A Bioengineered Three-Dimensional Cell Culture Platform Integrated with Microfluidics To Address Antimicrobial Resistance in Tuberculosis

ABSTRACT Antimicrobial resistance presents one of the most significant threats to human health, with the emergence of totally drug-resistant organisms. We have combined bioengineering, genetically modified bacteria, longitudinal readouts, and fluidics to develop a transformative platform to address the drug development bottleneck, utilizing Mycobacterium tuberculosis as the model organism. We generated microspheres incorporating virulent reporter bacilli, primary human cells, and an extracellular matrix by using bioelectrospray methodology. Granulomas form within the three-dimensional matrix, and mycobacterial stress genes are upregulated. Pyrazinamide, a vital first-line antibiotic for treating human tuberculosis, kills M. tuberculosis in a three-dimensional culture but not in a standard two-dimensional culture or Middlebrook 7H9 broth, demonstrating that antibiotic sensitivity within microspheres reflects conditions in patients. We then performed pharmacokinetic modeling by combining the microsphere system with a microfluidic plate and demonstrated that we can model the effect of dynamic antibiotic concentrations on mycobacterial killing. The microsphere system is highly tractable, permitting variation of cell content, the extracellular matrix, sphere size, the infectious dose, and the surrounding medium with the potential to address a wide array of human infections and the threat of antimicrobial resistance.

IMPORTANCE Antimicrobial resistance is a major global threat, and an emerging concept is that infection should be studied in the context of host immune cells. Tuberculosis is a chronic infection that kills over a million people every year and is becoming progressively more resistant to antibiotics. Recent major studies of shorter treatment or new vaccination approaches have not been successful, demonstrating that transformative technologies are required to control tuberculosis. We have developed an entirely new system to study the infection of host cells in a three-dimensional matrix by using bioengineering. We showed that antibiotics that work in patients are effective in this microsphere system but not in standard infection systems. We then combined microspheres with microfluidics to model drug concentration changes in patients and demonstrate the effect of increasing antibiotic concentrations on bacterial survival. This system can be widely applied to address the threat of antimicrobial resistance and develop new treatments.
T he progressive emergence of drug-resistant bacteria poses one of the most pressing threats to human health, with the development of totally resistant bacteria potentially leading to a return to the preantibiotic era (1)(2)(3). The pipeline of new antibiotics in development is inadequate to combat the rate of evolution of microbial resistance (4,5). To develop antibiotics, bacteria have traditionally been studied in broth culture, where bacilli are rapidly dividing under optimal growth conditions. However, an emerging concept is that studying pathogens in the context of the host is vital to fully understanding pathogenesis (6,7). Interaction with host cells modulates multiple facets of bacterial physiology and causes stress-induced changes in bacterial gene expression (8). In parallel, evidence is accumulating that host cell biology is modulated by three-dimensional (3D) extracellular matrix interactions, regulating key processes in the host-pathogen interaction such as cell survival, phagolysosomal fusion, autophagy, and cytokine secretion (9,10). In patients being treated for infection, the host-pathogen interaction occurs in three dimensions and antibiotic concentrations vary over time according to drug pharmacokinetics (11). Conversely, the vast majority of in vitro studies are done in the absence of human cells, without an extracellular matrix, and at static antibiotic concentrations.
Considering these concepts together, we concluded that a transformative system to address the threat of antimicrobial resistance requires the following elements: primary host cells infected with fully virulent bacteria, culturing within a 3D structure that incorporates a physiological extracellular matrix, and pharmacokinetic modeling of drug concentrations. These criteria represent a significant challenge in the context of virulent organisms because of the high biosafety containment level required and the complexity of bacteria being eluted under flow conditions. We utilized Mycobacterium tuberculosis, a pathogen that is inherently resistant to antibiotics and causes tuberculosis (TB) (12), to develop a system that addresses these technical obstacles and have recently reported on an investigation of the host immune response in this system (13).
TB is the leading cause of death from an infectious disease worldwide (14), and over the last 2 decades, multidrug-resistant, extensively drug-resistant, and totally drugresistant strains have sequentially emerged, posing the specter of a completely untreatable disease (15). Unfortunately, major recent trials of novel treatment-shortening regimens have not been successful (16), indicating that the model systems that were used to inform these approaches do not sufficiently reflect the disease in humans. Furthermore, pyrazinamide (PZA), one of the most critical antibiotics in human TB treatment, would not have been discovered by current screening approaches. Current models are principally reliant on microbiological broth or solid medium culture, 2D culture, zebrafish, and mice (17). Novel PZA-based regimens show promise (18), and so reliably understanding the action of PZA has become critical, principally focused on mutational analysis of PZA resistance (19). These approaches have limitations, especially in the context of PZA's complex activation, intracellular activity, and uncertain mode of action. For some other drugs, such as cycloserine, nearly all drug susceptibility systems are unreliable. M. tuberculosis is an obligate pathogen of humans and has a prolonged interaction with host cells, centered on adaption to survival within an intracellular niche (20). In addition, the host-pathogen interaction is spatially organized (21) and the extracellular matrix influences host cell survival (22), suggesting that a fully humanized system structured in three dimensions with an extracellular matrix is needed to identify novel treatments for TB. Therefore, we developed a platform utilizing M. tuberculosis as the prototype organism. Our system integrated genetically modified virulent reporter bacilli, primary human cells, and a human extracellular matrix by using a bioengineering approach and combined this with a multiparameter longitudinal readout. Within this microsphere system, we demonstrate cellular aggregation and upregulation of mycobacterial stress genes. Critically, PZA is efficacious in the 3D microsphere system but not in standard broth or 2D culture. We then combined microspheres with a microfluidic system to permit pharmacokinetic modeling. We observed more rapid M. tuberculosis killing with higher peak antibiotic concentrations, similar to outcomes in patients with TB (23). Therefore, this system models conditions in patients and can be readily applied to a range of drug-resistant organisms to address the global challenge of antimicrobial resistance.

RESULTS
Granulomas develop within microspheres, and M. tuberculosis stress genes are upregulated. We incorporated primary human cells, virulent M. tuberculosis, and type I collagen into 3D microspheres by bioelectrospray methodology (see Movie S1 in the supplemental material). Fluorescent staining of monocytes and T cells, followed by infection with mCherry-expressing M. tuberculosis, showed the distribution of cells and bacteria through the microspheres and early granuloma formation from day 4 (Fig. 1A). After 14 days of infection, large cellular aggregates resembling human granulomas Bioengineering To Combat Antimicrobial Resistance ® developed, while no aggregates formed in uninfected microspheres (Fig. 1B). Granuloma formation was associated with evidence of mycobacterial stress. Multiple stressrelated genes were upregulated at day 14 in comparison with M. tuberculosis in 7H9 broth culture analyzed by reverse transcription-quantitative PCR (RT-qPCR) (Fig. 1C), including lipF, the acid stress real-time response gene; recA encoding recombinase A, the key mediator of the SOS response to DNA damage; relA, the nutrient stress-related gene; and sodA, the oxidative-stress response gene. By infecting cells with genetically modified luminescent M. tuberculosis expressing the Lux operon (24), bacterial growth could be monitored longitudinally over time within microspheres in a nondestructive manner (Fig. 1D).
To determine the localization of mycobacteria, we studied microspheres longitudinally. We compared extracellular with cell-associated bacteria by decapsulating microspheres and performing differential centrifugation to separate extracellular mycobacteria from intracellular and cell-adherent mycobacteria. The proportion of cell-associated mycobacteria analyzed by luminescence or CFU counting progressively increased over time ( Fig. 2A and B), leading to a 20.7-fold increase in luminescence of cell-associated M. tuberculosis relative to extracellular M. tuberculosis within microspheres at day 15. Gentamicin treatment in a single experiment demonstrated that cell-associated mycobacteria were almost all intracellular, with no significant reduction in CFU counts after the killing of extracellular bacteria. Similarly, the fluorescence of cells infected with green fluorescent protein (GFP)-expressing M. tuberculosis progressively increased over time ( Fig. 2C; Fig. S1), demonstrating intracellular proliferation. M. tuberculosis infection did not increase cellular toxicity when measured by the CytoTox Glo 3D assay, with no significant difference in viability between the two conditions ( Fig. 2D).
Standard antibiotics kill M. tuberculosis under all conditions. Having demonstrated granuloma formation and the M. tuberculosis stress response within microspheres, we first studied standard first-line antibiotics in 2D cell culture and the 3D model to determine the tractability of the model and whether killing efficacy is the same under both conditions. Rifampin, isoniazid (INH), and ethambutol were added to cell culture medium around spheres at physiological concentrations (1, 0.25, and 4 g/ml, respectively). All three antibiotics inhibited M. tuberculosis growth in both 2D and 3D cell culture systems ( Fig. 3A and B, antibiotics added at day 6; Fig. S2, antibiotics added at day 1). Rifampin was the most efficacious at killing M. tuberculosis, and INH was consistently more efficient at controlling M. tuberculosis in the 3D microsphere system than in 2D cell cultures. M. tuberculosis growth analyzed by luminescence correlated closely with CFU counts on Middlebrook 7H11 agar (Fig. 3C).
PZA is only efficacious in microspheres and not in broth or 2D cell cultures. Next, we investigated PZA, which is a key antibiotic in treating human disease but has a poorly defined mechanism of action at the concentration described in epithelial cell lining fluid (25). PZA had no effect on M. tuberculosis growth in 7H9 broth without cells at neutral pH (Fig. 4A). In 2D primary cell cultures, PZA had a temporary effect but M. tuberculosis growth rapidly recovered (Fig. 4B). Critically, PZA killed M. tuberculosis in the 3D microsphere system, with luminescence falling to background levels by day 30 when we used the same antibiotic preparation that had no effect in broth and a transient effect in 2D cultures (Fig. 4C). A similar pattern of efficacy was observed when PZA was added to cultures on day 1, with PZA having no effect in 7H9 broth and a temporary effect in 2D cultures but complete control of M. tuberculosis growth in microspheres (Fig. S3). Colony counting on 7H11 agar confirmed that the efficacy of M. tuberculosis killing by PZA was equivalent to that of INH and moxifloxacin ( Fig. 4D;  Fig. S2D).
Second-line antibiotics are most efficacious in 3D microspheres. We then examined the effect of the second-line antibiotics D-cycloserine, moxifloxacin, and linezolid, which are of increasing importance with the emergence of drug-resistant TB. In 7H9 broth, D-cycloserine at a low concentration had a minor inhibitory effect but at a high concentration it was as effective as moxifloxacin (Fig. 5A). In the 2D and 3D systems, D-cycloserine effectively killed M. tuberculosis at low and high concentrations ( Fig. 5B and C). Linezolid and moxifloxacin effectively suppressed M. tuberculosis growth under all three conditions. INH, which was included as a control first-line antibiotic, consistently killed M. tuberculosis in 3D microspheres, but bacterial growth resumed in 7H9 broth and the 2D system. A similar pattern was observed when antibiotics were added at day 1, although the inhibition of M. tuberculosis growth by D-cycloserine was more rapid in microspheres than in 2D cultures (Fig. S4). To ensure that the antibiotics were not having a cytotoxic effect on host cells, we analyzed their viability and found no evidence of cytotoxicity after 21 days of culture compared to medium and dimethyl sulfoxide (DMSO) controls (Fig. S5). Pharmacokinetic modeling by integration with microfluidics. In patients, antibiotic concentrations fluctuate over time, as opposed to the static concentrations usually studied in the laboratory. Therefore, we integrated the microsphere system with a microfluidic platform to permit modulation of antibiotic concentrations over time to mimic in vivo pharmacokinetics in patients during treatment (Fig. 6A). We studied rifampin, as its concentrations in plasma correlate with treatment outcomes (23). We manufactured a microfluidic plate from milled poly(methyl methacrylate) (PMMA), providing each well with two inlets and one outlet, permitting a smooth flow of medium through the wells containing microspheres (Fig. 6B). Initially, bacterial luminescence from the 24-well plate was undetectable on a GloMax DISCover plate reader.
To overcome this, we used phenol red-free medium, optimized the microsphere density within wells, and placed a custom-made mirror under the plate. These modifications greatly improved the luminescence readout ( Fig. 6C and D), and we were able to monitor bacterial growth from experimental time zero. Bioengineering To Combat Antimicrobial Resistance ® To model pharmacokinetics in patients, rifampin at a range of concentrations (0.25, 1, and 4 g/ml) was used to irrigated wells from day 5, incubated for 6 h, and then washed out to leave a minimal antibiotic concentration overnight. A stepwise increase in the rifampin concentration in individual wells produced dose-dependent killing of M. tuberculosis (Fig. 6E). The highest 6-h peak rifampin concentration caused M. tuberculosis killing equal to that of a constant 1 g/ml antibiotic concentration or that of irrigation with medium with rifampin at 1 g/ml (Fig. S6A). Luminescence increased overnight in the absence of antibiotics, resulting in a sawtooth pattern of killing that did not occur at a fixed antibiotic concentration and demonstrated rapid recovery of mycobacterial growth once antibiotic pressure was removed. Colony counts on 7H11 agar confirmed that the luminescence data reflected the total bacterial load (Fig. S6B).

DISCUSSION
Antimicrobial resistance is rapidly emerging as one of the most pressing challenges to global society (2,3). Innovative approaches to studying the host-pathogen interaction are urgently needed to identify novel treatment approaches that counter the evolution of drug-resistant bacteria. To achieve this, an emerging paradigm is that bacteria should be studied in the context of the host (6). We hypothesized that a transformative platform will require multiple elements that are not currently available in a single system: virulent reporter bacteria, primary human cells, an extracellular matrix, 3D organization of the host-pathogen interaction, and pharmacokinetic modeling. We combined diverse methodologies, including genetically modified bacteria, primary human cell culture, electrostatic microsphere generation, multiparameter readouts, and microfluidics, to develop a platform with all of the requisite elements that could be used in a biosafety level 3 containment environment. In our investigation of the host immune response, we showed that host cell survival is improved in the 3D microsphere model, cellular aggregates form, and host-directed therapies can be studied (13). Here, we investigated the system from the pathogen's perspective and demonstrated that mycobacterial stress genes are upregulated and key antibiotics used to treat human disease are more efficacious in the microsphere system than in standard culture. The nondestructive readouts permit longitudinal analysis over prolonged periods and real-time pharmacokinetic modeling. This platform has the potential to revolutionize antibiotic discovery and to replace suboptimal animal model systems based on inappropriate host-pathogen combinations. Bioengineering To Combat Antimicrobial Resistance ® Miniaturized organoid systems and lab-on-a-chip technologies are rapidly evolving fields for drug discovery, resulting from the widespread belief that studying individual cells or infectious organisms in isolation does not sufficiently reflect conditions in vivo (26,27). The bioelectrosprayer can produce large numbers of identical microspheres rapidly in a patient-relevant model, fulfilling a principal requirement of such a system (28). 2D systems with flow have been developed to study respiratory epithelial cell infection (29), and while several TB cellular systems have been reported (30)(31)(32)(33)(34), none combine 3D culture, collagen, and the potential for high-throughput or longitudinal analysis. The microsphere system is highly tractable, with the ability to modulate infectious organism and doses, host cellular content, the extracellular matrix, microsphere size, and the surrounding medium dynamically.
The need for new antibiotics for TB is particularly acute, given the progressive emergence of drug resistance, recent disappointments in treatment-shortening regimens, and the ongoing global toll of infection (14). PZA is one of the most important antibiotics used to treat human disease, but it was discovered serendipitously in the 1950s and indeed would have not been found by current screening approaches based on the sequential study of MICs in broth culture, murine infections, and human disease (17). PZA has a complex mechanism of action, requiring intracellular acidification. We demonstrated that M. tuberculosis stress genes were upregulated in microspheres and that PZA killed M. tuberculosis in microspheres at neutral pH but not in standard 2D culture systems at a concentration found in epithelial lining fluid (25). This confirms that, within microspheres, bacilli are in a PZA-sensitive compartment and demonstrates the potential to identify other compounds only active against stressed mycobacteria in the correct microenvironment. Similarly, we found that INH was consistently more effective in the 3D system that in the 2D system, with a more rapid fall in luminescence and no failed treatments, supporting the relevance of the model to human TB. Therefore, the system can identify drug resistance in a more clinically relevant fashion and can be used to study novel regimens with variable concentrations in combination rather than single new agents at static concentrations.
The microsphere system has potential for high-throughput use, as Ͼ5,000 microspheres can be generated from a single blood donor within an hour and the diameter is compatible with a 384-well format. Mycobacterial killing curves during antibiotic treatment of patients suggest that there are diverse populations within patients' lungs (35), and therefore, the system can be used to study the separate physiological conditions that drive these within the same experiment, such as hypoxia and nutritional stress. Encapsulation within microspheres permits integration with a microfluidic system without cells and bacteria being lost during irrigation, resolving a significant technical hurdle for pharmacokinetic studies. We were able to show more rapid killing with increased rifampin concentrations, consistent with findings in patients (23,36). Microfluidics have been used to develop an array of organ-on-a-chip models (37), but we are not aware of development to study virulent containment level 3 pathogens such as M. tuberculosis, which represents additional challenges because of the infection risk.
The hollow-fiber model has been used to perform advanced pharmacokinetic modeling, but PZA is only efficacious in this system after acidification to pH 5.8 (38) and so cannot be studied in combination with other agents. Future potential developments for the bioelectrospray platform include dual encapsulation to permit a central lipid-rich caseous core, generating an additional layer of flexibility and modeling drug penetration into necrotic foci (39). Development of the microfluidic plate will permit optimization of combinations of multiple antibiotics in a fully humanized system, with pharmacokinetic modeling of each antibiotic within wells, to identify the best combinations with which to go forward to clinical trials (34).
We developed a bioengineered cell culture platform that replicates key features of human disease and incorporates primary human cells, an extracellular matrix, a 3D structure, virulent bacteria, and pharmacokinetic modeling. The microsphere system is highly tractable, permitting variation of cell content, the extracellular matrix, sphere size, the infectious dose, and the surrounding medium with the potential to address a wide array of human infections. The system can equally be applied to diverse inflammatory and malignant human diseases. Integration with molecular microbiology techniques and clustered regularly interspaced short palindromic repeat gene editing will provide genetically tractable host-pathogen pairings. Therefore, this platform has global applicability to address the threat of antimicrobial resistance and deliver new treatments.
Human PBMC isolation and infection. Ethical approval for these studies was provided by the National Research Ethics Service Committee South Central-Southampton A, ref. 13/SC/0043. Peripheral blood mononuclear cells (PBMCs) were isolated from single-donor buffy coats from the National Health Service Blood and Transplant, Southampton, United Kingdom. Leukocytes were isolated by density gradient centrifugation over Ficoll-Paque (GE Healthcare Life Sciences, United Kingdom). Isolated PBMCs were infected with M. tuberculosis Lux at a multiplicity of infection (MOI) of 0.1 and kept overnight at 37°C in a 5% CO 2 incubator in RPMI 1640 medium supplemented with 10 g/ml ampicillin, 2 mM glutamine, 25 g/ml kanamycin, and 10% fetal bovine serum (Labtech International Ltd.). The next day, infected PBMCs were transferred from vented flasks to 50-ml Falcon tubes after detachment with Versene solution (Sigma) for 10 min and scraping. After Hanks balanced salt solution (HBSS) without Ca/Mg (Gibco) was added, cells were centrifuged at 320 ϫ g for 8 min at 4°C and the supernatant was decanted. The pelleted cells were resuspended in appropriate volumes of RPMI 1640 medium supplemented with 10 g/ml ampicillin, 2 mM glutamine, 25 g/ml kanamycin, and 10% human AB serum (Sigma), referred to as complete RPMI medium.
2D culture. Infected cells were resuspended in 50 ml of complete RPMI medium, and 1 ml was equally distributed into 2-ml Eppendorf tubes at a final concentration of 3 ϫ 10 6 /ml. Cultures were incubated at 37°C in 5% CO 2 . M. tuberculosis luminescence was monitored with a GloMax 20/20 luminometer. Antibiotics were added at predetermined time points. For colony counts, cultures were treated with 1% saponin in HBSS and bacteria were plated onto 7H11 agar at serial dilutions. For RT-qPCR analysis, infected cells were plated into six-well plates at a final concentration of 2.5 ϫ 10 6 /ml. 3D culture. Infected cells were resuspended in complete RPMI medium, mixed with sterile alginatecollagen at 1 ϫ 10 6 cells/ml, and injected into an Electrostatic Bead Generator (Nisco, Zurich, Switzerland) to form microspheres via a Harvard syringe driver as described previously (41). After generation, microspheres were equally distributed into 2-ml Eppendorf tubes (microsphere volume, 0.4 ml), immersed in 1 ml of complete RPMI medium, and incubated at 37°C in 5% CO 2 . M. tuberculosis luminescence was monitored with a GloMax 20/20 luminometer. For CFU counts, microspheres were dissolved in 55 mM sodium citrate-10 mM EDTA with 1% saponin in HBSS and bacteria were plated onto 7H11 agar. For RT-qPCR analysis, microspheres were cultured in 50-ml Falcon tubes (microsphere volume, 10 ml) in complete RPMI medium.
Immunofluorescence and confocal imaging. PBMCs were separated into monocytes and lymphocytes with MACS Cell Separation Columns (Miltenyi Biotec, Inc., Surrey, United Kingdom). Cells were then labeled with CellTracker Blue or CellTrace CFSE (Thermo, Fisher Scientific, United Kingdom) separately in accordance with the manufacturer's recommendations before infection with mCherry-expressing M. tuberculosis H37Rv (40) at an MOI of 0.1. Microspheres were generated and fixed in 4% paraformaldehyde after 4 days. Confocal images were acquired on a Leica TCS SP5 confocal microscope and processed with ImageJ 1.5 0d (NIH, Bethesda, MD).
Transcription analysis by RT-qPCR. For bacteria grown in 7H9 broth (OD 600 of 0.25) and 2D cultures, total RNA was extracted by centrifugation at 13,000 rpm for 10 min and the addition of 500 l of RNAprotect Bacteria Reagent (Qiagen). The resuspended pellet was left for 10 min at room temperature prior to repeat centrifugation and resuspension of the pellet in 1 ml of TRIzol (Life Technologies, Inc.) and stored at Ϫ80°C. For 3D cultures, RNAlater solution (Ambion) was used to preserve RNA overnight at 4°C. Cells were decapsulated with 100 mM sodium citrate and centrifuged at 3,000 ϫ g for 30 min, and the pellet was resuspended in 1 ml of TRIzol (Life Technologies, Inc.) and stored at Ϫ80°C until use. Thawed samples were transferred to Lysis Matrix B tubes containing 0.1-mm silica beads (Q-Biogene) and homogenized in a MagnaLyser instrument (Roche) at 4,000 rpm for 5 ϫ 45 s with incubation on ice for 1 min between homogenizations. Samples were centrifuged for 1 min at 16,100 ϫ g at 4°C, and the supernatant was transferred to a new Eppendorf tube. After phenol-chloroform extraction, the nucleic acids were precipitated with isopropanol, washed with 75% ethanol, air dried for 10 to 15 min, and finally resuspended in nuclease-free water (Fisher Scientific). Genomic DNA was removed with a DNA-free kit (AM1906; Ambion) in accordance with the manufacturer's instructions. RNA was further purified with the Qiagen RNeasy minikit (Qiagen), subjected with on-column DNase digestion with the RNase-free DNase set (79254; Qiagen), repurified with an RNeasy minikit, and eluted in 50 l of RNase-and DNase-free water (Fisher Scientific). The first-strand cDNA was synthesized in 10-l reaction Statistical analyses. Statistical analyses were preformed with GraphPad Prism. Differences were considered significant at P Ͻ 0.05.