Abstract
Synthetic biology is transforming therapeutic paradigms by engineering living cells and microbes to intelligently sense and respond to diseases including inflammation1,2, infections3-5, metabolic disorders6,7, and cancer8,9. However, the ability to rapidly engineer new therapies far outpaces the throughput of animal-based testing regimes, creating a major bottleneck for clinical translation10,11. In vitro approaches to address this challenge have been limited in scalability and broad-applicability. Here, we present a bacteria-in-spheroid co-culture (BSCC) platform that simultaneously tests host species, therapeutic payloads and synthetic gene circuits of engineered bacteria within multicellular spheroids over a timescale of weeks. Long-term monitoring of bacterial dynamics and disease progression enables quantitative comparison of critical therapeutic parameters such as efficacy and biocontainment. Specifically, we screen S. typhimurium strains expressing and delivering a library of antitumor therapeutic molecules via several synthetic gene circuits. We identify novel candidates exhibiting significant tumor reduction and demonstrate high similarity in their efficacies using a syngeneic mouse model. Lastly, we show that our platform can be expanded to dynamically profile diverse microbial species including L. monocytogenes, P. mirabilis, and E. coli in various host cell types. This high-throughput framework may serve to accelerate synthetic biology for clinical applications and understanding the host-microbe interactions in disease sites.
Introduction
The abundance of naturally-occurring microbes living in and on the human body present a myriad of opportunities for engineering microbes to act as in situ therapies or diagnostics. As a result, an emerging focus of synthetic biology has been to engineer bacteria to intelligently sense and respond to disease states including inflammation1,2, infections3-5, metabolic disorders6,7, and cancer8,9. Notably, many bacteria have been found to selectively colonize tumors in vivo, prompting attempts to engineer bacteria as programmable vehicles to deliver anticancer therapeutics12. However, a major bottleneck for clinical translation in all these cases has been animal-based testing, which slows iterations of design cycles needed to produce robust microbial systems for in vivo applications. Consequently, typical development of engineered bacteria occurs in environments far from in vivo conditions, inevitably leading to failure of predicted functions in more stringent native niches10,13.
To bridge this gap, in vitro platforms have been developed to characterize small-molecule drugs and biologics in more physiologically relevant conditions, with examples including organs-on-a-chip, microfabricated cell patterning, and multicellular spheroids and organoids14-17. These systems allow for growth of cells in three-dimensions (3-D), which preserves characteristic features of in vivo environments such as gradients in oxygen and metabolites, cell-cell interactions, and intra-cellular variations, as compared to 2-D monolayer cultures18. Furthermore, analyses such as drug distribution and spatially heterogeneous responses can be performed that are otherwise unattainable in monolayer systems19. For bacterial therapies, three-dimensional disease models could provide an ideal testbed for quantitatively monitoring bacterial localization and circuit dynamics that are critical for accurate estimation of safety and efficacy in vivo.
Due to the rapid proliferation rate of bacteria, long-term co-culturing with mammalian cells has been a challenge, limiting assays to short time periods or necessitating the use of heat-killed bacteria20,21. Previous attempts to control the imbalance in growth rates have fluidically controlled excess bacteria, directly injected bacteria into multicellular aggregates, or utilized obligate anaerobes to prevent overgrowth22-24. As a result, these systems expectedly increase technical complexity and restrict species types, reducing throughput and broad-applicability, respectively. Furthermore, long-term analysis of bacteria circuit dynamics has yet to be employed in multicellular co-cultures or applied towards therapeutic development21,22. In order to narrow down the large space of microbial therapy candidates, simple and high-throughput 3-D testing platforms are needed to accelerate development for in vivo applications.
Here we present a bacteria-in-spheroid co-culture (BSCC) platform to test a large number of engineered microbial therapies that can be created from combinations of genetic circuits, therapeutic payloads, and host species in 3-D multicellular spheroids (Fig. 1A). By selectively confining bacterial growth within spheroids, we enable parallel and stable co-culture with diverse bacteria and cell types. Using the BSCC system, we screened clinically-relevant S. Typhimurium delivering a library of anticancer molecules via synthetic gene circuits and identified novel therapies for in vivo applications.
Results
Establishing A Stable Bacteria Co-Culture System in Multicellular Spheroids
To rapidly screen microbial therapies, we generated 3-D multicellular spheroids in 96-well plates for parallel testing16 (Movie S1). We first characterized attenuated S. typhimurium, a model bacterium well-characterized for its tumor colonization and anticancer applications in vivo12,25, and constitutively expressed sfGFP to track its dynamics. Following a short incubation of bacteria with spheroids, we screened several antimicrobial agents and found that optimal concentrations of gentamicin confined bacteria within spheroids without causing cell toxicity (Fig. S1 and Table S1). Here, bacteria infiltrated spheroids and subsequently localized and proliferated in the spheroid (Movie S2). To investigate the spatiotemporal dynamics of bacteria in detail, we used automated image analysis to quantify the distribution of fluorescence intensity within the spheroid over time. Bacteria fluorescence was observed to increase near the spheroid boundary (40 hours), reach a steady state level of fluorescence (110 hours), and eventually grow but remain contained within the spheroid (140 hours) (Fig. 1B and Fig. S2). Bacterial proliferation over time was confirmed by plating dissociated spheroids on selective agar (Fig. 1C). Bacteria achieved stable colonization for up to 2 weeks and localized to necrotic and hypoxic regions of spheroids (Fig. 1D and Fig. S2), analogous to bacteria colonization conditions in vivo26. These results suggest our BSCC platform enables long-term, quantitative characterization of bacterial dynamics within physiologically-relevant 3-D models in a highly parallel manner.
Quantitatively Monitoring Synthetic Gene Circuit Dynamics in BSCC
We first sought to characterize gene circuit dynamics relevant to in vivo therapeutic delivery using the BSCC platform. We constructed an acyl-homoserine lactone (AHL) inducible luxI promoter (Fig. 2A) and characterized circuit dynamics by monitoring downstream production of sfGFP from S. typhimurium within the spheroid. We applied 10 μM AHL to spheroids after bacteria localized within the spheroid, and observed >400% induction in sfGFP signal that reached steady-state expression within 10 hours (Fig. 2A, Fig. S3 and Movie S3), demonstrating the ability to rapidly trigger therapeutic production within a tumor environment. To sustain therapeutic production without repeated infusion of chemical inducer, we next sought to engineer positive feedback from luxI to build a quorum-sensing (QS) circuit (Fig. 2B). This circuit enables self-triggered gene expression when bacteria reach a critical density within the tumor core27. After 140 hours of bacterial colonization, we observed a sharp increase in sfGFP production, demonstrating QS activation (Fig. 2B, Fig. S3 and Movie S4). Although sfGFP signal continued to increase over the next 10 hours, maximum induction was approximately 2-fold less than the inducible system (Fig. S3), likely due to reduced levels of local AHL concentration. To enhance drug release from bacteria, we incorporated quorum-mediated bacterial lysis. We modified a recently created synchronized lysis circuit (SLC, Fig. 2C), which produces periodic cycles of self-lysis of host bacteria under QS control to efficiently release therapeutics into the surrounding environment28. To minimize bacterial lysis before tumor colonization, we designed the circuit in a single operon on a low copy number plasmid. Upon bacterial colonization of the spheroid, we detected fluctuations in fluorescence from SLC bacteria, indicating periodic lysis (Fig. 2C, Fig. S3 and Movie S5). In addition, spatiotemporal analysis indicated localization of bacteria in the inner spheroid core, while total sfGFP signal did not increase up to 170 hours of bacterial colonization (Fig. 2C and Fig. S4), suggesting a smaller and spatially-restricted bacterial population due to SLC. When gentamicin was removed from the media, we found that SLC maintained containment inside spheroids 2-fold longer than a non-lysing inducible circuit (Fig. S4).
Screening Efficacy of Therapeutic Payloads in BSCC
Since the vast majority of bacterial therapy studies test a small number of therapeutic payloads that are rarely compared to one another12, we sought to rapidly identify and compare novel bacterial therapeutics using the BSCC platform. We created a library of therapeutics including previously uncharacterized bacterial toxins and anti-cancer peptides (Table S2). Next, we tested bacterial therapy by selectively expressing therapeutics from bacteria within spheroids using the inducible system. After adding AHL to spheroids that contain bacteria, we observed varying degree of reduction in tumor spheroid growth (Fig. 3A). The three candidates that exhibited the highest reduction in spheroid growth were azurin29, theta-toxin30 and hemolysin E31. The former is a pro-apoptotic protein that demonstrated limited efficacy in monolayers. The latter two are pore-forming toxins, of which theta-toxin has not yet been studied as an engineered bacterial therapeutic to our knowledge. Histopathological analysis of treated spheroids revealed higher levels of tumor cell death following treatment with bacteria expressing pore-forming toxins compared to treatment with control bacteria (Fig. S5).
To compare the effects of genetic circuit on efficacy, we tested expression of the therapeutic library across inducible, QS, and SLC systems in the BSCC platform. Incorporating the QS circuit into bacteria expressing the therapeutic library produced limited efficacy in tumor spheroids (Fig. 3B), corroborating with our observation that quorum activation is limited compared to inducible circuits. In contrast, many therapeutics displayed significant efficacy when expressed from the SLC system (Fig. 3C). Comparing across the therapeutic library, we identified theta-toxin as a potent therapy when combined with SLC, resulting approximately 40% reduction in tumor spheroid growth (Fig. 3C). Given the toxicity of theta-toxin to host bacteria (Fig. S6), we varied AHL in the inducible circuit from 10 nM to 0 nM and found that decreasing inducer concentration did not increase efficacy (Fig. S6). Therefore, we reasoned that theta-toxin expression from SLC exhibited higher therapeutic efficacy compared to the inducible circuit due to efficient release.
Validating Efficacy of Bacterial Therapies in an Animal Model
To determine the predictive ability of the BSCC system, we assessed engineered bacteria in a syngeneic, hind-flank tumor mouse model harboring CT-26 cells identical to those used to generate spheroids. We investigated inducible therapeutics with predicted high efficacy (azurin, theta-toxin and hemolysin E), moderate efficacy (beta-hemolysin), and control (sfGFP). Approximately 3-fold reduced tumor growth was observed by bacterial therapeutics identified as highly effective from the in vitro screen compared to the control treatment (Fig. 3D). Tracking therapeutic responses over time revealed a high degree of similarity in trends of efficacy between BSCC and in vivo results (Fig. 3E). In contrast, results from bacteria co-cultured with a monolayer of the same CT-26 cells failed to predict trends of efficacy in vivo (Fig. S7). We performed histopathological analysis on treated tumors at trial termination. Consistent with the spheroid results, tumors treated with theta-toxin and hemolysin E showed increased cell death relative to tumors treated with control bacteria (Fig. S8).
We next examined the QS and SLC circuits, and combination therapy in mouse tumor models. We found that efficacy of therapeutics expressed under SLC control in vivo corresponded to those from the spheroid screen (Fig. 3F). Comparing efficacy of SLC and QS circuits, theta-toxin from SLC exerted the strongest response, similar to the results from the BSCC platform (Fig. 3G). Furthermore, overall animal health as measured by weight drop improved with SLC compared to bacteria carrying inducible circuits, similar to the observation of enhanced biocontainment by SLC in the BSCC platform (Fig. S9). Leveraging the high-throughput capability of the BSCC, we also investigated combination therapy. Applying pairwise combinations of bacteria carrying inducible circuits at equal proportions in the BSCC, we found theta-toxin and azurin produced the most effective combination (Fig. 3H), exhibiting higher therapeutic efficacy than the additive effect of each individual therapy. This combination therapy in vivo exerted significantly stronger anti-tumor effects than either therapeutic alone, yielding a 4-fold reduction in tumor graft growth compared to control (Fig. 3I). These findings indicate the BSCC platform predictively identifies potent genetic circuits and therapeutic combinations in a highly parallel manner.
Expanding Applicability of BSCC to Diverse Bacterial Strains, Species and Cell Types
Given existing 3D co-culture models for bacterial therapy are typically limited to particular types of bacteria and cells24,32, we assessed the ability of the BSCC platform to co-culture diverse bacterial strains, species and host tissue types. First, we examined the colonization capacity and sfGFP expression level of clinically-relevant strains of S. typhimurium: ELH430 (SL1344 phoPQ-), SL7207 (SL1344 aroA-), ELH1301 (SL1344phoPQ-/aroA-)33,34, and VNP20009 (14028S msbB-/purI-), which has been used in clinical trials25. All strains successfully colonized tumor spheroids, demonstrating stable co-culture (Fig. S10). ELH1301 exhibited the highest colonization and sfGFP expression levels within spheroids (Fig. S10). Analysis of dynamics revealed increased initial invasion by ELH1301 compared to others (Fig. 4A), possibly contributing to its high colonization capacity. Next, we investigated colonization of L. monocytogenes, P. mirabilis, and E. coli, bacterial species previously tested for cancer therapy26,35. Optimizing bacterial inoculation density, incubation time, and gentamicin concentrations (Table S3), we established long-term growth of all bacterial species, indicated by increase in sfGFP fluorescence and CFU from spheroids over multiple days (Fig. 4B-D). Spatiotemporal analysis allowed for comparison of fluorescence distribution within spheroids and respective dynamics. L. monocytogenes displayed a notably wide area of tumor colonization (Fig. 4B), possibly because of its cell-to-cell spreading ability within tumor mass36. P. mirabilis reached the spheroid core at earlier time compared to other bacteria (Fig. 4C), which might be attributable to its swarming ability. E. coli displayed a sharp increase in number at day 6 within the spheroid core (Fig. 4D), demonstrating similar dynamics to S. typhimurium. In addition to the ability to test host species, the BSCC platform can be used to study different tissue types. To do this, we generated spheroids derived from both human and mouse colorectal (HT-29 and CT-26/MC-26) and breast (MCF-7 and 4T1) cells. We observed increasing sfGFP fluorescence from S. typhimurium over time (Fig. S11), demonstrating growth of bacteria in various cell lines. As efforts to integrate additional bacterial species for therapeutics continue to expand, we anticipate that the modularity of the BSCC system allows rapid evaluation of broad microbial therapies and disease modalities.
Discussion
We present an approach to simultaneously profile a vast number of engineered bacteria in a physiologically-relevant, 3-D multicellular system. By utilizing a scalable methodology, the BSCC platform enables rapid ‘build-test’ cycles of bacterial dynamics, therapeutics, and species for in vivo applications of synthetic biology. The key features of our system are: 1) quantification of bacterial population and circuit dynamics in a 3-D disease model, 2) accurate prediction of long-term disease progression in vivo from in vitro screening, and 3) a simple and broadly-applicable system for high-throughput development of novel therapies.
The BSSCC platform has several advantages over existing co-culture of microbes and mammalian cells, which have been of limited utility for the development of bacterial therapeutics. Traditional co-culture with monolayer of cells typically rely on frequent dilution or physical compartmentalization to balance the rapid growth of bacteria21, and often do not mimic the stoichiometry, geometry, or environmental conditions observed in vivo. More complex in vitro platforms including spheroids and organoids have been introduced to recapitulate the 3-D spatial orientation seen in vivo, and have been used as a high-throughput screening tool for molecular-based therapeutics16,37. However, these systems have yet to be developed for testing bacterial therapies. The BSCC system combines the advantages of 3-D multicellular co-cultures with high-throughput drug screening and allows for a simple and broadly-applicable system to characterize bacterial therapies.
One of the promising applications of the BSCC technology is the ability to rapidly screen large libraries of therapeutics expressed by bacteria circuit variants to narrow candidate selection for further development. While previous studies have tested up to 5 bacteria-produced therapeutics in monolayers using bacterial lysate and culture supernatants38, our technology enabled screening of ~40 bacterial therapy candidates for efficacy while simultaneously monitoring bacteria dynamics and disease progression over a time scale of weeks. The long-term monitoring aspect is critical to the success of bacterial therapies, as it is necessary to characterize the effects of dynamic therapeutic expression on bacteria and cellular growth in a spatially-dependent manner. In this study, we were able to identify theta-toxin driven by the SLC as an efficacious bacterial therapy candidate compared to several previously-tested toxins. Although further studies are required to explore specific mechanisms involved, our results suggest this may be due to effective therapeutic release and minimization on host toxicity provided by the SLC. Furthermore, the BSCC system identified that theta-toxin combined with azurin was an efficacious combination, providing additional efficacy compared to either single therapy. In future studies, the platform can be used to screen even larger libraries and combinations of 3 to 4 agents as often utilized in chemotherapy regimens39. Since multiple bacteria strains expressing different therapeutics will compete for resources within a tumor, reducing the effective dosage of each therapeutic, it has been unclear how combination therapy will perform when increasing the number of bacterial therapies. The BSCC system is precisely suited to quantitatively explore this strategy and determine how specific selection of orthogonally-acting therapeutics can be used to increase efficacy.
While we focused on cancer therapy in this study, the BSCC may be expanded to characterize bacteria-based therapeutics for various diseases, and explore fundamental biological questions about bacteria in host tissue types. For example, recent findings have demonstrated the presence of bacteria in patients with pancreatic tumors, and identified their potential role in degradation of chemotherapies40. The BSCC system would be able to provide a high-throughput way to assess the interactions between bacteria, cancer cells, and chemotherapies for further exploration. Additionally, several studies have reported a role for bacteria to augment immunotherapies41,42. By incorporating immune cells into the co-culture platform, the BSCC system may also be amenable to investigate bacteria-based immunotherapeutics. Furthermore, while multicellular spheroid models enable a reproducible and high-throughput assay format, other 3-D disease models such as organoids incorporate cells with different lineages that may enable further investigation of host-microbe interactions and efficacy. As the field of engineered cell and microbial therapies evolves with the use of more complex synthetic gene circuits43,44 and incorporation of multi-cellular ecosystems45, we envision the BSCC platform to accelerate therapeutic development for various diseases towards clinical translation.
Author contributions
T.H., Z.S., and T.D. conceived and designed the BSCC platform. T.H., Z.S. and O.S.V. performed experiments and established the BSCC platform. T.H. built and tested bacterial strains, gene circuits, and therapeutic library. T.H., O.S.V., J.Z., S.C. and T.E.H. designed and performed in vivo experiments. T.H., W.M, and T.D. performed image and data analysis. T.H. and T.D. wrote the manuscript with input from all of the other authors.
Materials and Methods
Host strains and culturing
ELH1301 and ELH430 were kindly provided by Dr. Elizabeth Hohmann. SL7207 was kindly provided by Dr. Siegfried Weiss. L. monocytogenes was kindly provided by Dr. Eric Pamer. VNP20009, P. mirabilis and E. coli were obtained from ATCC (202165, 29906, 23506). For full strain information, please refer to the Table S4. S. typhimurium, P. mirabilis and E. coli were cultured in LB media (Sigma-Aldrich). L. monocytogenes was cultured in Brain Heart Infusion media (Fisher Scientific). All bacteria were grown with appropriate antibiotics selection (100 μg ml-1 ampicillin, 50 μg ml-1 kanamycin, 25 μg ml-1 chloramphenicol) at 37 °C. Synchronized lysis circuit strains were cultured with 0.2% glucose for less than 16 hours. The glucose was added in order to decrease expression from the luxI promoters.
Mammalian cells and spheroid generation
The MC-26 cell line was provided by K. Tanabe and B. Fuchs (Massachusetts General Hospital). Mammalian cells were cultured in DMEM/F-12 media with GlutaMAX supplement (Gibco; for HT-29 and MCF-7) or RPMI 1640 media (Gibco; for CT-26, CT-26-iRFP, MC-26, 4T1) and supplemented with 10% fetal bovine serum (Gibco) and 1% penicillin/streptomycin (CellGro), placed inside a tissue culture incubator at 37 °C maintained at 5% CO2. For full cell line information, please refer to the Table S4. CT26 cells were transfected with genomic integration of iRFP gene with nuclear localization signal sequence to construct CT26-iRFP cell line. Specifically, pNLS-iRFP670 plasmid was obtained from Addgene (Plasmid #45466). The plasmid was prepared by miniprep of an overnight culture and subsequently transfected into CT26 cells using lipofectamine (Invitrogen). Monoclonal cells were selected by serial limiting dilution into 96-well plates (Falcon). Selected iRFP-expressing monoclones were expanded and used to seed spheroids in all corresponding assays.
Tumor spheroids were generated by seeding cells in round-bottom ultra-low attachment 96 well plate (Corning). Each well contains 2,500 CT-26 cells in 100 μl of appropriate media without antibiotics. Number of cells seeded was adjusted for each cell line to maintain a similar diameter of spheroids generated across cell lines (MC-26 2,500 cells, 4T1 5,250 cells, HT-29 5,000 cells, MCF-7 12,500 cells). The plate was centrifuged at 3,000 rcf for 5 minutes to aggregate cells at the bottom of the plate and placed inside a tissue culture incubator for 4 days before co-cultured with bacteria.
Plasmids and therapeutic library constructions
Plasmids were constructed using Gibson Assembly or using standard restriction digest and ligation cloning and transformed into Mach1 competent cells (Invitrogen). Previously constructed pTD103 sfGFP plasmid containing kanamycin resistance cassette and ColE1 origin of replication was used to characterize the inducible gene circuit1. The QS circuit (pTH02) was constructed by switching antibiotic resistance cassette and origin of replication of previously used pTD103 LuxI sfGFP plasmid2 to ampicillin and p15A using the modular pZ plasmids3. The synchronized lysis circuit (SLC) plasmid (pTH03) was constructed by first amplifying a region containing the constitutively expressed luxR gene and luxI gene under the control of luxI promoter from pTD103 LuxI sfGFP plasmid. Next, the bacteriophage ϕ X174E was synthesized from IDT and cloned next to the luxI gene with intergenic RBS sequence between genes to allow for operon expression (GAGGCAGATCAA). Finally, the antibiotic resistance cassette and origin of replication was replaced with ampicillin and sc101* from a modular pZ plasmid. Maps of main plasmids used in this study (Fig. S12) are provided.
The therapeutics library was constructed by synthesizing therapeutic genes from IDT, except for the hemolysin E gene obtained via PCR from a plasmid in previous work4. Therapeutics were cloned under the control of the luxI promoter by replacing sfGFP gene in a previously used ColE1 pTD103 sfGFP plasmid to construct therapeutics with inducible control (pTH05) 1. To combine QS circuit and SLC with therapeutic expression, plasmids containing these circuits were co-transformed into bacteria and plated on full antibiotics. To make therapeutic characterization comparable between circuits, the QS circuit was swapped to a sc101* origin of replication (pTH06). A detailed table of therapeutics in the library (Table S2 and S4) and main plasmids used in this study (Fig. S12) are provided.
Bacteria co-culture with tumor monolayer cells
Cell viability experiment was performed in 96-well tissue culture plates (Falcon). CT26 cells were allowed to adhere to the wells for 24 hours before the addition of bacteria. 1 x 105 bacteria carrying therapeutic payloads were co-cultured for 4 hours before media was replaced by new cell culture media containing 50 μg/ml gentamici n (Gibco). After 48 hours, viability was assessed using an MTT assay by measuring the colorimetric output in a TECAN Infinite M200 Pro plate reader.
Bacteria co-culture with tumor spheroids
Bacteria were cultured in a 37 °C shaker overnight to reach stationary phase before use. 106 CFU S. typhimurium were inoculated into wells containing 4-day old tumor spheroids and placed back into the tissue culture incubator. After 2 hours of bacteria inoculation, media was removed and tumor spheroids were washed with 200 μl of PBS repeatedly while leaving spheroids at the bottom of plate. After washing, 200 μl of media containing 2.5 μg/ml gentamicin (Gibco) was added and tumor spheroids were monitored for growth. For long-term experiments, media was replaced every 3-4 days. Schematic of the overall protocol is provided (Fig. S1). Bacterial therapeutics expression was induced at day4 by replacing media containing 10 nM N-(β-Ketocaproyl)-L-homoserine lactone (AHL) (Sigma Aldrich). For other bacteria, identical procedures were followed with modifications in bacterial inoculation density, incubation time, and gentamicin concentration. For full protocol, please refer to Table S3. ELH1301 (SL1344 phoPQ-/aroA-) was used for all co-culture experiments unless otherwise noted.
Bacterial colonization quantification via colony counts
Spheroids containing bacteria were washed with 200 μl of PBS repeatedly while leaving spheroids at the bottom of plate. After washing, spheroids were re-suspended in 100 μl of PBS and homogenized using mechanical dissociation with sterile tips and repeated pipetting. Destruction of spheroids were confirmed by microscopy. Serial 10-fold dilutions of the samples were inoculated on appropriate agar plates.
Microscopy
Acquisition of spheroid still images was performed with EVOS FL Auto 2 Cell Imaging Systems. The scope and accessories were programmed using the Celleste Imaging Analysis software. For analysis of synthetic gene circuit dynamics, Nikon TiE microscope equipped with Okolab stage top incubator was used to maintain the culture at 37 °C with 5% CO2 for time-lapse movies. The scope and accessories were programmed using the Nikon Elements software and images were taken every 60 minutes. For the acquisition of images, we used an Andor Zyla sCMOS camera. The microscope and acquisition was controlled by the Nikon Elements software. Phase-contrast images were taken at 10x magnification at 50-200ms exposure times. Fluorescent imaging at 10x was performed at 70ms for sfGFP, 30% setting on the Lumencor Spectra-X Light Engine. Further information on the analysis of these images is presented in the Image Analysis section below.
Image alignment and localized fluorescence measurement for spheroids
Time-course imaging of tumor spheroids required image registration to properly align the resulting images, since the spheroid would both rotate and translate significantly within the field of view. For image registration, we leveraged the popular registration method ORB5 in the Python implementation of OpenCV to find matching key points in images at adjacent time points. An implementation of RANSAC6 in scikit-image7 determined the corresponding Euclidean transform between adjacent images. For each time point, we averaged edge-filtered transmitted light (TL) and edge-filtered sfGFP images to form the input for our registration pipeline, which allowed both TL and sfGFP to inform alignment. Once this method was used to align a time-course experiment, Fiji8 was used for data analysis. sfGFP fluorescence trajectories for a spheroid (Fig. S3a) are based on averaged signal of this aligned image set within several circular regions of interest (ROI’s) of the same size. These circular ROI’s are fixed in position and chosen to highlight representative dynamics of the spheroid. Images showing ROI’s used in this study are provided in Fig. S3.
sfGFP average fluorescence and radial histograms for spheroids
To measure the spatiotemporal dynamics of bacteria invading tumor spheroids, we first found a threshold brightness value for each TL image to distinguish the dark spheroid from the light background. We used scikit-image implementations of two popular thresholding methods: the minimum method9 for images taken daily, and Yen’s method10 for other images. We then identified the largest region within the resulting threshold-based image mask as the tumor spheroid, and we determined mean intensity of sfGFP fluorescence within this region. To compute radial histograms, we computed mean sfGFP fluorescence for many thin annuli with variable mean radius and centered on the centroid of the spheroid mask region. To compute radial fluorescence in Fig. 1d, fluorescence was calculated based on a line across the radius of the spheroid.
Animal models
All animal experiments were approved by the Institutional Animal Care and Use Committee (Columbia University, protocol AC-AAAN8002). The protocol requires animals to be euthanized when tumor burden reaches 2 cm in diameter, or under veterinary staff recommendation. Mice were blindly randomized into various groups. Animal experiments were performed on 4-6 weeks-old female BALB/c mice (Taconic Biosciences) with bilateral subcutaneous hind flank tumors from CT26 colorectal cells. The concentration for implantation of the tumor cells was 5x107 cells per ml in RPMI (no phenol red). Cells were injected at a volume of 100 μl per flank, with each implant consisting of 5 x 106 cells. Tumors were grown to an average of approximately 150 mm3 before experiments. Tumor volume was quantified using calipers to measure the length, width, and height of each tumor (V = L × W × H). Volumes were normalized to pre-injection values to calculate relative or % tumor growth on a per mouse basis.
Bacterial administration for in vivo experiments
Bacterial strains were grown overnight in LB media containing appropriate antibiotics and 0.2% glucose. A 1:100 dilution into media with antibiotics was started the day of injection and grown until an OD of approximately 0.1. Bacteria were spun down and washed 3 times with sterile PBS before injection into mice. Intratumoral injections of bacteria were performed at a concentration of 5 × 108 cells per ml in PBS with a total volume of 20–40 μl injected per tumor. For bacteria carrying the inducible circuit, 0.5 mL of 10 μM AHL was injected subcutaneously the day after bacterial treatment to induce therapeutic expression. For combination therapy, bacteria cultures were combined after washing with PBS at 1:1 ratio to reach total concentration of 109 cells per ml and injected the total volume of 20-40 μl per tumor.
Tissue histological analysis
Tumors were extracted from mice at the termination of trials according to protocol. Immediately after extraction, tumor tissues were rinsed with PBS and fixed in 4% paraformaldehyde for 24 hours in 4 °C. Tumor spheroids were fixed for 20 minutes to prevent over fixation. After fixation, the tissues were rinsed with PBS and preserved in 70% ethanol in 4 °C. For histological analysis, tissues were paraffin embedded and sectioned into 5 μl onto slides. Gram staining was performed to confirm presence of bacteria inside the tumors. TUNEL staining was performed to obtain measurement of apoptosis. Tumor cell death was quantified using Fiji software to measure the area of the viable and dead area by setting a pixel threshold to make binary images.
Statistical analysis
Statistical tests were calculated either in GraphPad Prism 7.0 (Student’s t-test and ANOVA) or Microsoft Excel. The details of the statistical tests carried out are indicated in the respective figure legends. Where data were approximately normally distributed, values were compared using either a Student’s t-test or one-way ANOVA for single variable, or a twoway ANOVA for two variables with Bonferroni correction for multiple comparisons. Mice were randomized in different groups before experiments.
Calculation of additional efficacy
We calculated additional efficacy, which represents the increased efficacy observed compared to expected from additive effect alone, based on past calculations of synergy and combination therapies for other drugs11. Unique to bacterial therapy combinations, when two different bacteria share the tumor space, each can only grow to half of the possible maximum bacterial volume of a single therapy. Efficacy of a single therapy (for example, therapy A) was calculated by measuring the percent reduction of spheroid growth as a result of treatment with that therapy alone. For example, 80% relative tumor growth after treatment with A was counted as A = 20% efficacy. The expected additive efficacy between two therapies was calculated as (A+B)/2. Thus, to calculate additional efficacy, we subtracted the expected additive efficacy from the measured efficacy from combinatorial treatment with A and B (“AB” term in the equation), with a formula of additional efficacy = AB - (A+B)/2, so that any result greater than 1 indicated the combinatorial therapeutic strategy had greater efficacy than additive effect of the component therapies.
Acknowledgements
We thank the Herbert Irving Comprehensive Cancer Center Molecular Pathology Shared Resources Facility for help with histological sample processing. We thank N. Arpaia, M. Omar Din, and J. Hasty for their helpful comments and suggestions. This work was supported in part by the Honjo International Foundation Scholarship (T.H.), NIH Ruth L. Kirschstein National Research Service Award (Z.S.), NIH Pathway to Independence Award (R00CA197649-02), DoD Idea Development Award (LC160314), and DoD Era of Hope Scholar Award (BC160541). T.H., Z.S. and T.D. have filed a provisional patent application with the US Patent and Trademark Office related to this work.