A High-Throughput Screen for Tuberculosis Progression

One-third of the world population is infected with Mycobacterium tuberculosis and multi-drug resistant strains are rapidly evolving. The noticeable absence of a whole organism high-throughput screening system for studying the progression of tuberculosis is fast becoming the bottleneck in tuberculosis research. We successfully developed such a system using the zebrafish Mycobacterium marinum infection model, which is a well-characterized model for tuberculosis progression with biomedical significance, mimicking hallmarks of human tuberculosis pathology. Importantly, we demonstrate the suitability of our system to directly study M. tuberculosis, showing for the first time that the human pathogen can propagate in this vertebrate model, resulting in similar early disease symptoms to those observed upon M. marinum infection. Our system is capable of screening for disease progression via robotic yolk injection of early embryos and visual flow screening of late-stage larvae. We also show that this system can reliably recapitulate the standard caudal vein injection method with a throughput level of 2,000 embryos per hour. We additionally demonstrate the possibility of studying signal transduction leading to disease progression using reverse genetics at high-throughput levels. Importantly, we use reference compounds to validate our system in the testing of molecules that prevent tuberculosis progression, making it highly suited for investigating novel anti-tuberculosis compounds in vivo.


Introduction
Tuberculosis (TB) is an ancient chronic disease caused by M. tuberculosis. With one-third of the world population infected, the predominant outcome is a latent and persistent infection controlled by type I immune responses [1,2,3,4]. An important characteristic of this infection is the formation of granulomatous lesions, consisting of clusters of infected macrophages and other immune cells [5,6]. Paradoxically, the main purpose of the host macrophages, which M. tuberculosis infects and where it persists, is to clear bacterial infection [7,8]. M. tuberculosis achieves persistent infection through rapid changes in its gene expression profile in order to counteract host cell biological and immune processes, such as antigen presentation, pro-inflammatory cytokine secretion and phagosome maturation [9].
The alarming rate of emergence of new drug resistant (MDR/ XDR) M. tuberculosis strains isolated from patients, in particular HIV-infected individuals, is cause for global concern, and the race for more efficient vaccines, as well as novel antibiotics targeting either the pathogen or the host, has begun [3,10,11]. While in vitro models have shed light on processes that are central to the uptake and survival of the bacterium, they cannot recapitulate the full phenotype of latent M. tuberculosis infection. This has been partly circumvented through the use of non-human primate models, which develop a form of TB that exhibits many of the hallmarks of the human infection [12]. Other in vivo systems include the guinea pig model, used to validate anti-TB vaccines and drugs [13], and mouse models, which offer extensive arrays of genetic tools. However, neither rodent model fully recapitulates essential aspects of TB lesion progression in man, including granuloma formation and maturation [10,14].
The low-cost and high clutch-size zebrafish (D. rerio) is, at the embryonal and larval stages, optically transparent, permitting visualization of pathogens and lesions in real time [15], as well as offering exciting possibilities for high-throughput imaging [16]. Zebrafish are also amenable to forward genetic screening, or reverse genetics techniques such as injection of morpholinos (inhibitory of mRNA translation) [17,18]. An ectotherm, the zebrafish is one of the natural hosts of M. marinum, the closest relative of the M. tuberculosis complex [19]. Of crucial relevance, as shown by the pioneering work of the Ramakrishnan group, M. marinum infection of zebrafish closely mimics the mammalian TB pattern of infection, both in terms of bacterial numbers which increase rapidly in early infection, and of the formation of caseous granulomas that present characteristics typical of their human counterparts [20,21,22,23].
The indirect study of human TB via the infection of the zebrafish embryo with M. marinum has already led to the clarification of many important processes in the life cycle of the infection, in particular those underlying the mechanisms of granuloma formation [22,23,24,25,26,27]. The importance of studying mycobacterial infections at a whole organism level was highlighted in the report that induction of mmp9 expression, enhancing macrophage recruitment to granulomas, was localized to epithelial cells near infected macrophages [26]. Another example of the use of zebrafish larvae to uncover a host-pathogen interaction relevant to human mycobacterial infection is the recent forward genetic screen by Tobin and Ramakrishnan, who mapped a hypersusceptibility mutation to the leukotriene biosynthesis gene, lta4h, and showed that heterozygosity at the LTA4H locus correlated with susceptibility of human populations to both TB and leprosy [28]. It is therefore clear that the zebrafish mycobacterial infection model is quickly becoming an attractive and advantageous alternative for analyzing granuloma and disease progression in vivo.
The common route of infecting zebrafish embryos with M. marinum is the injection of the pathogen into the caudal vein of the 1 day old embryo [23]. This method is labour-intensive and generally considered to be a low-throughput technique, leading to major bottlenecks in drug discovery, particularly in times of highthroughput technology. Since infection by immersion is not an effective alternative, we sought to achieve a reliable highthroughput automatic injection system, drastically reducing the man-hour requirement while vastly increasing the number of reproducibly infected embryos. Large quantities of similarlyinjected/infected embryos would then allow testing of sizeable drug libraries for anti-bacterial activity targeting either the pathogen or the host itself.
Here we show that the automatic injector we developed provides a powerful and reliable high-throughput system for infecting embryos with M. marinum. We also show that we can couple the injector to a flow cytometer capable of sorting live multicellular organisms (Complex Object Parametric Analyzer and Sorter, COPAS) and rapidly test the efficacy of known anti-TB drugs in infected embryos. Finally, and importantly, we demonstrate that this system is ideally suited to test proliferation and tissue spreading of the human pathogen, M. tuberculosis.

Results and Discussion
Proof of principle of the yolk sac as an early-stage embryo injection site We first demonstrated that the injection of 20-40 M. marinum colony-forming units (CFUs) into the yolk sac of embryos at several early developmental stages (up to 1,024-cell stage) precisely mimics the infection obtained with the well-established caudal vein injection method. In our set-up, all injections were performed using polyvinylpyrrolidone as a polymer-based carrier for the bacteria, which showed several benefits: (1) restriction of early bacterial spread into the embryo, precluding developmental problems arising from the early injection stage; (2) higher concentration homogeneity; (3) clear visibility of injected inoculum as a spheroid (Video S1). Besides extensive bacterial growth within the yolk, we witnessed frequent formation of aggregates of infected cells outside the site of injection, namely in the head, body and tail of the larvae at 5 days post-infection (dpi) ( Figure 1A). These aggregates were highly similar to those previously shown to represent initial stages of granuloma development [23]. No adverse developmental effects were seen in any of the conditions tested. Confirmation that yolk M. marinum injection resulted in granuloma formation was obtained through GFP-labelled gag (granuloma-activated gene) [23] activation at 7 dpi using an M. marinum strain also expressing mCherry ( Figure 1B-E). Additionally, immunohistochemistry using L-plastin showed clear colocalization of M. marinum and leukocytes ( Figure 1F-I). To functionally analyze the role of the immune system in the spreading and proliferation of mycobacteria after yolk injection, we co-injected a morpholino targeting pu.1 [29] and M. marinum at the 1-2 cell stage. The results revealed the presence of extracellular M. marinum and increased bacterial proliferation in Pu.1 morphants (Figure 2A and B), consistent with previous data demonstrating that macrophages in zebrafish embryos restrict mycobacterial growth [25]. At 2 dpi, we observed cording structures in Pu.1 morphants, characteristic of extracellular mycobacteria [27,28], in the tail region of infected embryos ( Figure 2C-E). Using a mag49 (macrophage-activated gene)-GFP [23] construct in mCherry-labelled bacteria, we were able to confirm their extracellular location through the lack of mag49-GFP expression, previously shown to be active only after phagocytosis by macrophages [23] ( Figure 2C and D).

High-throughput M. Marinum injection and drug screen
We subsequently developed an automatic injector system around the yolk injection concept (Figure 3 and 4). All tests performed demonstrated that this injector design, capable of 1,024 consecutive injections per run of 30 minutes, reproducibly reached a success rate of over 99% (sample in Video S1) and produced identical results to manual yolk injections of embryos. Importantly, embryos occupied the hemi-spherical wells of the agarose cast ( Figure 4B and S1) in a centred and completely reproducible manner, with the cell mass always resting to the side ( Figure 4C). No image recognition was thus required for the injections, unlike a previously reported design that operates at a throughput level of 25 consecutive injections per run of 2 minutes [30]. The choice of agarose as the casting material dramatically reduced light refraction, resulting in a better image during calibration, and helped maintain embryos humid and viable.
To demonstrate the applicability of our system to drug screens, two independent large sets of embryos were injected with M. marinum and treated with a combination of first-line anti-TB drugs (Rifampicin and Isoniazid). After 3 days, immediately prior to the start of the antibiotic treatment, embryos were run through the COPAS flow cytometry system to determine the total level of red fluorescence, representative of bacterial load. The embryos were subsequently split into two random groups, whereby one was subjected to the combinatorial treatment for 3 days. At 5 dpi, the average signal per larva in the untreated group was approximately 3-fold higher than that of the treated group, and this difference was even more pronounced (4-fold) at 6 dpi ( Figure 5A and Table S1). These results attest both to the efficacy of the combinatorial drug treatment and to the ability of COPAS to correctly discriminate treated and untreated groups.
Epifluorescence and bright-field imaging revealed little to no red signal outside the yolk region of the treated larvae, which looked healthy and phenotypically normal. By contrast, untreated embryos displayed varying bacterial loads in the head, body and tail regions ( Figure 5B, C, E and F). Additionally, the individual profiles generated by COPAS correctly indicated whether bacteria were present in the body of infected larva ( Figure 5D and G). Lplastin immunostaining further confirmed co-localization of M. marinum and leukocytes in the body of the untreated larvae ( Figure 1F-I).

High-throughput M. tuberculosis injection and drug screen
It is clear that much can be learned about TB from the study of M. marinum infections in zebrafish, and the use of this pathogen offers practical advantages when compared to M. tuberculosis, such as lower biosafety restrictions and faster growth rate. That notwithstanding, it was of interest to study the human pathogen, M. tuberculosis, directly in zebrafish. Using our system, we overcame all technical difficulties of manually injecting a BSL-3 pathogen into zebrafish embryos. Two independent sets of embryos were injected with M. tuberculosis, and at 3 dpi were split into treated (combinatorial Rifampicin and Isoniazid treatment) and untreated groups. To support growth of M. tuberculosis, embryos were maintained at a higher temperature (34uC) than in M. marinum infections (28uC).
Confocal imaging of fixed infected larvae revealed the presence of M. tuberculosis in their bodies after 5 dpi, indicating that the bacteria survived and were transported outside the injected area by macrophages, and that zebrafish larvae survive exposure to this pathogen ( Figure 6A). There was a highly significant correlation (p = 0.0004) between M. tuberculosis presence in the larvae and the absence of treatment ( Figure S2). Supporting the survival of M. tuberculosis in zebrafish, plating of lysates from 5 and 6 dpi untreated larvae resulted in growth of M. tuberculosis colonies. Noteworthy, treated larvae did not yield any colonies, implying that the bacteria were eliminated during treatment.
L-plastin immunostaining showed leukocytes clustering around infected regions throughout the body, suggesting the formation of granuloma-like aggregates similar to those observed in M. marinum infections ( Figure 6B-D). Leukocytes in these aggregates showed intracellular fluorescence of M. tuberculosis bacteria ( Figure 6C and D). In addition, we also observed bacterial accumulation in cord-like structures characteristic of extracellular growth [27,28].

Concluding remarks
Our work has shown that the automatic injector, coupled with COPAS analysis and sorting, provides an extremely powerful high-throughput pipeline for infecting and analyzing zebrafish embryos, and offers a new in vivo tool for rapidly testing the efficacy of large panels of molecules on the propagation of the pathogen studied. Gene-disruption tools, such as morpholinos, can be easily integrated into our set-up. Moreover, our results clearly demonstrate, for the first time, the potential of using fish larvae to investigate M. tuberculosis directly, and highlight the importance of the automatic injector in enabling a high biosafety-level study that would otherwise be technically extremely difficult to accomplish. Interestingly, we have recently demonstrated the applicability of

Automated injection system
A polycarbonate substrate featuring a honeycomb pattern of 1,024 hemi-spherical wells (1.3 mm diameter) was used to create a negative mould in flexible polydimethylsyloxane (PDMS, Sylgard 184, Dow Corning) using standard moulding techniques. A 1% agarose gel (Sphaero) was poured onto an agarose-coated glass plate and the PDMS mould was pressed to touch the glass. After gelling the mould was removed and the grid was placed in a leakage-free steel support (sized to a 96-well plate) ( Figure 4B).
The embryo grid was placed in a motorized stage (MTmot 2006100 MR, Mä rzhä user) connected to a controller (Tango, Mä rzhä user).
A motorized micro-manipulator (Injectman II, Eppendorf) was adjusted to a vertical position above the stage, and connected to a pump (Femtojet Express, Eppendorf) featuring an external compressor (lubricated compressor, model 3-4, JUN-AIR).
A firewire camera (DFK41BF02.H, The Imaging Source) equipped with a 46 macro lens (MR4/O, The Imaging Source) was placed beneath the stage for imaging.
All components were connected to the controlling computer (Ubuntu AMD64). A multi-threaded control program was written in  Python, using PySerial and wxPython. Coriander software (http:// damien.douxchamps.net/ieee1394/coriander) was used for imaging.
The camera height was adjusted to focus on the top plane of the agarose grid, and a grid calibration was performed.
The grid was removed for loading with embryos ( Figure 4C). The injection needle (pulled borosilicate glass capillary, Harvard Apparatus) was placed in the Injectman and moved to the central focal position. The x and y coordinates were stored and the needle was elevated to replace the filled grid.
The injection height was calibrated using the first embryo by moving the needle downwards through the chorion until touching the yolk (400 mm above injection point).
M. tuberculosis strain H37Rv stably expressing mCherry was maintained in logarithmic phase at all times in 7H9 medium (BD Difco Middlebrook) containing 50 mg/ml hygromycin in a BSL-3 laboratory. Prior to injection, optical density at 600 nm was assessed, the bacteria were washed three times in sterile water and resuspended in 2% PVP40.
The number of CFU in each inoculum was verified by plating out serial dilutions and the injected inoculum in triplicate.

Ethics statement
Zebrafish lines (wild-type, albino/fli1-egfp [33] and mpx-gfp [34]) were handled in compliance with the local animal welfare regulations and maintained according to standard protocols (zfin.org). The breeding of adult fish was approved by the local animal welfare committee (DEC) of the University of Leiden. All protocols adhered to the international guidelines specified by the EU Animal Protection Directive 86/609/EEC.

Zebrafish infections
Infections including the Pu.1 morpholino [29] (1 mM, Gene Tools) were performed by yolk injection (1 nl) at the 1-2 cell stage, whereas all other injections of M. marinum  or M. tuberculosis (100 CFUs) took place between 16 and 512 cells. Control embryos were injected with carrier solution.
After M. marinum infection, embryos were collected in 92616 mm petri dishes (Sarstedt), with a maximum of 100 embryos per dish, and maintained at 28uC in egg water. At 3 dpi embryos were analyzed by the COPAS system (see below) and randomly split into two equal groups. One group was exposed to a combination of 200 mM Rifampicin (Sigma-Aldrich) and 2 mM Isoniazid (Sigma-Aldrich) for 3 days (exposure to the drugs achieved by adding compounds to egg water; antibiotics refreshed once daily) and the other was followed without treatment (water refreshed once daily). Uninjected controls were similarly split into treated and untreated groups to account for antibiotic effects. At 5 and 6 dpi, the different larva groups were analyzed by the COPAS system, and the bacterial load was assessed by the total red fluorescence detected.
After M. tuberculosis infection, embryos were collected in tanks containing 1 litre of egg water with a maximum of 300 embryos per tank, and maintained at 34uC. At 3 dpi embryos were randomly split into treated and untreated groups as described above. Twenty larvae per group were homogenized at 5 and 6 dpi and plated out to assess the number of live bacteria per larva. Batches of 40-100 larvae per group were fixed at 5 and 6 dpi for optical analyses.

Immunohistochemistry
Larvae were fixed in 4% paraformaldehyde in PBS overnight at 4uC and immunolabeled using the L-plastin antibody as previously described [35].

Complex Object Parametric Analyzer and Sorter (COPAS)
The COPAS TM XL (Union Biometrica) large particle sorter has been designed for the analysis, sorting and dispensing of objects up to 1.5 mm in diameter based on size, optical density and fluorescence intensity. It is equipped with 488 nm and 561 nm Solid State lasers, and mCherry is detected through a 615/24 Band-Pass filter.
The Profiler II Option simultaneously detects and analyzes up to 8,000 data points per object for each of the channels of extinction and fluorescence, and includes advanced imaging options. The resulting profiles can be used to set sorting parameters.

Statistics
The effect of drug treatment on M. marinum mCherry fluorescence in zebrafish larvae was statistically analyzed using a 2-tailed T-test. The correlation between drug treatment and the presence or absence of M. tuberculosis was determined by a x 2 -test. Figure S1 Embryo holder designs. Possible designs of the agarose embryo holder, demonstrating that the hemi-spherical alternative provides the largest injection target volume, taking embryo size variability into account (indicated by arrows). Three embryo sizes are depicted, where the embryo designated by ''100'' is of average size. (TIF) Figure S2 Combinatorial treatment of M. tuberculosisinfected larvae. Effect of combinatorial treatment on presence or absence of bacteria in 6 day-old larvae from two independent experiments. (TIF)

Supporting Information
Table S1 Difference in mCherry-labelled M. marinum load between treated and untreated groups. Both experiments indicate a highly significant reduction of bacterial numbers (measured by COPAS) as a direct result of treatment with first-line anti-TB drugs, as determined by a 2-tailed T-test.

(DOC)
Video S1 Sample of automated injection of M. marinum into the yolk of early-stage zebrafish embryos. The left panel allows a side-view of the injection process, whereas the right panel allows the visualization of the inoculum as it is injected into the yolk (camera located beneath the embryo). The video runs at normal speed. (MP4)