The motility regulator flhDC drives intracellular accumulation and tumor colonization of Salmonella

Background Salmonella have potential as anticancer therapeutic because of their innate tumor specificity. In clinical studies, this specificity has been hampered by heterogeneous responses. Understanding the mechanisms that control tumor colonization would enable the design of more robust therapeutic strains. Two mechanisms that could affect tumor colonization are intracellular accumulation and intratumoral motility. Both of these mechanisms have elements that are controlled by the master motility regulator flhDC. We hypothesized that 1) overexpressing flhDC in Salmonella increases intracellular bacterial accumulation in tumor cell masses, and 2) intracellular accumulation of Salmonella drives tumor colonization in vitro. Methods To test these hypotheses, we transformed Salmonella with genetic circuits that induce flhDC and express green fluorescent protein after intracellular invasion. The genetically modified Salmonella was perfused into an in vitro tumor-on-a-chip device. Time-lapse fluorescence microscopy was used to quantify intracellular and colonization dynamics within tumor masses. A mathematical model was used to determine how these mechanisms are related to each other. Results Overexpression of flhDC increased intracellular accumulation and tumor colonization 2.5 and 5 times more than control Salmonella, respectively (P < 0.05). Non-motile Salmonella accumulated in cancer cells 26 times less than controls (P < 0.001). Minimally invasive, ΔsipB, Salmonella colonized tumor masses 2.5 times less than controls (P < 0.05). When flhDC was selectively induced after penetration into tumor masses, Salmonella both accumulated intracellularly and colonized tumor masses 2 times more than controls (P < 0.05). Mathematical modeling of tumor colonization dynamics demonstrated that intracellular accumulation increased retention of Salmonella in tumors by effectively causing the bacteria to bind to cancer cells and preventing leakage out of the tumors. These results demonstrated that increasing intracellular bacterial density increased overall tumor colonization and that flhDC could be used to control both. Conclusions This study demonstrates a mechanistic link between motility, intracellular accumulation and tumor colonization. Based on our results, we envision that therapeutic strains of Salmonella could use inducible flhDC to drive tumor colonization. More intratumoral bacteria would enable delivery of higher therapeutic payloads into tumors and would improve treatment efficacy. Electronic supplementary material The online version of this article (10.1186/s40425-018-0490-z) contains supplementary material, which is available to authorized users.


Introduction
Effective tumor colonization is essential for bacterial anti-cancer therapy. With poor colonization, insufficient treatment is delivered and the tumor response is hampered. For bacterial therapy, the tumor density is controlled more by the rate of colonization than the administered dose [1]. However, the mechanisms that control colonization are poorly understood. It has been well established that after intravenous injection into mice, Salmonella colonizes tumor tissue at ratios greater than 10,000:1 compared to other organs in the body [2]. It is this tumor specificity that makes Salmonella-based therapy particularly attractive as a targeted delivery agent [3]. Unfortunately, clinical trials showed that tumor colonization in humans was not sufficient to induce a lasting response [4]. Therefore, understanding and controlling the mechanisms that drive bacterial tumor colonization could greatly improve bacterial tumor therapy.
Two mechanisms that could affect tumor colonization are intratumoral motility and intracellular accumulation. We have previously shown that bacterial motility plays a critical role in the accumulation of Salmonella in tumors [5][6][7]. Upregulating motility by swim-plate selection increases distal tumor colonization of the bacteria [6,8] and altering chemotactic sensing increases bacterial penetration into tumor masses [7][8][9]. Salmonella motility is controlled by the master regulator flhDC [10][11][12]. The flhDC protein complex regulates expression of the functional flagellar components [13]. This regulator is one of the most tightly regulated transcription factors within bacteria [14][15][16][17][18][19]. Flagella-dependent motility is downregulated under nutrient deprivation in Salmonella, which helps Salmonella survive intracellularly where there is limited availability of nutrients [20].
Intracellular invasion and growth are important mechanisms that could also affect Salmonella colonization of tumors. Salmonella have two type three secretion systems, T3SS1 and T3SS2, that promote invasion, survival, and growth inside epithelial cells [21]. Other Salmonella invasion systems include the Rck system, which invades cells by binding to epidermal growth factor receptor [22]. In the gut, Salmonella use these systems to invade and grow inside intestinal cells [23]. Disabling T3SS2 limits the ability of Salmonella to inhibit tumor growth [24]. When T3SS2 genes are deleted by transposon insertion, bacterial accumulation in the spleen is reduced [25]. After serial passaging in mice, Salmonella with increased intracellular invasion had enhanced persistence [26]. We have seen similar effects in tumor cell masses in vitro. Compared to K-12 E. Coli that is T3SS deficient, Salmonella had considerably greater colonization [5].
The two Salmonella secretions systems have distinct functions. T3SS1 initiates invasion into epithelial cells and T3SS2 enables intracellular growth and survival [21]. Both systems are composed of a needle apparatus that spans the inner and outer membranes, and the peptidoglycan layer [27]. Effector proteins are injected into the mammalian cells through the T3SS1 [27,28]. Secretion of T3SS1 effectors into mammalian cell cytoplasm is required for T3SS dependent intracellular invasion of Salmonella [29]. Once injected, these effectors cause a rearrangement of the mammalian actin cytoskeleton and endocytosis of Salmonella [30,31]. One essential effector protein is sipB. When knocked out, Salmonella cannot invade using T3SS1 [32]. When Salmonella have internalized, the bacteria modify the endocytic vacuole by secreting T3SS2 effectors [33][34][35]. These modifications confer protection to the bacteria and enable intracellular growth and survival [36,37]. The T3SS-dependent intracellular invasion and survival of Salmonella confers protection against extracellular clearance mechanisms, like compliment and attack by macrophages and neutrophils [23,38]. A non-functional T3SS2 apparatus impairs in vivo colonization and anti-tumor efficacy of Salmonella [24,25], indicating the importance of intracellular growth for survival of bacteria in vivo.
Flagella-dependent motility and intracellular invasion are not regulated independently. Rather, both of these systems are intertwined and there is a complex feedback between them [39,40]. Increasing bacterial motility also increases intracellular invasion [41]. The flhDC transcriptional complex controls elements of both motility and cellular invasion. In addition to controlling expression of motility genes, it directly controls the expression of the dual regulatory element, fliZ. FliZ controls both flagellar hook assembly and upregulates the transcription factor hilD [39][40][41]. HilD expression directly upregulates T3SS1 expression and intracellular invasion [39,40]. The systems are further connected because flagella can act as physical cell surface sensors to determine the optimal extracellular location to initiate invasion [42]. These systems are connected in part because the T3SS evolved from the flagellar type three secretion system (fT3SS), which is used to assemble functional flagella [43,44]. The co-regulation of motility and intracellular invasion further supports the idea that both of these phenomena are important for bacterial tumor colonization.
In addition to affecting intracellular invasion, flagella-dependent motility also affects the intracellular lifestyle of Salmonella. Immediately after invasion the majority of Salmonella reside in intracellular vacuoles. A small but significant fraction of the intracellular bacteria escape from the vacuoles into the cytosol [45][46][47]. Some cytosolic bacteria are degraded by host ubiquitination machinery [48][49][50][51][52]. Those that escape degradation replicate rapidly and are extruded from the cell [45]. The T3SS1 system and functional flagella play important roles in the escape from the vacuole and the hyper-replication [45][46][47]. After extrusion, the bacteria are primed for reinvasion because of the expression of flagella and SPI-I invasion genes [45,46].
The goal of this study was to measure the effect of intracellular accumulation on bacterial tumor colonization and quantify the interplay between intracellular accumulation and motility. The interaction of these mechanisms has not been previously studied in relation to using bacteria for cancer therapy. We hypothesized that 1) overexpressing flhDC in Salmonella increases intracellular accumulation in tumor cell masses, and 2) intracellular accumulation of Salmonella drives tumor colonization in vitro. To test these hypotheses, Salmonella were transformed with genetic circuits that induce flhDC and express green fluorescent protein (GFP) after cell invasion. Genetically modified Salmonella were perfused into a microfluidic tumor-on-a-chip device to assess colonization and invasion using time-lapse fluorescence microscopy. The potential to use flhDC as a bispecific switch to increase tumor colonization was determined by inducing expression after initial penetration. A mathematical model was used to investigate why intracellular invasion and growth improved tumor colonization of Salmonella. Controlling Salmonella invasion into cells will increase overall tumor colonization and has the potential to make these therapeutic bacteria more effective in the clinic.

Bacterial Strains and Plasmid Construction
Eight strains of Salmonella Enterica serovar Typhimurium were used throughout the experiments ( Table 1). The control strain (Sal) was based on an attenuated therapeutic strain of Salmonella (VNP20009) that has three deletions, ΔmsbB, ΔpurI, and Δxyl, that eliminate most toxicities in vivo. The background strain was transformed with a plasmid containing two gene circuits, P lac /DsRed and P SSEJ /GFP, that constitutively express DsRed and express GFP after intracellular invasion (Table 1; Additional file 1: Figure S1-A). The constitutive lac DsRed gene circuit was created by adding the wild-type lac promoter and a ribosomal binding site (AAGGAG) to the 5' end of the forward DsRed primer. The SSEJ promoter was copied by PCR from VNP20009 genomic DNA using the following primers: forward-ACATGTCACATAAAACACTAGCACTTTAG C and reverse-TCTAGACCTCCTTACTTTATTAAA CACGCT. The second strain, F-Sal, was transformed with a plasmid that contains a third gene circuit that enables induction of flhDC with arabinose (Table 1; Additional file 1: Figure S1-B). PCR was used to amplify the flhDC genes from Salmonella genomic DNA using the following primers: forward-AAAAAACCATGGGT TAATAAAAGGAGGAATATATATGCATACATCCGAG  Figure S1-A 2. F-Sal Sal P lac /DsRed P SSEJ /GFP P BAD /flhDC Arabinose Inducible flhDC Constitutive DsRed Intracellularly inducible GFP Additional file 1: Figure S1-B

Salmonella (control)
ΔmsbB, ΔpurI, Δxyl P lac /GFP Constitutive GFP Additional file 1: Figure S1-C 6. Salmonella + pflhDC ΔmsbB, ΔpurI, Δxyl P lac /GFP P BAD /flhDC TTGCTAAAACA and reverse-AAAAAACTCGAGAA AAATTAAACAGCCTGTTCGATCTGTTCAT. The PCR product and PBAD-his-myc plasmid (Invitrogen, Carlsbad, CA) were digested with NcoI and XhoI and ligated with T4 DNA ligase. The flhDC expression cassette, which includes the AraC regulator and PBAD controlled flhDC, was amplified with PCR and combined with a plasmid containing SSEJ-GFP and Lac-DsRed using Gibson Assembly. Both S-Sal, which has a sipB deletion, and the ΔflgE strain were generated using lambda red recombination [53]. When the flagellar hook (flgE) is deleted, Salmonella are unable to produce functional flagella and are non-motile [54]. The S-Sal strain (strain three) was transformed with the plasmid containing P lac /DsRed and P SSEJ /GFP (Table 1; Additional file 1: Figure S1-A). The fourth strain, FS-Sal, was transformed with a plasmid that contains inducible flhDC (P BAD / flhDC), constitutive DsRed expression (P lac /DsRed) and intracellular GFP expression (P SSEJ /GFP) in a ΔsipB background (Table 1; Additional file 1: Figure S1-B). A second control Salmonella strain (strain five) was transformed with a plasmid containing P lac /GFP to constitutively express GFP (Table 1; Additional file 1: Figure  S1-C). The constitutive lac GFP gene circuit was created similarly to the lac DsRed circuit, by adding the wild-type lac promoter and a ribosomal binding site (AAGGAG) to the 5' end of the forward GFP primer. The sixth strain, Salmonella+pflhDC, expresses GFP constitutively and flhDC upon induction with arabinose (Table 1; Additional file 1: Figure S1-D). The seventh strain, ΔflgE, is non-motile and expresses GFP constitutively (Table 1; Additional file 1: Figure S1-C). The eighth strain, ΔflgE+pflhDC, expresses GFP constitutively and flhDC upon induction with arabinose (Table  1; Additional file 1: Figure S1-D). All cloning was performed with DH5α E. Coli (New England Biolabs, Ipswich, MA) and all plasmids contained a ColE1 origin and either chloramphenicol or ampicillin resistance (Additional file 1: Figure S1). Salmonella were transformed by electroporation. All cloning reagents, buffer reagents, and primers were from New England Biolabs, Fisher Scientific (Hampton, NH), and Invitrogen, (Carlsbad, CA), respectively, unless otherwise noted.

Cell Culture
MCF7 breast carcinoma cells and LS174T colorectal carcinoma cells (ATCC, Manassas, VA) were maintained in DMEM (Dulbecco's Modified Eagle Medium; Sigma Aldrich, St. Louis, MO) with 1 g/L glucose, 3.7 g/L sodium bicarbonate (pH 7.4) and 10% FBS using standard cell culture techniques. Between passages of LS174T cells, single cell suspensions were transferred to PMMA coated cell culture flasks (2 g/L PMMA in 100% ethanol, dried before use) in order to produce spheroids.

Fabrication and Operation of Microfluidic Devices
Photolithography was used to make silicon wafer masters as previously described [55]. Two silicon wafers were made: One silicon wafer was used to make the pneumatic valve layer (layer 1). The other wafer was to make the media perfusion layer (layer 2). The fabrication of multi-layer tumor-on-a-chip devices was based on a previous method [56]. The microfluidic device was fabricated in two parts. Layer 1 was created by mixing 9 parts of Sylgard 184 PDMS (Ellsworth Adhesives, Wilmington, MA) with 1 part of curing agent and poured onto the pneumatic valve layer silicon master wafer. Layer 2 was created by mixing 15 parts of PDMS with 1 part (weight by mass) of curing agent and spun coat onto the media perfusion silicon wafer to a height of 200 μm. Both layers of PDMS were cured at 65°C for 1.5 h and layer 1 was aligned on top of layer 2. Both layers were cured together at 95°C for 1 h. Holes were punched in the PDMS layers to receive fluidic and control tubing. The PDMS layers were bonded to a glass slide by plasma treatment (Harrick Plasma Cleaner). The valves were pneumatically actuated before bonding to prevent the valve from sealing. Devices were taped to a microscope stage adaptor and inlet and outlet tubes were inserted. A 10% bleach solution was perfused at 3 μl/min throughout the device for 2 h followed by 70% ethanol for 1 h. The device was prepared for spheroid loading by perfusing for 1 h with DMEM with 1 g/L glucose, 20 mM HEPES (pH 7.4), 10% FBS and 33 μg/ml chloramphenicol (henceforth referred to as DMEM-HEPES-chlor). For all experiments,~300 μm diameter LS174T spheroids were positioned into a microfluidic device and equilibrated in DMEM-HEPES-chlor for 6 h at a flow rate of 3 μl/min. Some spheroids were damaged in the insertion process and these cell masses were not included in the image analysis.

Quantifying Intracellular Invasion and Colonization of Salmonella in a Tumor-on-a-chip
Four experiments were performed with tumor-on-a-chip device to quantify colonization and intracellular accumulation for (1) induced F-Sal compared to Sal, (2) FS-Sal compared to S-Sal, (3) S-Sal compared to Sal, and (4) for intratumoral induction of F-Sal compared to Sal. Salmonella strains were grown in LB with chloramphenicol (33 μg/ml) to a density of approximately 250 million CFU/ml. Bacteria were resuspended in DMEM-HEPES-chlor at a density of 10 million CFU/ml. The bacterial suspension was perfused into the tumor-on-a-chip device for 1h at a flowrate of 3 μl/min followed by bacteria-free DMEM-HEPES-chlor at the same flowrate for 48 h. In experiments one and two, the F-Sal and FS-Sal conditions contained 0.4% arabinose to induce flhDC. Flowing bacteria-fee medium prevents over growth in the flow channel and mimics clearance in vivo. For experiment four, the procedure was the same (bacterial perfusion for 1 h, followed by perfusion with bacteria-free medium), except that after 11 h, medium containing 0.4% arabinose was perfused into the device to induce flhDC intratumorally.
Transmitted and fluorescent images (480/525 excitation/emission for GFPmut3 and 525/590 for DsRed) of tumor masses were acquired every hour with an Olympus IX71 or a Zeiss Axio Observer Z.1 microscope. Time lapse microscopy images of each tumor mass were cropped using the rectangular cropping tool in ImageJ and were analyzed in Matlab. Each image was background subtracted. Fluorescent intensities of ten spatially equal sections of each tumor mass were averaged to quantify bacterial profiles for each time point. Overall bacterial density as a function of time was determined by averaging fluorescent intensities for entire tumor masses per time point. Red fluorescence was used to calculate total bacterial colonization and green fluorescence was used to calculate intracellular bacterial density. Each experiment was normalized by dividing every calculated average fluorescence intensity by the highest fluorescent intensity observed, which occurred during the last time point.

Quantifying Aqueous Motility of Salmonella
Aqueous motility was determined by growing flhDC inducible Salmonella in 0.4% arabinose. Twenty microliters of 400 million CFU/ml of either flhDC induced or control Salmonella was placed between a coverslip and a glass slide. Transmitted light microscopy images were taken every 0.68 seconds for approximately 30 seconds. The automated particle tracking plugin in ImageJ, Trackmate, was used to analyze bacterial swimming velocity. Aqueous velocity histograms were generated by binning the fraction of total bacteria into three velocity categories: 0-15 μm/s, 15-30 μm/s and >30 μm/s. Motility assays were performed in triplicate.

Quantifying Intracellular Invasion and Growth inside MCF7 Cells in Monolayer
Intracellular invasion of Salmonella was quantified by growing in LB and adding to monolayer cultures of MCF7 cancer cells. Four strains were used to quantify the dependence on flhDC expression and flagella formation: control Salmonella, Salmonella+pflhDC, ΔflgE, ΔflgE+pflhDC. Two strains were used to show the intracellular specificity of the P SSEJ promoter and the dependence on T3SS: Sal and S-Sal, using a modified gentamycin protection assay. Each strain was grown in LB to a density of 5 × 10 8 CFU/ml and added to 6-well plates of MCF7 cells at a density of 5 × 10 6 CFU/ml. After two hours of incubation, each well was washed ten times with one milliliter of phosphate buffered saline. DMEM with 20 mM HEPES and 40 μg/ml gentamycin was added to each well to remove residual extracellular bacteria. For two hours following the addition of gentamycin, the cultures were observed microscopically to assess the effectiveness of the PBS washes to remove extracellular bacteria. The few remaining extracellular bacteria were observed over this period to ensure that they were eliminated by the gentamycin treatment. After two hours, intracellular Salmonella were imaged over time at 10X magnification with fluorescence microscopy. After 18 hours, bacterial invasion was quantified by randomly identifying 20 cells in each culture and counting the fraction of cells that contained intracellular Salmonella, as indicated by GFP fluorescence.
A similar invasion protocol was used to calculate the intracellular growth rate of Salmonella. Both control Salmonella and Salmonella+pflhDC constitutively expressed GFP (Table 1). Time lapse fluorescence microscopy was used to quantify the fluorescence from P lac /GFP Salmonella inside MCF7 cells over time. Salmonella density was determined by multiplying the average intensity by the area of all intracellular bacteria within a cell, as a function of time. It was assumed that the amount of GFP produced per bacterium was constant over time. Only MCF7 cells containing bacteria and that did not divide for a six hour interval were used. Intracellular growth rate was calculated by fitting an exponential growth function to the intracellular bacterial density.

Mathematical Modeling
A mathematical model was created to interpret the spatiotemporal dynamics of bacterial dispersion, growth and invasion in tumor masses. This model was based on a previous model of bacterial growth in tumor tissue [57].
The coupled PDE model incorporated a balance on extracellular (eq. 1) and intracellular (eq. 2) bacteria.  (1) there were no intracellular or extracellular bacteria initially within the tumor mass; (2) the flux into or out of the tumor mass was equal to the flux in the supply channel; and (3) there was no flux at the distal (x = 1) boundary. The supply of extracellular bacteria (C ex,0 ) is a stepwise function that was set to match experimental conditions: 10 7 CFU/ml of bacteria were administered for 2 h, followed by perfusion of bacteria-free media for the remaining time.
The variables in the model are as follows: C ex and C in are the normalized extracellular and intracellular densities (a value of one corresponds to 1x10 10 CFU/ml), D is the dispersion coefficient, μ g and μ g,in are the extracellular and intracellular growth rates, μ inv is the intracellular invasion rate, θ is the fraction of viable tumor cells, K aff is the chemotactic affinity to chemokines in the tumor mass, C chem is the normalized chemokine concentration, C ex,0 is the normalized density of bacteria that was perfused into the microfluidic device as a function of time (1x10 7 CFU/ml for t ≤ 2 h and 0 for t > 2 h), F 0 is the media flow rate in the perfusion channel, V is the volume of the section of the perfusion channel in front of the tumor chamber, and A is the cross-sectional area of the tumor chamber. All intracellular and total bacterial fluorescence values were normalized to the highest cross sectional fluorescence intensity that occurred during the experiment.
Equations were discretized in space and solved in Matlab (The MathWorks, Inc., Natick, MA) using a finite difference method. The spatially discretized coupled ordinary differential equations were solved with the built-in ode15s function in Matlab for all spatial (discretized in ten points in space) and temporal points between 0 and 40 hours in 1 hour intervals. The fraction of viable cancer cells within the tumor mass (θ) was calculated based on previous data [9]. The extracellular growth rate was calculated based on the growth rate in liquid culture.
Two datasets (F-Sal vs. Sal and S-Sal vs. Sal) were used for modelling and normalized to one another to match control (Sal) conditions. The bacterial dispersion coefficient was calculated by fitting the model (eq. 1-3) to the tumor-on-a-chip experimental data of GFP for all spatial and temporal points up to 40 hours. The fminsearch function in Matlab was used to minimize the sum of least squares error between the experimental data and model by adjusting (and calculating) the rates of intracellular invasion and dispersion for both Sal datasets. The intracellular invasion rate of S-Sal was calculated by fixing the dispersion coefficient to be the same as Sal. The dispersion coefficient and intracellular invasion rate of F-Sal were calculated by bounding the dispersion coefficient such that it could not be lower than that of Sal. The intracellular accumulation rate was determined by quantifying the total change in intracellular density between 47 and 48 h.

Data and Statistical Analysis
Image and statistical analysis was performed in Matlab software. Unpaired t-tests with unequal variance were used to determine statistical significance with a level of P < 0.05.

Induction of flhDC increases tumor colonization of Salmonella
Overexpressing flhDC in Salmonella increased intratumoral dispersion and colonization (Fig. 1). When administered to a tumor-on-a-chip device (Fig. 1A), F-Sal (induced flhDC) colonized tumor masses more than Sal (control) Salmonella (Fig. 1B). Both strains contained P lac /DsRed and expressed DsRed constitutively. In these images, red fluorescence indicates overall bacterial density. At 30 h, the size of the colony formed by F-Sal (white arrows) was considerably larger than the one formed by Sal (black arrows, Fig. 1B). The area of both colonies increased in size from 30 to 48 h after bacterial administration. Both colonies were located deep into the tissue, away from the perfusion channel (see Fig. 1A), indicating that both strains actively penetrated the tumor masses as we have described previously [5,6]. Across multiple cell masses (n = 3 for Sal and n = 5 for F-Sal), the average density of F-Sal was significantly greater than Sal within entire tumor masses between 29 and 45 hours of bacterial colonization (P < 0.05; Fig. 1C). After 48 hours of bacterial colonization, F-Sal colonized both proximal (x ≤ 0.5) and distal (x = 0.9) tumor tissue more than Sal (P < 0.05; Fig. 1D). The density of F-Sal was greater than Sal throughout the middle of tumor masses (0.6 ≤ x 0.8), but was not significant (0.05 < P < 0.08) because of heterogeneous localization of colonies between cell masses (Fig. 1D). Overall, F-Sal colonized tumor tissue five-fold more than Sal (P < 0.05, Fig. 1E).
Four of the Salmonella strains (Sal, F-Sal, S-Sal and FS-Sal; Table 1) were transformed with P SSEJ /GFP (intracellular GFP) and P lac /DsRed (constitutive DsRed) to identify and differentiate total (red only) and intracellular (red and green) Salmonella (Fig. 2E). This genetic circuit is necessary in tumor cell masses, because constitutive fluorescence would not differentiate intracellular and extracellular bacteria. A gentamycin protection assay was used to show that P SSEJ is a specific intracellular promoter. After applying control Salmonella (Sal) to a monolayer of cancer cells, all extracellular bacteria were removed with gentamycin. At early time points (2 h after gentamycin addition), GFP had yet to be translated (Fig. 2F, lower left) and all bacteria expressed DsRed (Fig. 2F, upper left). By 18 h, all intracellular bacteria (Fig. 2F, upper right) expressed both DsRed (Fig.   2F, upper right) and GFP (Fig. 2F, lower right), showing that the genetic circuits functioned as expected. In tumor-on-a-chip devices, overexpressing flhDC increased intracellular bacterial density (green, Fig. 2G). The high expression of GFP throughout the tumor masses (Fig. 2G) indicates that many of the Salmonella (both Sal and F-Sal) were intracellular (Additional file 2: Figure S2). Across all cell masses, the intracellular density of flhDC-induced F-Sal was significantly greater than control Sal from 29 to 45 h after administration (P < 0.05; Fig. 2H). Forty-eight hours after bacterial administration, the intracellular colonization of F-Sal was 2.5 fold more than Sal (P<0.05, Fig. 2I). In the middle of cell masses (0.5 < x < 0.6), induced F-Sal accumulated in cells 2.5 times more than control Sal (P < 0.05, Fig.  2J). Highly motile F-Sal also accumulated in distal tumor tissue (x ≥ 0.8) ten-fold more than Sal (P<0.05, Fig. 2J). These results demonstrate that flhDC induced Salmonella to accumulate in tumor cells.

Induction of flhDC does not increase tumor colonization in the absence of intracellular accumulation
To investigate the effect of flhDC induction in the absence of T3SS-based invasion, ΔsipB Salmonella (S-Sal)  were administered to a tumor-on-a-chip device (Fig. 3).
No difference was seen in the colonization pattern of extracellular (red) or intracellular (green) Salmonella (Fig. 3A). Across multiple tumor cell masses (n = 3), no differences were observed in the location of Salmonella colonization after flhDC induction, based on DsRed expression (Fig. 3B), and there was no effect on total bacterial density (Fig. 3C). Similarly, flhDC induction did not affect the location of intracellular Salmonella based on GFP expression (Fig. 3D) or overall density of intracellular Salmonella (Fig. 3E). The lack of difference between FS-Sal and S-Sal indicates that flhDC-mediated intracellular accumulation requires a functional T3SS-I.

Intracellular accumulation of Salmonella increases tumor colonization in vitro
Minimally invasive, ΔsipB Salmonella (S-Sal) colonized tumor tissue less than control Salmonella (Sal, Fig. 4). Both S-Sal and control Sal expressed GFP after intracellular invasion and constitutively expressed DsRed (Table  1). Without sipB, Salmonella invaded cancer cells considerably less than controls, as indicated by diminished GFP fluorescence (Fig. 4A). S-Sal invaded MCF-7 cells six-fold less than the Sal control (P < 0.05, Fig. 4B). When, S-Sal were administered to tumor-on-a-chip devices the amount of intracellular bacteria (green) was considerably less than for control Sal (Fig. 4C). The number of intracellular Sal increased from 30 to 48 hours as indicated by the increase in GFP intensity, but little increase was observed for S-Sal (Fig. 4C). Over multiple devices (n = 6), S-Sal accumulated within tumor masses 2.5 fold less than the Sal control (P<0.05, Fig. 4D) and the rate of GFP fluorescence increase of S-Sal was four fold less than Sal (P<0.05; Fig. 4E). Total tumor colonization was quantified through constitutive DsRed fluorescence. Thirty hours after administration, more control Sal bacteria were present in devices than S-Sal (Fig. 4F). The difference between Sal and S-Sal was due to the increase in intracellular invasion because knocking out sipB did not affect the growth rates of the strains (Additional file 3: Figure S3 (P<0.05, Fig. 4G) and grew four fold slower than the Sal control (P<0.05; Fig. 4H). Sal visibly grew between 30 and 48 hours after bacterial administration, while the S-Sal density remained relatively unchanged during the same time period (Fig. 4F). These results demonstrated that intracellular accumulation is an essential component of Salmonella tumor colonization in vitro.

Intratumoral induction of flhDC improves colonization and intracellular accumulation of Salmonella
To determine if flhDC could be induced intratumorally, F-Sal was grown without arabinose and administered to tumor-on-a-chip devices. After induction with arabinose, F-Sal were 1.2 times faster in aqueous media compared to uninduced F-Sal (P<0.05; Fig. 5A).
To test intratumoral induction, F-Sal were administered to devices for one hour in arabinose free medium (Fig. 5B). Twelve hours after administration, 0.4% arabinose added to the medium delivered in the flow channel to induce flhDC (Fig. 5B). Twelve hours was chosen as the time to induce, because this was the time when bacterial colonies could first be seen in the tumor cell masses (red arrows, Fig. 5C). At 47 h after administration, colonies grew in both uninduced and induced devices, but the induced colonies were visibly larger and located farther from the flow channel (Fig. 5C). Over multiple devices (n = 5 for uninduced and n = 6 for induced), intratumorally induced F-Sal colonized distal tumor tissue (0.8 ≤ x ≤ 1) five-fold more than the Sal control after 47 hours (P<0.05, Fig. 5D). The total amount of intratumorally induced F-Sal was two-fold greater than Sal (P <0.05, Fig. 5E). Similar to overall density, induction increased the amount of intracellular F-Sal (Fig. 5F). Intracellular accumulation of intratumorally induced F-Sal was five-fold greater (P< 0.

Intracellular accumulation improves bacterial retention in tumors
A model of bacterial dispersion, growth and intracellular invasion was used to determine how modulating intracellular accumulation affected tumor colonization. The model includes balances on extracellular and intracellular bacteria (eq. 1-2). Extracellular bacteria (eq. 1) could accumulate, disperse, chemotax, invade cells, or be convectively transferred into the perfusion channel at the x = 0 boundary (eq. 3 middle). The number of intracellular bacteria increase because of either growth or cell invasion (eq. 2). The model was used to calculate rates of intracellular accumulation and the bacterial dispersion coefficient in tumor masses. The model was fit to the spatiotemporal profiles of intracellular bacterial density for S-Sal, Sal and F-Sal (Fig. 6A-C). The dispersion coefficient (D) was calculated to be 23.5 μm 2 /s, by fitting to the Sal data set. The dispersion coefficient did not increase when the mathematical model was fit to the F-Sal dataset. The rate of intracellular accumulation for F-Sal was 4.47 times greater than Sal, and the accumulation rate of S-Sal was 2.39 times less than Sal ( Table 2).
The model prediction of overall colonization as a function of the intracellular accumulation closely matched experimental data (Fig. 6D). When intracellular accumulation increased, overall tumor colonization increased. Theoretically extrapolating to bacteria that neither (See figure on previous page.) Fig. 5 Intratumoral flhDC induction increases colonization, dispersion and intracellular accumulation of Salmonella. a) When flhDC was induced in Salmonella, aqueous motility increased by 18% compared to uninducedSalmonella containing the same pBAD-flhDC construct (*, P<0.05). b) Graphical depiction of the dosing scheme. One hour after tumors were placed into devices, Salmonella was administered for 1 hour. Eleven hours after bacterial administration, media with 0.4% (w/v) arabinose was administered to the devices to induce bacterial flhDC expression. c) When F-Sal was administered to devices, bacteria colonies (red arrows) were first detected at 12 hours. At 47 h, colonies formed by F-Sal with intratumorally induced flhDC were larger than control Salmonella (Sal). Scale bar is 100 μm. invade nor grow intracellularly suggests that they would not colonize tumors (Fig. 6D). Based on the model, the increase in bacterial density with higher rates of intracellular accumulation occurred primarily in intermediate regions of the cell masses (0.4 ≤ x ≤ 0.6; Fig. 6E). The calculated amounts of extracellular bacteria was greater for bacteria with lower rates of intracellular accumulation (i.e. S-Sal and Sal compared to F-Sal; Fig. 6F). Based on the model, this higher extracellular density (Fig. 6F) lead to greater leakage from the tumor and a lower overall density (Fig. 6G).

Discussion
The results of this study demonstrate key mechanisms that control Salmonella colonization of tumors. Using in vitro tumors that can be monitored for bacterial infiltration and proliferation in real time, we demonstrated that overexpressing the master motility regulator, flhDC, increased tumor colonization (Fig. 1). As expected, induction of flhDC increased the motility of Salmonella, but it also increased the accumulation inside cancer cells (Fig.  2). In Salmonella with impaired invasiveness, flhDC induction did not affect colonization (Fig. 3) showing that flhDC enhances colonization by increasing the number of intracellular bacteria. Similarly, when Salmonella were modified to impair their invasiveness, tumor colonization was dramatically reduced (Fig. 4), showing that intracellular invasion and growth is important for Salmonella colonization of tumors, independent of flhDC overexpression. Integrating the spatial and temporal tumor penetration data into a mathematical model enabled calculation of the intracellular accumulation rate and showed that invasion promotes colonization by increasing bacterial retention in tumors (Fig. 6). These mechanisms could be used to improve therapeutic efficacy by enhancing bacterial tumor colonization. When flhDC was induced after initial penetration, intracellular accumulation and tumor colonization both increased (Fig. 5).
The difference between these effects shows that the major contribution of flhDC was to produce flagella, which in turn improved accumulation. The increase in accumulation of non-motile ΔflgE+pflhDC Salmonella, however, shows that flhDC control of T3SS synthesis does play a role in controlling accumulation.
Two primary mechanisms could have increased intracellular accumulation after flhDC induction: cell invasion and intracellular growth. The T3SS1 system and functional flagella are important for both. The injection of T3SS1 effectors into mammalian cells is critical for cell invasion [29]. Similarly, T3SS1 plays an important role in the escape of Salmonella from intracellular vacuoles and hyper-replication in the cellular cytoplasm [45][46][47]. In addition to T3SS, invasion could have been mediated by alternate mechanisms, such as the EGFR-dependent Rck system. The contribution of alternate mechanisms was considerably less than the T3SS system (Fig. 4B). T3SS-deficient Salmonella (S-Sal) colonized tumor masses three-fold less than T3SS-competent control bacteria (Sal; Fig. 4G), although residual intracellular accumulation (Fig. 4D) and colonization (Fig. 4G) was observed.
The intracellular niche provides Salmonella with an environment to proliferate (Additional file 3: Figure  S3B-C) and that is protected from convective clearance (Fig. 6G). In MCF7 cells in monolayers, Salmonella grew with a doubling time of 3.6 h (Additional file 3: Figure  S3C), which is considerably faster than the doubling time within tumors in mice (16.8 h) [58]. Overexpressing flhDC increased bacterial density inside cells (Fig. 2D) and in distal tumor tissue (Fig. 1D). The fact that T3SS-deficient Salmonella accumulated far less in tumor masses than control Salmonella (Sal, Fig. 4F, G) suggests that intracellular and distal tumor tissue are protected from convective clearance (Fig. 6E, F).
The mathematical model of bacterial invasion and colonization shows how intracellular accumulation would improve bacterial retention (Fig. 6). Convection continuously clears bacteria from tumor tissue located near the perfusion channel (Fig. 6F). This mechanism is analogous to convective clearance of bacteria from tumors by the bloodstream. By invading tumor cells, fewer bacteria would reside extracellularly (Fig. 6F) and fewer would be cleared (Fig. 6G). As the rate of intracellular accumulation increases, more bacteria are retained within the tumor (Fig. 6D), a mechanism similar to the 'binding' of small-molecule drugs to cancer cells [59]. With small molecule drugs, it has been shown that drug/receptor binding improves retention within tumors once the drug clears from the blood [59]. By 'binding' to cancer cells, the model suggested that Salmonella with higher rates of intracellular accumulation are less prone to leaking out of tumors (Fig. 6G).
A distally located reservoir of extracellular bacteria could serve as a continuous source for intracellular invasion and colonization of tumors. Within in vitro tumor masses, there is a considerable amount of bacterial colonization in necrotic and quiescent tissue, which is located between necrotic and actively dividing tumor tissue [7]. Of the total population of colonized bacteria, the majority of extracellular bacteria were located in necrosis (Fig. 6F). Neither intracellular nor extracellular bacteria resided in tissue near the channel because of the high rate of convective clearance (Fig. 6E, F). Due to the high dispersion coefficient, extracellular bacteria would rapidly clear out of proximal tissue close to the perfusion channel. However, extracellular bacteria residing in necrosis could grow faster than the rate of dispersion (Fig. 6F) allowing for high regional accumulation and migration to viable tissue to invade cells.
Controlling intracellular accumulation by inducing flhDC would increase tumor colonization. It would be beneficial to suppress flagellar expression outside of tumors. Flagella biosynthesis is an energetically costly process and can consume as much as 2% of bacterial energy [10,60]. In addition , Salmonella flagellin is an immunogenic agonist that facilitates accelerated bacterial clearance [61]. Inducing flhDC selectively after initial penetration into tumors would improve fitness prior to administration, while promoting invasion and colonization within tumors (Fig. 7).

Conclusion
This study demonstrates that overexpressing flhDC increases intracellular accumulation within tumor cell masses, which drives tumor colonization. Robust tumor colonization is necessary for Salmonella to be an effective drug delivery vehicle. Intracellular accumulation increased colonization by causing Salmonella to 'bind' to tumor cells. This binding prevented bacteria from being convectively cleared from tumor masses. Selectively inducing flhDC expression within tumor masses would promote fitness prior to administration and enhance colonization after initial penetration. We envision that therapeutic strains of Salmonella will utilize inducible flhDC to drive colonization in human tumors. After intravenous administration and a period of initial penetration, an inducer would be provided to activate the flhDC regulator. Intracellular invasion enables Salmonella to deliver a wide range therapies directly into the intracellular space of tumors. Measuring the mechanisms of intracellular bacterial accumulation and tumor colonization has identified a key regulator, flhDC, that could be used to amplify colonization and make Salmonella an effective anticancer therapeutic.

High Intracellular Accumulation Low Intracellular Accumulation
A B Fig. 7 Graphical depiction of how intracellular accumulation could mechanistically improve tumor colonization. a) When Salmonella have a low intracellular accumulation rate, the rate of dispersion back into tumor vasculature is high, thus reducing bacterial tumor colonization due to a lack of "binding" to cancer cells. b) When Salmonella have a high intracellular accumulation rate, more bacteria would be retained in the tumor and not leak back into tumor microvasculature, thus increasing overall tumor colonization.