Mycobacterium bovis Bacillus Calmette–Guérin–Infected Dendritic Cells Induce TNF-α–Dependent Cell Cluster Formation That Promotes Bacterial Dissemination through an In Vitro Model of the Blood–Brain Barrier

One of the leading causes of death among infectious diseases is tuberculosis (TB) caused by Mycobacterium tuberculosis. These bacteria primarily infect the lung but can disseminate to other parts of the body, including the CNS, leading to tuberculous meningitis (1). CNS TB (CNSTB) is a devastating disease that causes serious neurologic damage and severe disability with a high mortality rate (2–4). It has been proposed that mycobacterial lesion formation on the vessels of the CNS facilitates the dissemination (5); however, the mechanism of how mycobacteria traverse the blood–brain barrier (BBB) to access the brain parenchyma is still unknown.

There are three biological barriers in the CNS, including the meninges, the choroid plexus, and the BBB, that could serve as an entry route for mycobacteria from blood vessels (6). A study using a human in vitro BBB model showed that mycobacteria are capable of invading endothelial cells (ECs) and traverse the endothelial monolayer without a cellular carrier (7). In contrast, previous studies have described mycobacteria crossing the cellular layers within carrier cells (4) in a lung in vitro system (8) and in zebrafish models (9, 10), providing a basis for the “Trojan horse” theory. Mycobacteria are intracellular microorganisms, and in the body, they reside mainly in macrophages and dendritic cells (DCs) (11). Previous studies have shown that CD11c^high DCs are the sole emigrating cell type from primary granulomas carrying live mycobacteria that result in the dissemination and formation of new granulomas through capturing of Ag-specific T cells (12). Mycobacterium carrying DCs promote granuloma formation (12) that can be facilitated by their limited migration (11–13).

In vitro granuloma models offer a unique opportunity to elucidate the mechanisms of early granuloma formation and analyzing host–mycobacteria interactions in a controlled environment (14). Several groups have established in vitro granuloma models using human PBMCs infected with M. tuberculosis or the attenuated strain M. bovis bacillus Calmette-Guérin (BCG) Pasteur. With these models, the following multiple aspects of the granuloma pathogenesis were studied: 1) secretion of cytokines promoting granuloma formation (such as TNF-α) (15, 16), 2) dormancy (16), 3) formation of multinucleated giant cells (17), 4) differences between active and latent TB infections (18), and 5) three-dimensional (3D) environment to study host–pathogen interactions (19). These models are suitable for analyzing intercellular interactions within an artificial granuloma but lack the possibility of studying M. tuberculosis dissemination in situ or across a vessel wall. To address this, we developed a novel, to our knowledge, combination model of an in vitro granuloma and BBB coculture system (20). This model allows us to study the interaction of Mycobacterium-infected DCs, Mycobacterium Ag-specific P25 T cells, PBMCs, and the BBB. In this study, we show that in vitro granulomas form on the surface of brain ECs when BCG-infected DCs are cocultured with PBMCs and P25 T cells. These in vitro cell clusters are complex in cellular composition, resemble in vivo lung granuloma morphology, and facilitate in situ bacterial dissemination. We also confirmed that DCs are restricted in migration postinfection (12, 13), which we show to be a possible cause of altered intracellular actin redistribution. The in vitro cell aggregates interfere with BBB and brain EC functions. The in vitro granuloma cell cluster formation leads to reactive oxygen stress of ECs, cluster-associated damage (CAD), and increased transmigration of infected cells across the EC monolayer. Treatment with the TNF-α blocker MP6-XT22 decreased cluster formation, adherence to brain ECs, and diminished the CAD effect. Our data contribute to the understanding of mycobacterial dissemination across brain endothelial monolayers, giving insight for developing therapeutic targets to inhibit TB meningitis.

All experimental procedures were performed in accordance with the guidelines of the University of Wisconsin Institutional Animal Care and Use Committee using approved protocols. Wild-type C57BL/6J, CD11c-eYFP (B6.Cg-Tg(Itgax-Venus)1Mnz/J) (21), and CX3CR1-GFP (B6.129P-Cx3cr1tm1Litt/J) mice were purchased from The Jackson Laboratory (Bar Harbor, ME) and housed in microisolator cages at the University of Wisconsin pathogen-free facility Animal Care Unit (Madison, WI). The P25 mice express a transgenic T cell Ag receptor recognizing M. tuberculosis Ag 85b^(250-254) epitope restricted to MHC class II IAb (11) were a gift of Dr. A.G. Rothfuchs and Dr. A. Sher (National Institutes of Health, Bethesda, MD). The mT/mG tdTomato-expressing P25 T cells were isolated from the F1 generation of a cross between homozygous tdTomato and P25 mice. The LifeAct-RFP mice (22) were a gift from Dr. J. Jacobelli (Department of Biomedical Research, National Jewish Health; Department of Immunology and Microbiology, University of Colorado School of Medicine). LifeAct-RFP mice were generated to study F-actin dynamics in living cells that express the 17-aa LifeAct peptide driven by a chicken actin promoter and CMV enhancer. These are transgenic global reporter mice that express LifeAct fused to mRFPruby on F-actin. LifeAct-RFP⁺ mice were crossed with CD11c-eYFP mice, generating CD11c-eYFP⁺LifeAct-RFP⁺ mice, from which primary cells were isolated.

The M. bovis BCG Pasteur strain was provided by Dr. G. Fennelly (Albert Einstein College, Bronx, NY). BCG was electroporated with tdTomato (pTEC27) or E2-Crimson (pTEC19) plasmids, which were generated by Dr. L. Ramakrishnan (Addgene plasmid 30182 and 30178, respectively; Cambridge, MA) (23). Strains were stored as frozen aliquots at −80°C. All BCG was grown at 37°C in Middlebrook 7H9 media (BD, Franklin Lakes, NJ) supplemented with 10% oleic acid–albumin–dextrose–catalase enrichment (BD, Franklin Lakes, NJ) and 0.05% Tween 80 (Sigma-Aldrich, St. Louis, MO) in the presence of Hygromycin B Gold (100 μg/ml; InvivoGen, San Diego, CA) in a shaking incubator at 37°C.

DCs were generated from the bone marrow of adult wild-type C57BL/6J or CD11c-eYFP or CD11c-eYFP⁺LifeAct-RFP⁺ mice as described previously (24–27). Briefly, bone marrow was extracted from the femurs of adult wild-type or CD11c-eYFP mice, cell suspensions were treated with ammonium chloride potassium–containing lysis buffer (STEMCELL Technologies, Vancouver, Canada) to lyse the erythrocytes, washed, then plated in RPMI 1640 supplemented with 10% FBS, 100 U/ml penicillin/streptomycin, 1% HEPES, 1% GlutaMAX (Thermo Fisher Scientific, Waltham, MA), 1% nonessential amino acids (Thermo Fisher Scientific), 1% essential amino acids (Thermo Fisher Scientific), 1% Na-pyruvate (Thermo Fisher Scientific), 50 μM 2-ME (Sigma-Aldrich, St. Louis, MO), and 20 ng/ml GM-CSF (PeproTech, Rocky Hill, NJ). Cells were seeded onto 100-mm nonadherent petri dishes (Falcon; Corning, Corning, NY), and maintained at 37°C in a humidified incubator with 5% CO₂. Cells were fed on day 3, and nonadherent and loosely adherent bone marrow–derived DC (BMDC) precursors were passaged on day 6 using the same medium containing 10 ng/ml GM-CSF. Surface expression of CD11b, CD11c, and MHC class II were confirmed by flow cytometry. Cultures were infected with BCG on day 8 for 1–4 h with a multiplicity of infection 1:1 in 2% FBS–RPMI in a 48-well nonadherent plate. Cells were washed postinfection and used for experiments. Infection was confirmed with both flow cytometry and fluorescent microscopy.

3D collagen chemotaxis assays were performed with μ-Slide Chemotaxis^3D following the manufacturers protocol (Ibidi, Fitchburg, WI). The μ-Slide Chemotaxis^3D device is composed of three observation areas, each of which are flanked on both sides by a reservoir such that once the observation area is seeded and the collagen has solidified, a concentration gradient develops through the addition of media plus cellular stimuli to one reservoir and media without stimuli to the other, which serves as a no-chemokine control for each experiment. Infected or uninfected CD11c-eYFP or CD11c-eYFP⁺LifeAct-RFP⁺ BMDCs at a concentration of 3 × 10⁶ cells/ml were suspended in 5 mg/ml type I collagen gel (Ibidi) at a final concentration of 1.5 mg/ml supplemented with 10× DMEM (Sigma-Aldrich, St. Louis, MO), 1× DMEM (Sigma-Aldrich, St. Louis, MO), NaOH in ultrapure H₂O, 7.5% NaHCO₃ (Sigma-Aldrich St. Louis, MO), and ultrapure H₂O on ice. Immediately after suspension, the cell–collagen mixture was injected into the observation area of the μ-Slide and placed into an incubator for 30 min to allow for gelation. Once fibrils are visible, 200 ng/ml recombinant mouse CCL2/JE/MCP-1 (PeproTech, Rocky Hill, NJ) chemokine in RPMI 1640 (Corning, Corning, NY) was added to one reservoir and only RPMI 1640 to the other as a no-chemokine control. For measuring distance and velocity, the slide was placed in a stage-top Tokai Hit biochamber at 37°C and 5% CO₂, and pictures were taken every 2 min for 3 h on a Leica SP8 STED 3X Super-Resolution Confocal Microscope at ×10 magnification. For morphodynamic analysis, CD11c-eYFP⁺LifeAct-RFP⁺ DCs were used in the same setup with pictures of BCG-infected and -uninfected cells taken at original magnification ×63 every 30 s for 10 min. Single-cell protrusion outputs and actin localization were analyzed using the ImageJ open source plugin Automated Detection and Analysis of ProTrusions (ADAPT) version 1.185. Observation of assembly and evaluation of the timing of in vitro granuloma formation was performed using a two-well culture insert microdish (Ibidi). ECs were grown on the ibiTreat surface, and the model was assembled on the top of the confluent EC monolayer using 7 × 10⁴ tdTomato BCG-infected DC cocultured with 7 × 10⁴ P25 PBMCs. Before the assembly, PBMC cultures were labeled with anti-CD4–Alexa Fluor 647 conjugated Ab. Cell imaging was performed using confocal live-cell imaging (Leica SP8, Leica Microsystems, Wetzlar, Germany). Stable temperature and CO₂ levels were maintained by a stage-top incubator (Tokai Hit, Fujinomiya, Japan). Cell aggregate formation was followed 1–4 h after assembly, and pictures were taken every 2 min for 3 h. Picture analysis was performed using ImageJ (National Institutes of Health open source software).

Primary mouse brain ECs were isolated from adult wild-type C57BL/6J mice, according to previously described methods (28–30). Harvested cells were plated on plastic surfaces coated with collagen type IV and fibronectin (100 µg/ml each; Sigma-Aldrich, St. Louis, MO). Cell culture medium consisted of DMEM/F12, FBS (15%; Corning, Corning, NY), heparin (100 µg/ml; Sigma-Aldrich, St. Louis, MO), insulin (5 µg/ml), transferrin (5 µg/ml), sodium selenite (5 ng/ml; Sigma-Aldrich, St. Louis, MO), basic fibroblast growth factor (1 ng/ml; Sigma-Aldrich, St. Louis, MO), and gentamicin (50 µg/ml; Sigma-Aldrich, St. Louis, MO). Cells were cultured for 4 d in the presence of 4 µg/ml puromycin (Sigma-Aldrich, St. Louis, MO) to eliminate contaminating cell types (31). Primary mouse brain ECs were seeded on Transwell, polycarbonate membrane (3-µm pore size; Corning) and cocultured with glial cells (28–30). After 2 d of coculture, 550 nM hydrocortisone (Sigma-Aldrich, St. Louis, MO) was added to the medium to facilitate intercellular junction formation (32, 33). Primary cultures of mouse glial cells were obtained from 1- or 2-d-old C57BL/6J wild-type mice, as described previously (29, 30).

Transendothelial electrical resistance (TEER) was measured by the EVOM2 Voltohmmeter (World Precision Instruments, Sarasota, FL) using STX2 electrodes. Resistance was calculated according to the surface of the Transwell inserts (Ω × cm²), and TEER of cell-free inserts (70 Ω) was subtracted from these values. TEER values were confirmed prior to all experiments (150 Ω × cm²), and a low permeability value was validated with showing 3.36 ×/10⁻⁶ cm/s permeability coefficient (P_e) for sodium fluorescein and 0.45 ×/10⁻⁶ cm/s for Evans blue–labeled albumin.

Primary blood lymphocytes were isolated from 8- to 12-wk-old C57BL/6J wild-type, P25, CD11c-eYFP reporter or CX3CR1-GFP reporter mice using the Ficoll gradient method (GE Healthcare, Chicago, IL). Blood from anesthetized mice was collected in 10-ml PBS with 20 U/ml heparin (Sigma-Aldrich, St. Louis, MO) and layered to the top of 7-ml Ficoll gradient in a 50-ml centrifuge tube. The gradient was centrifuged at 740 × g for 30 min at 20°C with the brake off. The upper layer of the gradient containing blood plasma and platelets was discarded, and the mononuclear cell layer was collected and processed for in vitro granuloma studies.

The in vitro coculture BBB model was established as described previously (20). For each experiment performed using the Transwell inserts, 1 × 10⁵ uninfected or infected DCs were cultured on the top of brain ECs with or without the addition of 1 × 10⁵ P25 PBMCs. All groups received 20% liver granuloma supernatant to enhance in vitro granuloma formation (20). The cell culture medium consisted of DMEM/F12, 10% FBS (Corning, Corning, NY), heparin (100 µg/ml; Sigma-Aldrich, St. Louis, MO), insulin (5 µg/ml), transferrin (5 µg/ml), sodium selenite (5 ng/ml; Sigma-Aldrich, St. Louis, MO), and gentamicin (10 µg/ml; Sigma-Aldrich, St. Louis, MO). When TNF-α neutralization was tested, neutralizing Ab MP6-XT22 (34) was added at 50 ng/ml (BioLegend, San Diego, CA).

The cellular composition, morphology, and functional changes of the in vitro aggregates were characterized by immunohistochemical staining. The in vitro granuloma models were fixed 1 or 2 d after the assembly with 2% paraformaldehyde (PFA; Electron Microscopy Sciences, Hatfield, PA) in PBS for 1 h, and cells were stored in PBS until staining. To define cellular composition of the in vitro aggregates, cultures were stained with Abs for CD4 conjugated to PE (clone RM4-5, 1:200, 553048; BD Biosciences, Franklin Lakes, NJ), CD11b conjugated to PE (clone M1/70, 1:200, 553311; BD Biosciences) or allophycocyanin (clone M1/70, 1:200, eBioscience, Santa Clara, CA, 17-0112-82), and B220 conjugated to PE (clone RA3-6B2, 1:200, 553090; BD Biosciences). Nonconjugated primary Abs against integral membrane tight junction marker Claudin-5 (Cl-5; rabbit anti-Cl5, 1:400, SAB4502981; Sigma-Aldrich, St. Louis, MO), junctional-associated protein zonula occludens-1 (rabbit anti–ZO-1, 1:400, 61-7300; Thermo Fisher Scientific), VCAM-1 (rat anti-VCAM-1, CD106, 1:100, 14-1061-82; Thermo Fisher Scientific), and biotinylated intercellular adhesion molecule-1 (ICAM-1) and CD54 (1:200, BAF796; R&D Systems, Minneapolis, MN) were used to test barrier characteristics. Secondary Abs goat anti-rabbit Alexa Fluor 568 (1:400, A-11008, Thermo Fisher Scientific), goat anti-rat Alexa Fluor 568 (1:400, A-11077; Thermo Fisher Scientific), and streptavidin–Alexa Fluor 647 (1:200, 405237; BioLegend, San Diego, CA) were used to detect stainings. All samples were permeabilized with 0.1% Triton X-100 (Sigma-Aldrich, St. Louis, MO) in PBS detergent for 10 min, 4°C, and blocked in 1% BSA–PBS for 3 h at room temperature. Conjugated Abs were incubated for 1 h at room temperature with the nucleus dye. Cells were washed three times with PBS between each step, and incubations were always performed in blocking buffer. Samples were mounted on slides using the Fluoromount-G Mounting Medium (SouthernBiotech, Birmingham, AL). Images were obtained using the Olympus FV1200 IX83 confocal microscope (Olympus, Shinjuku, Japan) and analyzed with ImageJ (National Institutes of Health open source software).

Cell cluster formation was analyzed with two methods to determine cluster size and cell number threshold. During the first 24 h of incubation, floating and attached immune cell cluster formation was observed using the 2-Well Culture-Insert microdish (Ibidi) with phase-contrast microscopy. Before the addition of infected and uninfected DCs and PBMCs, primary brain ECs were grown to confluency. Pictures were taken after 30 min of assembly at four sides of the microdish, and the cluster number was quantified with the ImageJ software. Cell clusters were defined as cellular aggregates consisting of at least five cells per cluster. This threshold was kept uniform during the analysis.

In subsequent experiments, coculture of cells was kept together using cell culture Transwell inserts. After 24 and 48 h, supernatants were removed, and the barrier forming primary brain ECs with attached cells and cell clusters was fixed and stained using immunohistochemistry. Images were taken of entire inserts for which aggregates consisting of at least five cells were counted manually.

The mitochondrial network was visualized using the MitoTracker Orange dye (Thermo Fisher Scientific) according to the manufacturer’s protocol. Briefly, primary brain ECs were grown on collagen type IV–coated round cover slips (VWR International, Radnor, PA) until confluency, and the in vitro granuloma model was established using 3 × 10⁵ uninfected or BCG-infected DCs with or without 3 × 10⁵ P25 PBMCs. After 1 d of coculture, supernates were removed, cultures were washed once with PBS, and cells were incubated with 0.1 µM of MitoTracker Orange (Thermo Fisher Scientific) in serum free DMEM for 40 min at 37°C. After staining, cells were washed with PBS and fixed with 2% PFA in PBS for 1 h. Samples were mounted and analyzed with confocal microscopy.

To count CD11c-eYFP cells that transmigrated the membrane and remained adherent underneath the cell culture inserts, we used Z-stack fluorescent confocal imaging. In this study, only CD11c-eYFP that carried bacteria and associated with cell clusters was counted.

Migration of CD11c-eYFP cells across the EC monolayer was studied by cytofluorometry of the cells in the bottom wells. A total of 1 × 10⁴ CellTrace Violet–labeled wild-type DCs was added to the bottom well as a buffer cell population. CellTrace Violet labeling was performed according to the protocol of the manufacturer (Invitrogen, Life Technologies, Waltham, MA). Absolute cell numbers were determined with the AccuCheck Counting Beads for flow cytometry (Life Technologies, Waltham, MA). Data were collected using a BD LSR II flow cytometer (BD Biosciences) and analyzed with FlowJo v.8.7 (FlowJo, Ashland, OR).

Cells were incubated with cell viability Ghost Dye (1:100; Tonbo Biosciences, San Diego, CA) in PBS for 10 min at 4°C, then fixed for 40 min with 2% PFA–PBS at room temperature. Specification of anti-CD4, CD8, B220, CD11c, and CD11b Abs are listed in the Immunohistochemistry section. Data were collected using an LSR II flow cytometer and analyzed with FlowJo v.8.7.

For statistical analysis, GraphPad Prism 5.0 software (GraphPad Software, San Diego, CA) was used. Results are given as mean ± the SEM. Multiple comparisons were made using one-way and two-way ANOVA with Bonferroni tests. Two-tailed unpaired t test analysis was used to compare measures made between two groups. The p values < 0.05 were being considered statistically significant. All types of experiments were repeated at least two to three times using independent cell isolations for all types of primary cells with n ≥ 3.

The migration of infected DCs and dissemination of BCG from primary granulomas by infected DCs can result in Ag-specific T cell arrest and the subsequent formation of secondary and tertiary granulomas (12). To study the effect of mycobacterial infection on the migratory capacity and morphokinetics of DCs, we infected BMDCs from the CD11c-eYFP reporter mice with BCG E2-Crimson. Infected and uninfected DC migration was analyzed using Ibidi μ-Slide chemotaxis chambers with media supplemented with 200 ng/μl CCL2 on one side and only media on the other. CCL2 contributes to the recruitment of immune cells to the site of infection and is necessary for DC migration into the CNS in experimental autoimmune encephalomyelitis (35–37). A mixture of infected (CD11c-eYFP⁺BCG E2-Crimson⁺) and uninfected (CD11c-eYFP⁺BCG E2-Crimson^-) DCs were placed in a 3D collagen suspension and loaded into the migration chamber, where a picture was taken every 2 min for 3 h to create a video (Supplemental Video 1) in which single-cell migration kinetics were tracked using the manual tracking plugin from ImageJ. Tracking data from a total of 52 uninfected and 93 infected cells from three separate experiments showed a reduction in track length by infected cells compared to uninfected cells (Fig. 1A) that correlated with a reduction in the accumulated distance by infected DCs (Fig. 1B). Additionally, the migration velocity of uninfected DCs was roughly four times greater than that of BCG-infected DCs within the 3 h time frame (Fig. 1C). These data show that the migration kinetics of DCs during infection with BCG is diminished both in velocity and in distance traveled.

As we have shown, BCG has the ability to restrict DC migration during infection (Fig. 1). Next, we wanted to determine the effect of BCG infection on DC polarity and actin distribution to find a possible cause for the reduction of migration. To study actin localization and distribution in regard to DC cell polarity during BCG infection, DCs from CD11c-eYFP⁺LifeAct-RFP⁺ mice were used to visualize the actin distribution and kinetics in combination with live-cell imaging. DCs were infected with E2-Crimson BCG for 1 h at a multiplicity of infection of 1, then suspended in a 3D collagen matrix and injected into the μ-Slide migration chamber. Pictures were taken of 10 infected and 11 uninfected cells at the single-cell level every 30 s for 10 min to create a video for analysis. The ImageJ plugin ADAPT was used to measure actin distribution and kinetics within each cell. Our data show that infected DCs have a round morphology with reduced continued polarization in any specific direction, whereas uninfected cells usually expressed polarization in a single direction (Fig. 2A). For cells to polarize and subsequently migrate toward a chemotactic signal, they first must be able to protrude in one direction (leading edge) of the cell while simultaneously retracting the other end of the cell (38). We measured the protrusion and retraction of the cell membrane and found that infected cells displayed a tendency to either protrude or retract (Fig. 2A, bottom row, cell outline all green or red), whereas uninfected cells were able to accomplish both actions simultaneously and more often (Fig. 2A, top row, cell outline green and red). Analysis showed that, in infected DCs, ∼60% of the time the DC cell membrane either protruded or retracted, whereas uninfected DCs experienced this movement only around 5% of the time (Fig. 2B). Additionally, uninfected DCs were able to simultaneously protrude and retract ∼75% of the time, in contrast to infected cells that were observed less than 20% of the time (Fig. 2C). Next, we studied the actin distribution and expression by measuring the mean fluorescence intensity at the center of the cell (arbitrary units = 0) to the outer boundary (arbitrary units = 1) (Fig. 2E). We found that actin distribution was visibly expressed at the center of the infected DCs through the outer boundary, whereas uninfected cells had most of the actin localized to the outer boundary (Fig. 2D). Compared with uninfected DCs, mean fluorescence intensity analysis confirmed our observations that infection increased actin expression toward the center of the cell and through to the outer cell boundary (Fig. 2F). Our data show that infected DCs exhibit different morphokinetic and actin distribution than uninfected DCs, which may play a role in the infection-mediated reduction of cell migration.

In vitro granuloma models are useful to understand the formation of granulomas and define cellular interactions in mycobacterial infections (14, 39). By combining BCG-infected or -uninfected CD11c-eYFP reporter DCs with or without PBMCs from mycobacterial Ag-specific P25 TCR transgenic mice and the primary mouse brain EC BBB model, we were able to visualize and characterize early cellular interactions of infected DCs with other immune cell types on brain ECs (Fig. 3A). We detected cellular cluster formation as early as 30 min when infected DCs were used (Fig. 3B). However, after 1 h, small clusters began to form more frequently in the presence of P25 PBMCs with infected DCs (Fig. 3C). To observe the dynamics of cluster formation, live-cell imaging was performed between 1 and 4 h after the assembly (Supplementary Video 1). We found that when BCG-infected DCs and PBMCs from P25 TCR transgenic mice are cocultured on brain endothelial monolayers, cluster formation is dynamic. Single cells come together to form smaller clusters, which then form larger aggregates (Supplemental Video 1). These data show that BCG-infected DCs induce rapid cellular cluster formation on brain EC monolayers.

Next, we wanted to understand whether the formation of these clusters on the in vitro BBB affects the migration of infected cells across the barrier. To answer this question, we quantified the amount of BCG-infected DC migration across the brain EC monolayer. 3D rendering of fluorescent confocal images organized in Z-stacks of clusters from BCG-infected DCs cocultured with PBMCs shows CD11c-eYFP⁺ and CD11c-eYFP⁻ cells traversing the EC monolayer carrying bacteria (Fig. 4A, 4C). Analyzing cellular migration of BCG-infected DCs and infected DCs with PBMCs for 24 and 48 h using cytofluorimetry showed that infected DCs were able to migrate across the monolayer (Fig. 4B). However, BCG-infected DC migration across the EC monolayer significantly increased in the presence of PBMCs: ∼8-fold after 24 h and ∼3-fold after 48 h (Fig. 4B). Confocal microscopic analysis of the inserts showed that some infected DCs remained adherent to the bottom side of the membrane (Fig. 4Ci–iii). This action was amplified in the presence of PBMCs when compared with groups without PBMCs (Fig. 4D). Together, these results indicate that cluster formation facilitates the dissemination of infected CD11c-eYFP cells across the brain EC monolayer.

To determine the cellular composition and distribution of cells within the clusters formed on the surface of brain endothelial monolayers, BCG-infected CD11c-eYFP DCs and PBMCs were cocultured for 24 and 48 h and analyzed using confocal fluorescent microscopy (Fig. 5A). Eighty cell clusters were analyzed to determine cell composition and distribution throughout the cluster. Our data show that CD11b⁺ cells were the most abundant cell type, making up, on average, 56% of the cells present in the clusters (Fig. 5A, 5B). CD11c-eYFP DCs and CD4⁺ T cells make up 23% and 15% of the cellular aggregates, respectively (Fig. 5B). DCs were found throughout the cluster and in contact with CD4⁺ cells (Fig. 5A). The number of B220⁺ cells present in these clusters was below 4% (Fig. 5B). These results point to similarities in cellular formation and localization between clusters in our in vitro model and granulomas of pulmonary TB infections (40).

Upon analysis of the clusters, CD11c-eYFP–negative cells were harboring BCG E2-Crimson despite only infecting CD11-eYFP⁺ DCs (Fig. 5A). Pulmonary granulomas provide a niche that can be conducive for the localized dissemination of bacteria from one cell to another. To fully understand whether localized dissemination was occurring within cell aggregates on the BBB, PBMCs from CD11c-eYFP and fractalkine CX3CR1-GFP reporter mice were isolated and cocultured with BCG-infected DCs from C57BL/6J wild-type mice on brain endothelial monolayers. BCG bacilli were found in both CD11c-eYFP (Fig. 6A) and CX3CR1-GFP cells (Fig. 6C), indicating that both DCs and macrophages contribute to in situ dissemination of BCG within clusters on brain EC monolayers (Fig. 6B, 6D). These results support local in situ dissemination of mycobacteria within cellular clusters of DCs and PBMCs on the in vitro BBB.

M. tuberculosis–induced pulmonary granulomas, although protective against TB infection, can also considerably contribute to disease pathology by inducing damage to the surrounding tissue (41, 42). Our next aim was to understand the damaging effects of cell clusters on the brain EC monolayer. To study this, inserts were cocultured with uninfected DCs, BCG-infected DCs, and BCG-infected DCs with PBMCs for 24 and 48 h, then fluorescently labeled to identify any cluster-mediated inflammation on the brain EC monolayer.

We examined the effects of cluster formation on barrier integrity by analyzing the morphology of two essential interendothelial junctional proteins supporting physiological BBB function. Fluorescently labeled tight junction protein Cl-5 and tight junction–associated protein zonula occludens-1 (ZO-1) were analyzed for morphological changes using fluorescent confocal microscopy (Fig. 7A). In the absence of infection, both ZO-1 and Cl-5 expression remained intact and localized to the cell borders, whereas infection led to a more perturbed Cl-5 expression, showing small, discontinuous gaps in the staining (Fig. 7A, white arrows) and holes (Fig. 7A, asterisk) on the monolayer, whereas the ZO-1 staining was mildly affected. This morphology was more noticeable in groups containing PBMCs with infected DCs, suggesting that cluster formation likely plays a role in regulating barrier integrity.

Another critical characteristic of inflammation is the expression of adhesion molecules. ECs facilitate leukocyte transmigration to sites of inflammation through increasing adhesion molecule expression, specifically ICAM-1 and VCAM-1, ligands for LFA-1 and VLA-4, respectively (43). Fluorescent Ab labeling revealed a slight increase in both VCAM-1 (Fig. 7B) and ICAM-1 (Fig. 7D) expression on brain EC monolayers with infected DCs only. However, there was a substantial increase in the expression of both adhesion molecules in the presence of PBMCs, which was validated using image analysis and quantifying mean gray value staining intensity of the VCAM-1 and ICAM-1 labeling (Fig. 7C, 7E ). These data support that Mycobacterium infections lead to brain EC inflammation and possible barrier function loss.

To further understand the cellular mechanism of BBB damage as a result of Mycobacterium infection, we visualized cluster-induced cellular stress. Mitochondria play many roles in the cellular process, including contribution to endothelial dysfunction and vascular disease (44). To understand mitochondrial stress in the presence of BCG-induced cluster formation, we stained inserts with MitoTracker dye and evaluated mitochondrial network organization. BCG-infected DCs have disorganized mitochondrial network morphology with more punctate mitochondrial staining both in the cytoplasm of infected DCs (arrow heads) and the underlying EC monolayer (asterisk), an effect that was exacerbated by the addition of PBMCs (Fig. 8A). DCs and the underlying ECs of uninfected samples expressed an evenly distributed and continuous network of mitochondria, suggesting that BCG-infected DC-induced cluster formation on the EC monolayer causes cellular stress in both infected cells and in the underlying ECs as well.

Next, to assess barrier damage directly associated with the clusters over time, BCG-infected CD11c-eYFP DCs were cocultured with wild-type PBMCs and purified P25 transgenic CD4⁺ T cells in the in vitro BBB system for 24 and 48 h. P25 T cells were stained with CellTrace Far Red before assembly. After 24 and 48 h, inserts were fluorescently labeled for ICAM-1 to identify areas of damage in the EC monolayers. Areas that lacked ICAM-1 expression and had direct contact with the cell clusters were termed CAD. Nine and thirteen clusters from 24- and 48-h time points were selected in an unbiased manner, and the endothelial monolayer directly below was examined. CAD was observed at clusters formed both after 24 and 48 h (Fig. 8B). Using ImageJ, a region of interest was drawn around the clusters. Our analysis indicates that clusters are similar in size after 24 and 48 h (Fig. 8C). Similarly, to measure the CAD area, the area of each cluster was subtracted from the region of interest around all areas of damage/decreased ICAM-1 expression that showed an increase in CAD area after 48 h when compared with 24 h (Fig. 8C). CAD appeared to be the result of prolonged cluster formation on the EC monolayer rather than the size of the clusters because CAD area grew independently of cluster size (Fig. 8C). The collection of this data indicates that clusters are capable of inducing increased cellular stress in brain ECs and cause an increase in CAD, which indicates a decrease in barrier function and integrity.

Cytokines play an important role in protection against M. tuberculosis. TNF-α is among the most potent cytokines affording this protection, as TNF-α–deficient mice are quickly overcome by the infection and lack granuloma formation (45). As TNF-α is required for pulmonary granuloma formation, we tested whether TNF-α affects cluster formation on brain ECs.

TNF-α neutralizing Ab MP6-XT22 is frequently used to block the effects of TNF-α during M. tuberculosis infection in mouse studies (34, 45). To understand the mechanism of BCG-infected DC-induced cluster formation on brain ECs, BCG-infected CD11c-eYFP DCs, wild-type–isolated PBMCs, and purified P25 CD4⁺ T cells stained with CellTrace Far Red dye were cocultured in the presence and absence of MP6-XT22 for 24 and 48 h, then analyzed using immunohistochemistry and confocal imaging. Confocal fluorescent microscopy analysis of the brain EC monolayers showed a pronounced absence of clusters in the presence of TNF-α neutralizing Ab at both time points (Fig. 9A). Multiarea mosaic photos of four to six inserts over 24 and 48 h were stitched and tightly compact clusters consisting of ≥five cells were counted, revealing that, on average, there were 150 clusters per brain endothelial monolayer in the absence of TNF-α neutralizing Ab after 24 h, with a modest decrease to slightly over 100 clusters after 48 h (Fig. 9B). Neutralization of TNF-α decreased cluster formation to under 50 per insert over 24 and 48 h, which was a significant decrease compared with inserts containing TNF-α (Fig. 7B). The absence of clusters in the presence of TNF-α neutralizing Ab MP6-XT22 reveals that cluster formation on brain ECs is a TNF-α–dependent process.

Of the various manifestations of the TB infections, CNSTB is the least understood, in part because of the lack of appropriate in vivo or in vitro models. The first observations regarding CNSTB were described by pathologists Rich and McCordock (5) in 1933 during autopsies of M. tuberculosis–infected individuals. It was proposed that M. tuberculosis bacilli travel systemically through the blood and reach the vasculature of the CNS, where it can traverse meningeal vessels. In the CNS, inflammatory lesions (Rich focus) are formed and damage the underlying vasculature, leading to a rupture and spread of bacilli into the perivascular space (5). Recent findings of Brilha et al. (46) showed that tight junction expression decreases in an in vitro BBB model, leading to reduced barrier integrity in the presence of M. tuberculosis–infected monocyte-conditioned medium. In another cell culture BBB system, Jain et al. (7) published that M. tuberculosis can invade and transverse the human in vitro BBB monolayer through actin rearrangement of the brain ECs.

In this study, we present an in vitro BBB Mycobacterium-induced granuloma coculture model that enables the study of the early interactions between Mycobacterium-infected CD11c-eYFP DCs, CX3CR1-GFP macrophages, PBMCs, and Mycobacterium-specific CD4⁺ T cells to understand how these effects can contribute to the dissemination of M. tuberculosis in the CNS. Although it has been proposed that extracellular bacilli travel to the CNS through systemic circulation and traverse the BBB (7), we found that M. tuberculosis–carrying DCs populate the brain parenchyma and provide a carrier for BCG (4), suggesting intracellular Mycobacterium drives dissemination across the BBB. This Trojan horse theory has been implicated as the transport mechanism of other diseases involving neuroinvasion, such as West Nile virus and Cryptococcus neoformans (47, 48). In a mouse model of TB infection, inflammatory DCs can be seen carrying bacilli away from BCG-induced liver granulomas, leading to the dissemination of infection and formation of multiple secondary and tertiary granulomas as infected DCs encounter localized M. tuberculosis–specific CD4 T cells (12).

Human DCs that disseminate M. tuberculosis infection have been shown to express reduced levels of cell surface integrins and the ability to migrate toward a chemokine gradient (13). We have demonstrated that BCG-infected DCs showed a severely impaired migratory capacity that was correlated with two distinct actin profiles associated with infected and uninfected cells. Cells migrate through protruding and retracting forces mediated by actin polymerization at the front (or leading edge) of the cells and depolymerization at the rear of the cell, usually resulting in the polarization of the cell (38, 49–51). During infection with BCG, infected DCs were less likely to protrude and retract simultaneously, making it rare to find BCG-infected DCs that could polarize in a manner that was observed in migrating uninfected DCs. As actin expression in infected DCs was increased toward the center and maintained throughout the cell, this was rarely observed in uninfected migrating DCs, which mostly expressed actin toward the edges. Multiple studies have identified the effect of Mycobacterium on actin polymerization within multiple cell types (7, 52, 53). Bacterial surface protein heparin-binding hemagglutinin binds to G-actin and alters actin dynamics (54), and the M. tuberculosis–secreted tyrosine phosphatase (MptpA) has been shown to affect macrophage phagocytosis through inhibition of actin-mediated processes (55). DCs encounter pathogens, process Ags, and travel to the lymph nodes to prime CD4 T cells and elicit the adaptive immune response (11, 56). Data from our laboratory supports that BCG may at least delay this process, as BCG-infected DCs showed impaired migration toward CCL21 in a CCR7-dependent manner, and CD11c⁺BCG⁺ liver granuloma-derived DCs had a reduction in CCR7 expression (12), a receptor that is used for trafficking to the lymph nodes. Although it is apparent that Mycobacterium infection impairs DC migration, it seems to have an opposite effect on an infected DC’s ability to recruit PBMCs and M. tuberculosis–specific T cells to the site of infection. Although infection limits DC mobility, it increases the ability of DCs to initiate local granuloma-like aggregates on the brain ECs. These aggregates diminish the integrity of the barrier and promote bacterial dissemination into the brain.

Lung granuloma formation is the hallmark immune response to pulmonary M. tuberculosis infection in vivo (57–59). During early lung granuloma formation, bacteria is engulfed by alveolar macrophages that secrete proinflammatory cytokines, leading to the recruitment of monocytes, DCs, T cells, B cells, and neutrophils (41, 58, 60, 61). Typical pulmonary granulomas mainly consist of macrophages located at the center, whereas T cells accumulate and distribute around the periphery (58, 62). As early as 30 minutes into culture assembly, BCG-infected DCs induce cluster formation on the brain endothelial monolayer that increased in quantity and size in the presence of Mycobacterium-specific CD4 T cells and PBMCs. The multicellular clusters morphologically resembled pulmonary TB granulomas, as a tight cluster of cells containing BCG rods with CD11b⁺ macrophages (56%) and CD11c⁺ DCs (23%), whereas CD4⁺ T cells (15%) were found consistently toward the edges in contact with other cells.

Furthermore, in situ bacterial dissemination from CD11b⁺/CD11c⁺ cells to CD11c-eYFP or CX3CR1-GFP cells was observed in the present model, further drawing similarities to pulmonary lesions. Likewise, similar to granuloma formation of pulmonary M. tuberculosis infections, cluster formation on the brain EC monolayer is also dependent on the presence of TNF-α. TNF-α–dependent clusters on the brain endothelial monolayer induce inflammation and impairment of the basic function of the BBB. Clusters increased brain endothelial oxidative stress shown by mitochondrial network disassembly (44), reduction in tight junction protein Cl-5 integrity, and increased expression of adhesion molecules VCAM-1 and ICAM-1. Likewise, examination of brain ECs directly in contact with clusters revealed substantial CAD to the barrier that increased over time, conceivably because of sustained proinflammatory cytokine secretion, as cluster size did not significantly change the damaged area. Lung granulomas are well known to contribute to the pathology of pulmonary M. tuberculosis infections through the sustained secretion of proinflammatory factors, causing damage to the surrounding tissue (5, 63, 64). The nature of our in vitro coculture systems permits a more in-depth investigation into the effects BCG-infected DC-induced cluster formation has on the BBB. Bacterial meningitis has been shown to compromise tight junction integrity, which was also seen in an in vitro model of CNSTB using M. tuberculosis–infected monocyte-conditioned medium (46). ICAM-1 and VCAM-1 expression is elevated on ECs in the presence of inflammatory stimuli to increase adherence and enhance leukocyte transmigration to the inflamed tissue (65), corresponding with our findings and demonstrating BCG-infected DC-induced clusters promote an inflammatory phenotype on the brain endothelial monolayer. The inflammatory characteristics demonstrated in this study present a potentially devastating situation. Clusters comprising M. tuberculosis niche-creating macrophages and DCs promote bacterial growth and localized dissemination directly in the vicinity of barrier integrity loss, resulting in a possible opportunity for DC-mediated dissemination of M. tuberculosis across the barrier.

DCs play a key role in tuberculosis infection through M. tuberculosis Ag presentation in the lymph nodes that induce an M. tuberculosis–specific immune response (66). Continuous Ag sampling by DCs controls disease progression (67). Inflammatory DCs are less capable of engulfing and killing M. tuberculosis compared with macrophages, although both cell types provide a niche that fosters the survival and replication of M. tuberculosis. Our studies show that cell clusters forming on the BBB are TNF-α dependent and contribute to the dissemination of mycobacteria across the barrier by inducing damage to the endothelial layer.

Overall, these findings suggest that Mycobacterium employs a host-mediated mechanism for dissemination into the CNS. Infected DCs are arrested at the BBB and induce cellular cluster formation. The resulting clusters, as increasing numbers of PBMCs arrive, will release proinflammatory cytokines, thus increasing the concentration gradient at a single inflammatory focus. Sustained secretion of these factors results in interendothelial junction protein ZO-1 and Cl-5 dysregulation, upregulation of ICAM-1 and VCAM-1 expression, cellular stress, as well CAD, leading to the loss of barrier integrity. Over time, clusters foster bacterial growth and simultaneously promote barrier damage, thus providing an opportunity for the dissemination of Mycobacterium-harboring immune cells to spread infection across the BBB and into the CNS. DC-induced local lesions on vessels may represent a general mechanism for mycobacterial dissemination.

We thank Satoshi Kinoshita, from the University of Wisconsin Translational Research Initiatives in Pathology laboratory, for assistance. We also thank Dr. Lance Rodenkirch, Managing Director of the University of Wisconsin-Madison Optical Imaging Core for advice in helping with the live-cell imaging experiments. We thank members of our laboratory for helpful discussions and constructive criticisms of this work. We especially thank Khen Macvilay for expertise provided during flow cytometry measurements, Aisha Mergaert for help in dendritic cell culture optimization, Weixuan Chen for assistance in mouse genotyping and image analysis, as well as Bailey Spellman and Kelsey Wigand for cell tracking.

The authors have no financial conflicts of interest.