis and caseation, upon encountering Type I IFN from macrophages
   in their migration into the granuloma [[232]69]. Therefore, studying
   neutrophil interactions with other immune cells in CNS-TB is crucial.
   We also demonstrate that ECM remodeling of collagen and ECM degradation
   which is pathognomonic to CNS-TB are similar to the findings from TB
   lung and lymph node transcriptomic analyses [[233]66, [234]70].
   Down-regulated neuronal pathways in human and murine CNS-TB brain
   granulomas indicate that M. tb infection may interfere with CNS
   neuronal function. The five consistently up-regulated DEGs in both
   human and murine CNS-TB granulomas point to prolonged immune responses,
   with GBP5 implicating inflammasome activation, IL7R and CXCL9
   implicated in the regulation and recruitment of immune cells, and BIRC
   implicated in the inhibition of apoptosis. Additionally, COLA1
   indicates ongoing fibrosis and tissue remodeling for which upregulation
   of MMP genes are also implicated. The transcriptomic findings uncover a
   potential therapeutic intervention using MMP inhibitors to modulate
   host tissue remodeling and inflammation in CNS-TB granulomas.

   In our CNS-TB mouse model [[235]12], we demonstrate adjunctive MMP
   inhibition reduced immunopathology and improved survival. SB-3CT is an
   MMP inhibitor that is selective for gelatinases MMP-2 and MMP-9
   [[236]71], and can cross the blood-brain-barrier [[237]42], but is not
   available for human use. Although another murine CNS-TB study has shown
   that adjunctive SB-3CT with ATT treatment decreased M. tb burden, the
   pathology was not comprehensively analyzed and the mortality benefit
   was not demonstrated [[238]53]. Doxycycline, a U.S. Food and Drug
   Administration (FDA)-approved broad-spectrum MMP inhibitor is an
   effective adjunctive antibiotic that suppressed MMPs and reduced
   pulmonary cavity volume in pulmonary TB patients from our own Phase 2
   trial [[239]72]. An in vitro study showed that SB-3CT did not
   significantly suppress growth of M. tb [[240]16]. Doxycycline exhibited
   bacteriostatic activity against drug-sensitive M. tb [[241]73].
   However, the in vitro bacteriostatic effect was not observed in vivo
   relative to the ATT alone arm. Using an appropriate control
   (ATT-treated), we demonstrated that both SB-3CT (ATT + SB-3CT) or
   doxycycline (ATT + doxycycline) had no significant antibacterial
   activity in vivo. Treating murine CNS-TB with ATT together with
   adjunctive SB-3CT or doxycycline, suppressed NETs, down-regulated
   MMP-3, −8, −9, −10 and − 12 in the brain, with a corresponding
   reduction in CNS immunopathology with a corresponding significant
   improved survival. By reducing MMPs, particularly MMP-9, which can
   cleave IL-8 to generate a more potent chemoattractant, there would be
   less cellular infiltration with reduced neutrophil recruitment
   [[242]74]. This feed-forward mechanism would further decrease cytokine
   and chemokine secretion, reduce inflammation and tissue damage.
   Adjunctive MMP inhibition in our study also altered the immune cell
   fractions in murine CNS-TB, with significantly decreased neutrophils,
   increased basophils and a trend of increased T cells. Of note, previous
   studies have highlighted the importance of T cells in mediating M. tb
   control [[243]75, [244]76]. In line with MMP inhibition which
   demonstrated increased pericyte-covered blood vessels and improved TB
   drug delivery and retention on murine pulmonary TB [[245]77], we now
   extend MMP inhibition to murine CNS-TB and showed increased well-formed
   pericyte-covered blood vessels indicating that doxycycline modulates
   angiogenesis and normalizes vasculature in CNS-TB granulomas.
   Suppressing ECM remodeling may be the key mechanism by which MMP
   inhibition improves blood vasculature. MMP inhibitors markedly reduced
   collagen lysis, associated with stabilized endothelial networks with
   accumulation of collagen fibrils surrounding the microvessels
   [[246]78]. As MMPs also detach pericytes from vessels during
   angiogenesis [[247]79], inhibiting MMPs increased pericyte coverage and
   attenuated vessel leakage [[248]77, [249]80]. The modulation of blood
   vessels may facilitate drug penetration and retention in the
   granulomas, while concurrently suppressing inflammatory responses to
   hasten the resolution of infection.

   Our study has several limitations. Given the rarity of CNS-TB, a small
   proportion of our TBM patients had microbiologically-confirmed
   diagnosis, with the remaining being possible TBM [[250]81].
   Nonetheless, all patients received anti-TB treatment. There was
   unavailability of information regarding the disease stage of TBM
   patients, which limits the ability to correlate this variable with CSF
   findings. With the limited volume of CSF, gelatin zymography could not
   be performed to assess the endogenous activity of MMP-2 and − 9. The
   human CNS-TB histological samples were archived for up to six years,
   which may explain the lower gene counts compared to murine histological
   samples that were processed within a year. The limited availability of
   human CNS-TB tissues also restricted its validation via RT-PCR. Despite
   the age difference between pediatric CSF and adult CNS-TB biopsy
   samples, both CSF protein and tissue transcriptomic data indicate
   inflammatory and ECM dysregulation in CNS-TB patients. The
   transcriptomic data from murine CNS-TB mice of different treatments did
   not show clear clusters in PCA, which may be due to different
   harvesting time points. Another limitation is the difference in age
   group between our murine CNS-TB model and human CSF samples. Adult mice
   aged 6 to 8 weeks were used in the CNS-TB model due to the significant
   technical challenges associated with cannula implantation into the
   brain of smaller younger mice and the associated increased risk of
   procedural mortality. Furthermore, younger mice would likely succumb
   earlier to M. tb infection before the infection could be established in
   the brain with visible granulomas, limiting their use as a CNS-TB model
   to investigate potential host-directed therapy. In contrast, human CSF
   samples were obtained from pediatric populations due to the higher
   prevalence of CNS-TB in children compared to adults. We did not include
   mouse models with other bacterial or viral CNS infections as controls
   to distinguish TB-specific findings. Consistent with previous studies,
   healthy mice were used to establish a clear baseline [[251]52, [252]55,
   [253]64]. However, future studies should explore the inclusion of other
   CNS infection models for improved comparison. The mice were not treated
   with steroids which is the standard of care for human CNS-TB; we
   avoided steroids to determine the true anti-inflammatory effects of MMP
   inhibition. This is crucial to determine the clinical translational
   value of MMP inhibition without the masking effect of dexamethasone and
   its effects on a disease with high mortality in our mouse model. Our
   current Phase 2 human CNS-TB clinical trial ([254]NCT06446245) now
   investigates whether adjunctive doxycycline plus steroids with ATT
   compared to standard of care improves patient outcomes, and so will
   address this question in part. Given the limitations of the BSL-3
   facility which permitted only animal monitoring during weekends, daily
   drug administration to mice was not feasible. This is also in keeping
   with other preclinical M. tb studies where drugs were administered five
   times per week [[255]82, [256]83]. We were also constrained by the
   biosafety requirements to process M. tb-infected tissue as FFPE and
   were unable to detect other TB drugs after the process of dewaxing the
   tissues. Positron emission tomography (PET) imaging may be an
   alternative approach for studying the drug levels and reportedly
   demonstrated excellent CNS penetration of pretomanid or rifampicin in
   mouse and rabbit TBM models [[257]55, [258]84]. Increasing the dose of
   rifampicin may enhance the drug distribution in the granulomas.
   High-dose rifampicin was shown to significantly improve bactericidal
   activity compared to standard-dose rifampicin [[259]55], and may be
   explored with adjunctive MMP inhibition in future studies. Infarcts and
   hydrocephalus, a late presentation in human CNS-TB, were not observed
   in our murine CNS-TB model. Due to lack of neuro-imaging facility in
   our BSL3, we were unable to detect basal leptomeningitis. Histological
   analysis in our previous study indicated meningeal inflammation near
   the basal brain regions [[260]12]. However, partial brain sectioning
   limited comprehensive evaluation and prevented definitive confirmation
   of basal leptomeningitis. Future studies should perform more focused
   sectioning of the entire brain, including ventral regions, to better
   characterize basal meningeal involvement.

   Altogether, we demonstrate the detrimental role of MMPs and NETs in
   driving CNS-TB immunopathology both in a human cohort and in a
   pre-clinical murine model. Critically, adjunctive CNS-TB therapy with
   MMP inhibitors improves survival and reduces immunopathology in murine
   CNS-TB. Our findings show that MMP inhibition has excellent potential
   to reduce the great toll of morbidity and mortality caused by CNS-TB,
   and doxycycline is readily available and cost-effective, as
   host-directed therapy in CNS-TB to improve both neurological and
   survival outcomes.

Materials and methods

Participants

   This was a prospective observational cohort study conducted at Hospital
   Likas, Kota Kinabalu, Sabah, East Malaysia. The study was approved by
   the Medical Research Sub-Committee of the Malaysian Ministry of Health
   (Reference NMRR-11-623-9629), the Menzies School of Health Research,
   Australia (Reference 2011 − 1636), and the Domain Specific Review
   Board, Singapore (Reference 2015/00067). Subjects have given prior
   consent for their samples to be used in future research stored in
   biobank applications NHG DSRB 2013/01259, CIRB 2013/374/E,
   NUH/2012 − 00116 for the conduct of this study. Enrollment occurred
   from year 2013 till 2014. A standardized data sheet was used by
   research clinicians to collect demographic data, epidemiological
   information (animal exposure, family history of TB, travel history),
   clinical history, vital signs, biochemical and microbiological data,
   imaging reports, treatment, outcome, and final diagnosis. Childhood
   vaccinations for potential CNS pathogens in Sabah, Malaysia included
   diphtheria-pertussis-tetanus, measles-mumps-rubella, polio, BCG, and
   Haemophilus influenza type B vaccine [[261]85].

   All pediatric patients with a clinical suspicion of meningitis or
   encephalitis were initially enrolled and subsequently excluded if they
   did not meet the case definitions for meningitis and encephalitis.
   Meningitis was defined as having two or more of the following: (i)
   fever or history of fever (≥ 38 °C), (ii) neck stiffness, (iii)
   headache, or (iv) cerebrospinal fluid (CSF) pleocytosis (≥ 5 WBC/µL).
   Encephalitis was defined as having encephalopathy (altered level of
   consciousness, lethargy, irritability, and change in behavior or
   personality) plus two or more of the following: (i) fever or history of
   fever (≥ 38 °C), (ii) CSF pleocytosis (≥ 5 WBC/µL), (iii) seizures or
   focal neurological findings, (iv) abnormal neuroimaging consistent with
   encephalitis, or (v) abnormal electroencephalogram findings compatible
   with encephalitis. This definition has been used previously and will
   allow us to compare our results to published studies [[262]85]. A
   published consensus case definition was used to classify TBM into
   confirmed, probable or possible TBM based on clinical, CSF,
   microbiological and radiological criteria [[263]81]. Clinically
   diagnosed and treated TBM includes probable (n = 9) and possible
   (n = 9) cases.

   Other exclusion criteria were (i) patients with non-infectious CNS
   disorders due to hypoxic, vascular, toxic and metabolic causes, (ii)
   patients with CNS disorders lasting less than 24 h, and (iii) patients
   with malaria diagnosed on microscopy. Patients with an immune-mediated
   post-infectious etiology such as acute disseminated encephalomyelitis
   (ADEM) were enrolled as well.

Clinical neurological scoring, radiological abnormalities and outcomes
assessment

   The modified Glasgow Coma Scale (GCS) was used to assess the conscious
   state of patients ≥ 3 years old, with scores ranging from 3 (indicating
   deep unconsciousness or coma) to 15 (fully awake). The scale is
   composed of 3 tests: eye response (score 1–4), verbal response (score
   1–5) and motor response (score 1–6). In younger children (patients < 3
   years old), the Blantyre Coma Scale (BCS) was used instead, with scores
   from 0 to 5 with 0 being equivalent to GCS 3 and 5 being equivalent to
   GCS 15. Similar to GCS, the BCS is also composed of 3 tests: eye
   response (score 0–1), verbal response (score 0–2) and motor response
   (score 0–2). CT brains of TBM patients were assessed for radiological
   abnormalities including leptomeningeal contrast enhancement and
   ventricular dilatation by two neuroradiologists (S.C.T and T.C.C.L)
   blinded to patient outcomes. TBM patients with full recovery or good
   recovery with mild deficit are categorized under good clinical outcome,
   whereas TBM patients who died or survived with severe disability were
   categorized under poor clinical outcome.

Mouse cannula implantation and infection

   Six-to-eight-week-old male C57BL/6 Nos2^−/− (Stock No. 002609) and
   C3HeB/FeJ mice (Stock No. 000658) (Jackson Laboratory) were infected
   with M. tb H37Rv or CDC1551 at seven days after brain cannulation, via
   intra-cerebroventricular (i.c.vent.) route as previously described
   [[264]12]. Briefly, a motorized stereotaxic instrument (Neurostar) was
   used to implant a 26-gauge guide cannula (RWD) into the third ventricle
   (coordinates from the bregma: − 1.6 mm posterior, 0 mm lateral, −
   2.5 mm ventral) of mice at one week before infection. Mice were
   injected with 0.5 µL of sterile 0.9% NaCl or 2 × 10^8 CFU/mL M. tb
   (administered at a dose of 10^5 CFU to each animal) through the brain
   cannula (over 5 min) using the syringe pump (Harvard Apparatus). All
   mice were observed for mortality and weight change. Humane endpoints
   included ≥ 20% weight loss, complete hind limb paralysis and repeated
   seizures. Infected mice were also monitored daily after infection for
   clinical signs indicative of CNS-TB, such as limb weakness, tremors,
   and twitches.

Drug treatment for M. tb-infected mice

   Anti-tuberculous (ATT) drugs Isoniazid (INH), Rifampicin (RIF),
   Pyrazinamide (PZA), and Ethambutol (EMB) were purchased from Sigma. All
   drugs were administered five days a week for a total duration of eight
   weeks via oral gavage in the following dosages as previously used in
   other murine TB studies; RIF(10 mg/kg), INH(25 mg/kg), PZA(150 mg/kg),
   and EMB(100 mg/kg) [[265]86, [266]87]. The eight weeks of duration of
   ATT treatment was similar to those used in other murine TB studies
   [[267]83, [268]88]. Drug solutions were prepared such that the desired
   concentration would be delivered in a 0.2 mL total volume. The 25 mg/kg
   dose of INH produces a serum area under the concentration-time curve
   from 0 to 24 h (AUC[0−24]) value similar to those observed in humans
   with a slow acetylator phenotype [[269]89]. RIF was given 45 min prior
   to administration of the other drugs to avoid any adverse
   pharmacokinetic interaction [[270]90–[271]92].

   SB-3CT (MedChemExpress) (50 mg/kg) or doxycycline (Sigma) (100 mg/kg)
   were administered into mice by intraperitoneal (i.p.) injection for
   five days per week for a total duration of eight weeks. Based on
   allometric scaling calculation [[272]93], the human equivalent dose for
   doxycycline use here is 8.1 mg/kg, which exceeds the maximum clinical
   dose for doxycycline at 100 mg twice daily. However, it has previously
   been reported that C57BL/6 mice given an oral doxycycline dose of
   100 mg/kg daily had similar mean plasma concentration of doxycycline as
   patients treated with 200 mg doxycycline daily [[273]94], thus we used
   a similar dose in this study. SB-3CT or doxycycline was diluted in the
   diluent comprised of 10% DMSO, 75% PEG-200 and 15% milli Q water daily
   prior to i.p. injection. Drug suspension was prepared such that the
   desired concentration would be delivered in a 50 µL total volume.

Histopathological analysis

   Histopathology was performed on the left hemisphere of the murine
   brain, which was fixed in 10% neutral buffered formalin,
   paraffin-embedded and sectioned to 4 μm thickness. Brain slices were
   stained with hematoxylin-eosin (Vector Laboratories, Burlingame,
   California) to characterize pathological lesions and Ziehl–Neelsen
   staining (Sigma-Aldrich, St. Louis, Missouri) to detect mycobacterium
   according to manufacturer’s instructions. Histopathology was evaluated
   in a blinded fashion by a histopathologist (R.R.) based on the presence
   of pathological changes including inflammation, perivascular cuffing,
   gliosis, granuloma (not well developed), pyogranuloma, and neuronal
   degeneration. Grading of severity was assigned on the following scale:
   0: no abnormalities detected; 1—minimal; 2—mild; 3—moderate; 4—marked
   and 5—severe. The presence of liquefactive necrosis and the presence of
   bacilli were also accessed. The presence of bacilli was indicated with
   a “+” for presence and a “-” for absence, without quantification.

Organ harvesting and processing

   Mouse brain was harvested at eight weeks after infection, or eight
   weeks post-treatment for mice in the drug study. Half of the brain was
   fixed in 10% neutral buffered formalin for histology, and the other
   half was homogenized for bacterial enumeration and characterization of
   immunological markers. The organ was homogenized by high-speed shaking
   in 2 mL microcentrifuge tubes filled with sterile PBS and 5/7 mm
   stainless steel beads using TissueLyser LT (Qiagen, Hilden). For
   bacterial enumeration, brain homogenates were cultured on selective
   Middlebrook 7H11 agar supplemented with 0.05 mg/mL carbenicillin
   (Sigma), 0.01 mg/mL amphotericin B (Sigma), 0.02 mg/mL trimethoprim
   (Sigma), 200 units/mL polymyxin B (Sigma), 0.5% glycerol and 10% OADC.

Sterilization of human CSF and murine brain homogenates

   Following experiments involving M. tb, samples required removal of M.
   tb by double filtration before analysis. CSF and organ homogenates are
   transferred to 0.22 μm centrifugal filter units (Millipore) and
   centrifuged at 12,000 x g for 5 min. The filtrate is again centrifuged
   using a new filter unit so each sample will be sterile filtered twice.

Total protein concentration quantification

   Total protein concentration was quantified using DC assay (Bio-Rad)
   according to manufacturer’s instructions. Five microliters of standard
   or samples were loaded into a 96-well plate. Next, 25 µL of Reagent A
   was added to each well, followed by an additional 200 µL of reagent B
   added into each well. The plate was placed on a shaker and mixed for
   5 s. After 15 min incubation at room temperature, the absorbance was
   read at 750 nm using a microplate reader.

Luminex bead array for human MMPs and TIMPs

   Human CSF MMP-1, −2, −3, −7, −9, −10 and EMMPRIN concentrations were
   analyzed by Fluorokine multianalyte profiling kit (R&D Systems). Human
   CSF TIMP-2 and − 4 concentrations were analyzed by human TIMP multiplex
   kit (R&D Systems). All analytes were analyzed using the Bio-Plex 200
   platform (Bio-Rad), according to the manufacturer’s protocol. The
   minimum detection Limits for the nine analytes were 1.1, 12.6, 7.3,
   6.6, 13.7, 3.2, 5.6, 14.7, 1.29 pg/mL, respectively. CSF samples were
   diluted with calibrator diluent with 1 in 5 dilution for the array of
   MMPs, and 1 in 100 dilution for the array of TIMP-2. Neat CSF was used
   to analyze TIMP-4 concentration. Twenty-five microliters of standard or
   samples were loaded into a 96-well plate, and incubated with 25 µL of
   microparticle cocktail (diluted 100-fold with Microparticle Diluent) at
   room temperature for 2 h shaking at 800 rpm. The microplate was washed
   three times, 25 µL of diluted Biotin Antibody cocktail was added and
   the plate was incubated for 1 h on the shaker. After three washes, 25
   µL of streptavidin-PE was added, incubated for 30 min, and further
   washed three times. The microparticles were resuspended with 100 µL of
   wash buffer, incubated for 2 min on the shaker, and read using the
   Bio-Plex analyzer. CSF concentrations of analytes were normalized to
   the total CSF leukocyte count. For patients who have 0 leukocytes in
   the CSF, the MMP concentration was normalized to 1 leukocyte.

Luminex bead array for mouse MMPs, adhesion molecules, cytokines and
chemokines

   Analysis of mouse analytes was performed using the Mouse Premixed
   Multi-Analyte Kit (R&D Systems) according to manufacturer’s
   instructions. The analytes with respective minimum detection limit are
   MMP-2 (48.8 pg/mL), MMP-3 (0.332 pg/mL), MMP-8 (2109 pg/mL), MMP-9
   (47.1 pg/mL), MMP-12 (0.424 pg/mL), TIMP-1 (12.4 pg/mL), TIMP-4 (14.4
   pg/mL), TNF-α (1.47 pg/mL), IFN-ɣ (1.85 pg/mL), IL-1α (8.17 pg/mL),
   IL-1β (41.8 pg/mL), IL-6 (2.3 pg/mL), IL-17 A (7.08 pg/mL), IL-27 (1.84
   pg/mL), CXCL1 (32.9 pg/mL), CXCL2 (1.97 pg/mL), CXCL10 (6.85 pg/mL),
   CXCL13 (19.3 pg/mL), CCL2 (134 pg/mL), CCL3 (0.452 pg/mL), CCL4 (77.4
   pg/mL), CCL5 (19.1 pg/mL), CCL7 (1.69 pg/mL), CCL11 (1.46 pg/mL), CCL12
   (0.613 pg/mL), CCL19 (2.28 pg/mL), and CCL22 (0.965 pg/mL). Neat
   concentrations of mouse brain homogenates were used. Twenty-five
   microliters of standard or samples were loaded into a 96-well plate,
   and incubated with 25 µL of microparticle cocktail (diluted 10-fold
   with Assay Diluent RD1W) at room temperature for 2 h shaking at
   800 rpm. The microplate was washed three times, 25 µL of diluted Biotin
   Antibody cocktail was added and the plate was incubated for 1 h on the
   shaker. After three washes, 25 µL of diluted streptavidin-PE was added,
   incubated for 30 min, and further washed three times. The
   microparticles were resuspended with 100 µL of wash buffer, incubated
   for 2 min on the shaker, and read using the Bio-Plex 200 analyzer.

Gelatinase inhibition assay using MMP-2 and MMP-9 antibodies

   Gelatin degradation by the neat concentration of CSF was assessed.
   Samples were activated with 2 mM of aminophenylmercuric acetate (APMA)
   for 2 h at 37 °C. Next, 80 ng of inhibitor MMP-2 antibody (Thermo
   Fisher [274]MA513590) or MMP-9 antibody (Sigma-Aldrich IM-09 L) was
   added per well and incubated for 1 h at 37 °C. Subsequently, 1 µg of DQ
   Gelatin was added per well and incubated for 6 h in the dark at room
   temperature. Gelatin degradation activity was measured at absorption
   485 nm and emission 528 nm with a fluorescence plate reader (BioTek
   Instruments). Results were normalized by the total protein
   concentration.

DQ gelatin degradation assay

   Gelatin degradation by the neat concentration of murine brain
   homogenates was assessed. Samples were activated with 2 mM of
   aminophenylmercuric acetate (APMA) for 2 h at 37 °C, before mixing with
   1 µg of DQ Gelatin. After 16 h of incubation in the dark at room
   temperature, Gelatin degradation activity was measured at absorption
   485 nm and emission 528 nm with a fluorescence plate reader (BioTek
   Instruments). Results were normalized by the total protein
   concentration.

NETs quantification

   NETs concentrations were quantified by measuring extracellular DNA
   concentration, a commonly used method to evaluate NETs levels in both
   in vitro and clinical studies [[275]10, [276]95, [277]96]. To take into
   account the number of neutrophils entering the CNS, extracellular DNA
   concentrations were corrected for CSF neutrophil count. NETs were
   quantified using QuantiT PicoGreen (Invitrogen) according to
   manufacturer’s instructions. In brief, a standard was made using serial
   1 in 10 dilutions of Lambda DNA commencing from 20 µg/mL to 2 ng/mL
   diluted in 1 X TE buffer. Fifty microliters of samples or standards
   were loaded onto 96-well black plates, and 50 µL PicoGreen at a 1 in
   200 dilution was added in each well. The plate was then incubated in
   the dark at room temperature for 5 min, and then read using the Synergy
   H1 microplate reader (BioTek).

Enzyme-linked immunosorbent assay (ELISA) for Citrullinated Histone 3 (CitH3)

   CitH3 concentration was assayed using Citrullinated Histone 3 (Clone
   11D3) ELISA kit (Cayman Chemical), according to the manufacturer’s
   instructions. CSF or brain homogenates were diluted 1 in 2 with Assay
   Buffer. Serial 1 in 2 dilutions of standards were made commencing from
   10 to 0.15 ng/mL. One hundred microlitres of standards or samples were
   added per well and incubated for 2 h at room temperature on a
   horizontal orbital microplate shaker. The plate was washed five times
   with Wash Buffer before adding 100 µL of 1X HRP Conjugate per well, and
   the plate was incubated for another hour at room temperature. The plate
   was washed five times, prior to the addition of 100 µL of TMB Substrate
   Solution and incubated for 30 min in the dark at room temperature. One
   hundred microliters of Stop Solution were added to each well, and the
   optical density was read using a microplate reader set to 450 nm. The
   lower Limit of detection is 0.1 ng/mL.

Mycobacterium tuberculosis culture for infection

   M. tb strain H37Rv was kindly provided by Professor Nicholas Paton
   (National University of Singapore, Singapore). M. tb was cultured in
   Middlebrook 7H9 medium (BD Difco) supplemented with 10% (v/v)
   Middlebrook albumin-dextrose-catalase (ADC) (BD Difco), 0.2% (v/v)
   glycerol (Sigma) and 0.05% (v/v) Tween-80 (Sigma) or on complete
   Middlebrook 7H11 agar (BD Difco) supplemented with 0.5% (v/v) glycerol
   and 10% (v/v) oleic acid-albumin-dextrose-catalase (OADC) (BD Difco).
   M. tb cultures were sustained in square PETG bottles (Nalgene) at 37 ᴼC
   with agitation at 80 rpm or plated on complete or selective 7H11 agar
   in petri dishes (Thermo Fisher Scientific). For infection experiments,
   M. tb was cultured to the mid-logarithmic phase at an optical density
   of 0.6 to 0.8. Prior to infection, the M. tb was centrifuged at 3200 g
   for 10 min and resuspended in 1 mL sterile 0.9% NaCl. The M. tb
   inoculum was then plated to determine the dose of infection.

ELISA for mouse TIMPs

   Mouse TIMP-2 concentration was measured using the Duoset ELISA
   Development System (R&D Systems) according to the manufacturer’s
   instructions. The lower limit for detection of TIMP-2 is 7.8 pg/mL.
   ELISA plates were coated with 100 µL of Capture Antibody at 1 µg/mL
   overnight at room temperature, then washed three times with Wash Buffer
   (0.05% Tween-20 in PBS). Free binding sites were blocked for 1 h with
   300 µL of Reagent Diluent (1% BSA in PBS) at room temperature, then
   washed three times. Fifty µL of standards or brain homogenates (diluted
   1 in 100 to 200 with water) were added to the appropriate wells and
   incubated for 2 h at room temperature. After three washes, 50 µL of
   Detection Antibody (50 ng/mL) was added for 2 h at room temperature.
   Three further washes were performed, then 50 µL of Streptavidin-HRP (1
   in 40 dilutions) was added into each well and incubated for 20 min in
   the dark at room temperature, before washing three times again. Fifty
   µL Substrate Solution (1:1 mixture of H[2]O[2] and
   Tetramethylbenzidine) was then added per well and the plate was
   incubated for 20 min in the dark at room temperature. The reaction was
   then stopped with 25 µL per well of Stop Solution (2 N H[2]SO[4]) and
   the plate read at 450 nm and 570 nm using the Synergy H1 microplate
   reader (BioTek, US). The 570 nm reading was subtracted from the 450 nm
   reading to account for imperfections in the plate.

   Mouse TIMP-3 concentration was measured using ELISA kit (LSBio),
   according to the manufacturer’s instructions. The lower limit for
   detection of TIMP-3 is 0.156 ng/mL. Brain homogenates were diluted 1 in
   50 with water. One hundred microliters of standard or sample was added
   per well and incubated for 2 h at 37 ᴼC. After 2 h incubation, mixture
   of each well was aspirated without washing. One hundred microliters of
   1X Biotinylated Detection Antibody was added into each well, incubated
   for 1 h at 37 ᴼC, and then washed three times. Each wash step was
   allowed to sit for 2 min before completely aspirating. One hundred
   microliters of 1X HRP-streptavidin was added per well and incubated for
   a further hour at 37 ᴼC. The plate was washed five times, prior to the
   addition of 90 µL of TMB substrate and incubated for 15 to 30 min in
   the dark at 37 ᴼC. Fifty microliters of stop solution were added to
   each well, and the optical density was read using a microplate reader
   set to 450 nm and 570 nm. The 570 nm reading was subtracted from the
   450 nm reading to account for imperfections in the plate.

Immunohistochemistry

   Tissue sections were deparaffinized and rehydrated for 5 min each in
   sequential order of Clearene, Clearene, 100% Ethanol, 100% Ethanol, 95%
   Ethanol, 70% Ethanol, and distilled water. Subsequently, heat-induced
   epitope retrieval was performed by incubating the sections in 1x
   Citrate Buffer, pH 6.0 (Sigma), for 30 min at 95 ᴼC. After antigen
   retrieval, sections were left in the Citrate Buffer at room temperature
   to cool for 20 min, before washing with PBS three times for 5 min each.
   Sections were surrounded with PAP pen (Vector laboratories) and treated
   with blocking buffer for 1 h at room temperature to prevent
   non-specific binding. The blocking buffer used was 10% goat serum in
   PBS. After blocking, primary antibodies were added to the sections. The
   antibodies with respective incubation condition are anti-histone H3
   (citrulline R2 + R8 + R17) antibody (Abcam Ab5103, 1:50 dilution for
   mouse samples or 1:100 dilution for human samples, 37 °C, 2 h
   incubation), anti-MMP9 antibody (Abcam Ab38898, 1:250 dilution for
   mouse samples or 1:100 for human samples, RT, 2 h incubation), CD16
   (D1N9L) rabbit monoclonal antibody (Cell Signaling Technology 24326 S,
   1:100 dilution, RT, 1 h incubation), CD68 clone PG-M1 mouse monoclonal
   antibody (Agilent M0876, 1:100 dilution, RT, 1 h incubation), CD4 mouse
   monoclonal antibody (Leica Biosystems NCL-L-CD4-368, 1:20 dilution,
   4 °C, 16 h incubation), rat α-Ly6G primary antibody (Abcam Ab25377,
   1:100 dilution, RT, 1 h incubation), and anti-NKR-P1C antibody
   EPR22990-31 (Abcam [278]Ab289542, 1:20 dilution, 4 °C, 16 h
   incubation). Blocking buffer alone was used as a control. Upon primary
   antibody incubation, sections were washed three times with PBS for
   5 min. This was followed by incubation with appropriate secondary
   antibodies for 30 min at RT: ImmPRESS goat anti-rabbit HRP (Vector
   Laboratories, MP7451) for rabbit monoclonal antibody, goat anti-rat HRP
   polymer (Abcam [279]Ab214882) for rat monoclonal antibody, and HRP
   polymer detector reagent (Abcam [280]Ab269452) for mouse monoclonal
   antibody. All primary and secondary antibodies were diluted in 10% goat
   serum in PBS.

   For chromogenic staining, the DAB Substrate Kit (Abcam) was used
   according to the manufacturer’s instructions. Sections were covered
   with DAB Substrate for 1 to 10 min until the desired color was
   achieved, then washed under running tap water for 2 min before
   counterstaining with hematoxylin for 5 min. Sections were washed with
   tap water again for 2 min followed by dehydration with 3 min each in
   sequentially order of 50% Ethanol, 70% Ethanol, 90% Ethanol, 100%
   Ethanol, 100% Ethanol, Clearene, and Clearene. Histological sections
   were covered with Organo/Limonene Mount and coverslip, air-dried
   overnight, and then stored at room temperature. Negative control for
   all IHC stains showed no immunoreactivity.

   For double immunofluorescence staining involving the use of two primary
   antibodies Raised in the same host, Opal 3-Plex Anti-Rb Manual
   Detection Kit (Akoya Biosciences) was used according to the
   manufacturer’s instructions. After dewaxing and rehydration, antigen
   retrieval of sections was performed using microwave treatment, 1 min at
   100% power followed by 15 min at 20% power. To block endogenous HRP
   activity, sections were incubated in 1% H2O2 in TBS for 1 h and washed,
   followed by blocking in 10% goat serum in TBS. All washing steps
   include washing in 1X TBS-Tween 20 (0.05%) three times in 5 min each.
   Anti-CD31 antibody (Abcam [281]Ab182981) was added to the sections at
   1:500 dilution, for 16 h incubation at 4 °C. After washing, secondary
   anti-rabbit HRP from the kit was added for 15 min incubation at RT.
   Sections were washed, then incubated with Opal 520 dye for 15 min at
   room temperature and further washed. Prior to the addition of the
   second primary antibody, the first primary antibody was eluted by
   repeating the microwave treatment. After washing, anti-αSMA antibody
   (Abcam [282]Ab124964) was added to the sections at 1:500 dilution, for
   16 h incubation at 4 °C. After washing, secondary anti-rabbit HRP from
   the kit was added for 15 min incubation at RT. Next, the sections were
   incubated with Opal 570 dye for 15 min at room temperature. After
   washing, nuclei were counterstained with DAPI solution for 5 min before
   the final washing step. The sections were mounted in ProLong Gold
   Anti-fade aqueous mounting medium. Negative control for all IHC stains
   showed no immunoreactivity.

Laser capture microdissection RNA sequencing

   Archival FFPE (formalin-fixed paraffin-embedded) tissue from adult
   patients with clinical biopsies for CNS-TB and patients with
   demyelination as controls were obtained from the Department of
   Pathology with ethics approval (Reference 2015/00067). For murine
   samples, TB-infected mouse brains and controls were fixed in formalin
   and embedded in paraffin blocks.

   Paraffin sections of 7 μm thickness were mounted on polyethylene
   naphthalate (PEN) membrane slides for laser-capture microdissection,
   with an adjacent H&E section used to confirm the region of interest.
   PEN membrane slides were deparaffinized with a series of xylene and
   ethanol wash, then stained with hematoxylin (Dako #S3309) and
   counterstained with bluing reagent (Thermofisher Catalog #6769002).
   Laser-capture microdissection was performed on the Leica LMD6 system,
   targeting regions of interest, including granulomas, regions of
   inflammation, and areas of healthy tissue as controls. Biological
   duplicates were collected in separate tubes. RNA-Seq libraries were
   prepared using Smart-3SEQ [[283]97]. Libraries were evaluated for their
   fragment size using the Agilent TapeStation4200 and quantified by qPCR
   on the Applied Biosystem StepOne Plus real-time PCR machine. Libraries
   were pooled and sequenced on an Illumina NextSeq500.

Bioinformatics analysis

   Quality control checks of raw sequencing data were executed using
   FastQC software (version 11.9). Alignment was performed using Kallisto
   (version 0.46) [[284]98], with sequence-based bias correction. The
   alignment was performed against the human transcriptome (Homo sapiens
   GRCh38), mouse transcriptome (Mus musculus GRCm39), and M. tb strain
   H37Rv transcriptomes, which served as index transcriptomes. Data were
   annotated using the ensembldb package (version 2.24.0). The transcript
   abundance was summarized to gene level using Sleuth (version 0.30)
   [[285]99]. Raw counts from RNA sequencing were processed using edgeR,
   Bioconductor version 3.17 [[286]100], with 3 variance estimated and
   size factor normalized using Trimmed Mean of M-values (TMM).

   Differential analysis was performed using edgeR (version 3.17) and
   filterByExpr function was used to filter genes that have a low number
   of reads. Samples with a median that is beyond twice the median
   absolute deviation (MAD) from the median of medians and any
   observations that are more than two IQR below Q1 or more than two IQR
   above Q3 are considered outliers were excluded from downstream
   analysis. Design matrix coded by model.matrix(~ 0 + variable) was used
   to identify the DEGs comparing two groups of interest. Genes with a
   false discovery rate (FDR)-corrected p-value < 0.05 are considered
   differentially expressed, following on from a likelihood ratio test
   using a negative binomial generalized linear model fit. Pathway
   enrichment analysis of significant up- and down-regulated DEGs was
   performed using web-based ShinyGO v0.77 [[287]101]. Limma with its
   voomWithQualityWeights function (version 3.56.2) was performed and the
   counts were used for principle component analysis. Gene signature
   enrichment was performed using Gene Set Variation Analysis (GSVA)
   [[288]102] using the gene sets from Molecular Signatures Database
   (MSigDB) (v2023.1) [[289]103] and based on literature. Key
   transcription regulators and their target genes are analyzed using
   web-based TRRUST v2 [[290]104].

   In silico cell fraction deconvolution was performed with web-based
   CIBERSORTx to estimate the abundance of immune cell types. LM22
   signature matrix which distinguishes 22 purified immune cell types is
   used for the human dataset [[291]30], and the ImmuCC signature matrix
   is used to infer the relative proportions of 10 major immune cells for
   the mouse dataset [[292]105].

Granuloma and blood vessel analysis

   The brain tissues harvested from M. tb H37Rv-infected C57BL/6 Nos2^−/−
   mice treated with ATT alone, or ATT with adjunctive MMP inhibitors, or
   vehicle control (n = 7 to 8 per group) were processed as FFPE sections.
   The first, tenth, twenty-fifth and fiftieth sections away from M. tb
   infection site were stained with hematoxylin-eosin. The number of
   granulomas present in each section was quantified manually in a blinded
   fashion by H.T.C. and T.H.H. Ziehl–Neelsen staining was performed on
   adjacent tissue sections and evaluated for the presence of AFB. Full
   slide HE and ZN stains were scanned using TissueFAXS Slide Scanner
   (TissueGnostics). The section that presented the most severe
   histopathology was selected as the representative section for each
   group, and the adjacent sections were used for vasculature analysis and
   drug distribution analysis.

   Blood vessels of both human and murine CNS-TB tissues were
   double-stained with antibodies detecting CD31 for endothelial cells and
   α-SMA for pericytes, which maintain the integrity of vessels and the
   BBB [[293]106]. IHC stains were scanned using Vectra 2 (PerkinElmer).
   Blood vessels were analyzed using customized semiautomatic software
   applications (Visiopharm). Deep learning application was trained for
   automatic selection, followed by manual verification and modification
   to generate outputs. The granuloma areas were outlined manually as
   regions of interest (ROI). Within the ROI, the area with CD31 stain was
   quantified as vessel area, while the area with CD31 and αSMA double
   stains was quantified as vessel area with pericyte coverage. The
   distance of blood vessels from the granuloma rim within the maximum
   range of 300 μm away from the granuloma rim was quantified using ImageJ
   software.

Liquid chromatography-tandem mass spectrometry

   Twenty-five microliters of mouse brain homogenates and standard
   calibrators were transferred into a deep 96-well plate and treated with
   180 µL of ice-cold Acetonitrile containing 0.1% Formic acid. The plate
   was mixed on a shaker at 1000 rpm/min for 10 min, then centrifuged at
   2270 x g for 50 min at 4 °C. One hundred microliters of the supernatant
   were carefully transferred to a 96‐microwell plate and loaded into the
   auto‐sampler (7 °C) for analysis by LC-MS/MS. Ion counts or areas under
   the MRM chromatogram were utilized for the construction of a standard
   curve for partial quantification.

   Multiple reaction monitoring (MRM) methods, analyses, and data
   acquisition using Agilent’s 1290 Infinity II LC/6495 tandem triple
   quadrupole mass spectrometer equipped with electrospray ionization
   (ESI) source in MRM mode and HILIC chromatographic conditions were as
   follows: Liquid chromatographic (LC) separation of the biomarkers was
   carried out on an Agilent 1290 Infinity II LC system (Agilent
   Technologies) with PEEK coated SeQuant^®ZIC^®-cHILIC 3 mm,100Å
   100 × 2.1 mm HPLC column (Merck Pte Ltd, Singapore) maintained at
   40 °C. The organic solvent was Methanol containing 2 mM Ammonium
   Formate and 2% Formic acid (Solvent A) and the aqueous solvent used was
   water containing 2 mM Ammonium Formate and 2% Formic acid (Solvent B).
   A linear LC gradient on Binary Pump (Agilent Model G7120A) was set up
   with a percentage of Solvent B as follows: 60% between 0 and 0.05 min,
   10% between 1.00 and 2.00 min, and back to 60% between 2.01 and
   3.5 min, with a flow rate of 0.4 mL/min. The column was further
   equilibrated for a further 7 min with 60% Solvent B. The sample
   injection volume was 10 µL. For mass detection, the LC eluent was
   connected online to an Agilent 6495 Triple Quadrupole MS system
   (Instrument model G6495A, Agilent Technologies) operated with the
   electrospray source in positive ionization mode. The electrospray
   ionization source conditions were as follows: capillary voltage of
   4.0 kV, nozzle voltage of 500 V, iFunnel parameter high/low-pressure RF
   of 200 V, nebulizer pressure of 45 psi, the gas temperature of 290 °C,
   sheath gas temperature of 290 °C, and sheath gas flow of 20 L/min. For
   quantitative analyses, the analytes were detected by means of
   characteristic product ions formed from protonated molecules by
   collision-induced dissociation (CID) using the MRM mode after
   fragmentation with 99.9995% pure nitrogen. The MRM transitions are
   shown in Supplemental Table [294]S1. Dwell time was set to 100 ms and
   all transitions were in positive mode. Acquisitions and quantitative
   analysis were performed using the Agilent MassHunter Workstation
   Acquisition (10.0.127) and Agilent MassHunter Workstation Quantitative
   Analysis (10.0) software.

   Calibration curves were constructed using standard stock solutions in
   water. For quantification, a calibration curve for each compound was
   constructed by serial dilution at decreasing concentrations. QCs were
   set at low, mid, or high concentrations. A linear curve was plotted,
   and the limits of quantitation (LOQ) were determined. The linearity
   coefficients of determination (r2) for all metabolites were > 0.9.

Mass spectrometry imaging

   The section that presented severe histopathology with the presence of
   granulomas was selected as the representative section for each mouse
   group. FFPE tissue sections (5 μm thickness) embedded on the conductive
   glass slide, Intellislide (Bruker Corporation) underwent a dewaxing
   process. The slides were placed in an oven at 60 °C for 60 min to melt
   the parafilm. Then the slides were immersed in solution in the
   following order to remove the parafilm: 100% xylene twice for 3 min
   each time, 100% EtOH twice, 95% EtOH once, 70% EtOH once, for 1 min
   each time, and lastly in water twice for 3 min each time. The slides
   were placed in a vacuum desiccator for about 15 min to dry.

   After drying, external calibrants of poly-DL-alanine (pA)
   (Sigma-Aldrich) (0.15 mg/mL in H2O) and a mixed standard of INH, PZA,
   EMB, RIF, Doxycycline, and SB-3CT (2.5 to 25 µg/mL in H2O, stock SB-3CT
   in DMSO) were spotted (0.5 µL) at two locations on each slide and dried
   again in the vacuum desiccator. The slides were then scanned with the
   flatbed scanner Epson Perfection V39 (Nagano) at 3200 dpi. After
   scanning, the slides were sprayed with HCCA solution (Bruker
   Corporation) (10 mg/mL in 70%/30%/0.2% ACN/H2O/TFA) with the M3 TM
   sprayer (HTX Technologies LLC) with solvent flow of 0.12 mL/min, nozzle
   temperature of 75 °C, four passes in horizontal spray pattern,
   1200 mm/min nozzle velocity with a track spacing of 3 mm. After
   spraying, the slides were stored in 50 mL Falcon tubes with silica gel
   as desiccants until MSI experiment.

   The tissue sections were scanned with the rapifleX TOF/TOF MALDI mass
   spectrometer (Bruker Daltonik). The mass spectrometer was operated in
   positive, reflector mode. External mass calibrations were performed
   using spots of pA and mixed standard spotted on the slide. Mass spectra
   were acquired in the mass range of 100 to 1000 Da, at a spatial
   resolution of 50 μm, with the Smartbeam 3D laser operated in Single
   beam mode at 40% laser fluence power. A blank area of matrix and a
   small mixed standard section (< 200 positions) were also scanned.

   Post-acquisition analysis was done using the SCiLS Lab 3D (v. 2016b)
   software (Bruker Daltonik). Root mean square (RMS) algorithm was used
   for normalization. The TB drug ions within the tissue sections data
   were annotated by matching ion peaks with the previously identified
   mixed standards peaks. Automatic hotspot removal, and weak denoising
   were used for ion image processing.

Statistical analysis

   The statistical analyses performed for continuous variables depended on
   the distribution (parametric or nonparametric) of the data. Where the
   data followed a normal distribution (based on D’Agostino-Pearson or
   Shapiro-Wilk normality test), the Student’s t-test for two groups or
   one-way ANOVA with Dunnett’s multiple comparison test against control
   group was used; Mann-Whitney or Kruskal-Wallis with Dunn’s multiple
   comparisons test was used for data with non-normal distribution. In
   correlation analyses, normally distributed data was analyzed with
   Pearson correlation coefficient, and non-normal distributed data was
   analyzed with Spearman correlation coefficient. Paired t-test was used
   to compare the differences in matched groups. The log-rank test was
   used for the analysis of Kaplan-Meier survival curves. Data are
   presented as median and interquartile range or mean and standard
   deviation as stated in the Fig. legend. A two-sided value of p < 0.05
   was considered statistically significant. All analyses were performed
   using GraphPad Prism version 7.05 (GraphPad).

Supplementary Information

   [295]Supplemental File S1^ (14KB, xlsx)
   [296]Supplemental File S2^ (17.7KB, xlsx)
   [297]Supplemental File S3^ (121.8KB, xlsx)
   [298]Supplemental File S4^ (17.6KB, xlsx)
   [299]Supplemental File S5^ (987.2KB, xlsx)
   [300]Supplemental File S6^ (12.1KB, xlsx)
   [301]Supplemental File S7^ (17.7KB, xlsx)
   [302]Supplemental File S8^ (11.7KB, xlsx)
   [303]Supplemental Figures and Tables^ (8.2MB, docx)

Acknowledgements