The CT26 (mouse colon carcinoma) and B16F10 (mouse melanoma) cell lines were purchased from the Cell Bank of Chinese Academy of Sciences Type Culture Collection. Cells were grown in RPMI 1640 (CT26) or high-glucose DMEM (B16F10) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin–streptomycin and cultured at 37°C in a humidified atmosphere of 5% CO₂. Human umbilical vein endothelial cell (HUVEC) was purchased from Mingzhoubio (MZ-0746, Ningbo, China) and grown in endothelial cell medium (ECM) supplemented with 5% FBS, 1% penicillin–streptomycin, and 1% endothelial cell growth supplements (ECGS). Cells were counted using a Fuchs-Rosenthal counting chamber and seeded into 96-well plates (1 × 10⁵ cells per well), 24-well plates (2 × 10⁵ cells per well), 6-well plates (5 × 10⁵ cells per well), 25 cm² culture dishes (5 × 10⁶ cells per dish), or 75 cm² culture dishes (2 × 10⁷ cells per dish). The culture medium was changed until cells reached 80% confluence.

Bacterial strains, plasmids, and primers used in this study are listed in Table 1 and Supplementary Table S1. The suicide plasmid-mediated mutation method was used to introduce precise scarless mutations into Salmonella as described previously with minor modifications (Edwards et al., 1998). For the deletion of purM, two pairs of primers, DpurM-1F/DpurM-1R and DpurM-2F/DpurM-2R, were used to amplify upstream and downstream DNA fragments (≈ 400 bp) of purM gene, respectively, from the genome of wild type Salmonella UK-1. These two fragments were fused via their homologous part and inserted into pYA4278 following the protocol of circular polymerase extension cloning (CPEC; Quan and Tian, 2011) to construct the desired suicide plasmid (pSS263). Then, asd⁻ E. coli χ7213 was transformed with pSS263 by electroporation and grown on LB agar plates containing chloramphenicol (25 μg/ml) and Diaminopimelic acid (DAP; 50 μg/ml). After the conjugation of E. coli χ7213 carrying pSS263 and the parent Salmonella strain, recombination of suicide plasmids into the chromosome were selected on LB agar plates supplemented with chloramphenicol (negative selection). Screened Salmonella clones were grown for 2–4 h until an OD₆₀₀ value of 0.4–0.6 was reached, diluted appropriately, and spotted on NaCl-free plates supplemented with 5% sucrose (positive selection). Colonies were replica patched onto two plates, one of which was supplemented with antibiotics used to select for the suicide plasmid. Those sensitive to the chloramphenicol antibiotic were further purified on sucrose plates and their genotypes were examined by PCR. A similar strategy was used to construct other suicide plasmids and mutant strains (Table 1).

The plasmid pYA4088, which encodes asd gene and contains bla secretion signal sequence under P_trc promoter, was used for construction of different expression plasmids. The codon-optimized cDNA of endostatin and its fusion fragments with different targeting peptides (RGD4C, RGD10 and anti-PSMA scFv) were independently cloned into pYA4088 immediately downstream bla ss, generating plasmids pYA3342-ES, pYA3342-RGD4C-ES, pYA3342-RGD10-ES, and pYA3342-PSMA-ES. The commercial plasmid PET-28(a) was engineered for the expression of another anti-tumor molecule TRAIL involved in this study. Briefly, we replaced kanamycin resistance gene with asd and used the P_BAD promoter together with araC to regulate the expression of lacI from the plasmid. The gene encoding T7 RNA polymerase with its lac UV5 promoter was cloned from BL21(DE3) and inserted into the plasmid. The codon-optimized cDNA of TRAIL was inserted under the control of T7 promoter, generating TRAIL-expression plasmid named as “pYA3342-TRAIL.” For the co-expression of endostatin and TRAIL, open reading frames (ORFs) of endostatin without or with targeting peptides together with P_trc promoter and rrnB terminator (T1T2) were, respectively, inserted into pYA3342-TRAIL between asd and T7 terminator. Finally, empty pYA3342 and different expression plasmids (asd⁺) were transformed into the engineered Salmonella bacterial strain, of which asd was deleted. In this expression system, P_BAD promoter is used to regulate the expression of the repressor LacI. Since the promoter of T7 RNA polymerase, T7 promoter and P_trc promoter contain lac operator sequence, none of these promoters will be activated when bacteria are grown with the supply of sufficient arabinose. In the absence of arabinose or in in-vivo conditions, both T7 promoter and P_trc promoter will be activated to initiate the transcription of downstream genes encoding anti-tumor molecules because of no repressor expressed.

All bacterial strains were cultured on LB agar plates or in LB broth containing appropriate supplements and/or antibiotics. One single clone of bacterial strains was picked, inoculated into LB broth, and grown overnight in a shaking incubator (37°C, 180 rpm). Next day, the overnight culture was diluted 100-fold into fresh medium and grew to exponential phase with an optical density value at 600 nm (OD₆₀₀) of 0.8–0.9. Bacterial cells were then harvested by centrifugation (4,000 × g for 10 min), washed with phosphate buffer saline (PBS), and diluted to obtain desired concentration in an appropriate volume for in vitro and in vivo experiments. The concentration of bacterial cells was calculated as follows: 0.8 OD₆₀₀ ≈ 1.0 × 10⁹ CFU.

The phenotypes of bacterial strains were determined in vitro, and all assays were repeated at least twice. The growth of bacterial strains in LB broth (37°C, 180 rpm) was measured every hour, with an initial OD₆₀₀ value of less than 0.05. Bacterial swimming motility was assessed on LB plates solidified with 0.3% agar (wt/vol) as described previously (Kong et al., 2011c). In brief, 5 μl of bacterial suspension (approximately 5 × 10⁶ CFU) was inoculated into the middle of semi-solid plates and subsequently incubated at 37°C for 8 h. The swimming diameter was measured with a ruler. Outer membrane proteins (OMPs) were purified from bacterial strains as described previously (Kang et al., 2002), subjected to sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and stained using coomassie brilliant blue. The lipopolysaccharide (LPS) profile of Salmonella bacterial strains was determined by silver staining following the method of Hitchcock and Brown (Hitchcock and Brown, 1983).

The invasion assay was performed for testing bacterial infection in cancer cells as described previously (Liang et al., 2018). Cancer cells were seeded into 24-well plates 16 h prior to infection to obtain a density of 5 × 10⁵ cells per well. A total of 5 × 10⁷ CFU of Salmonella bacteria prepared as described above was then added to achieve the desired multiplicity of infection (MOI) of 100:1, and the mixture was incubated at 37°C under 5% CO₂ for 2 h. After washing three times with PBS, cancer cell line-optimal medium containing gentamycin (200 μg/ml; Sigma) was added for additional 1 h of incubation to kill extracellular bacteria. Intracellular bacteria were harvested after washing and extraction with the lysis buffer (0.05% Triton X-100 diluted in PBS). Finally, the diluted lysate was plated onto LB plates and the plates were incubated at 37°C overnight for counting alive bacterial cells (CFU).

The indirect immunofluorescence assay was performed to further confirm the invasion of Salmonella bacteria into cancer cells. Cells were seeded into 6-well plates containing coverslips and cultured for 16 h. Prepared bacterial strains were washed twice with PBS, diluted in cell culture medium without antibiotics, and added to cancer cells at a ratio of 100:1. After co-incubation for 2 h and killing of extracellular bacteria by gentamycin, cells were washed gently with PBS and fixed in 4% paraformaldehyde. Then, fixed cells were permeabilized by 0.1% Triton X-100 and incubated with the rabbit anti-Salmonella polyclonal antibody (ab156656, Abcam) for 2 h at room temperature (RT) or overnight at 4°C. After washing, the goat anti-rabbit IgG polyclonal antibody conjugated with Alexa Fluor 488 (ab150077, Abcam) was added for 1 h of incubation at RT. Cytopainter phalloidin-iFluor 555 Reagent (ab176756, Abcam) and DAPI (R37606, Invitrogen) was used to indicate cell boundaries and nuclei, respectively. Finally, cell slides were observed and photographed under an inverted fluorescence microscope (Olympus, IX73P2F, Tokyo, Japan).

To test the cell-killing activity of Salmonella, cell counting kit-8 (CCK-8) assay was performed. Cells were seeded in 96-well plates and incubated overnight to reach about 80% confluence. After washing twice with PBS, Salmonella bacteria resuspended in fresh medium was added for co-incubation with cells at different MOI. After 6 h of incubation, cells were washed gently to remove bacteria. Then, 10 μl of CCK-8 solution (Solarbio, CA1210) per well was added and cells were incubated for additional 2 h. Then, the absorbance was measured at 450 nm with a microplate reader. The cell viability was expressed as the percentage of viable cells compared to the untreated group.

The expression of anti-tumor molecules endostatin and/or TRAIL by engineered Salmonella bacteria was analyzed by western blotting. When freshly cultured bacteria grew to an OD₆₀₀ value of 1.0, 2 ml of bacterial culture was taken. Bacterial pellets were obtained after centrifugation (12,000 × g for 10 min) and proteins present in the supernatant medium were precipitated by trichloroacetic acid (TCA; Sanchez et al., 2009). Bacterial pellets and concentrated supernatant proteins were boiled in 200 μl loading buffer for 10–15 min, and 10 μl of samples were taken for separation by 12% SDS-PAGE gel. Proteins were then transferred to 0.22 μm nitrocellulose membranes using semi-dry electrophoretic transfer cell (Bio-Rad, China). The membranes were probed with mouse anti-endostatin (MA1-40230, Themofisher) or anti-TRAIL monoclonal antibody (ab2219, Abcam), followed by a horseradish peroxidase (HRP)-conjugated anti-mouse IgG secondary antibody (1030-05, Southern Biotech). Immunoreactive proteins were detected using enhanced chemiluminescence (ECL) substrate (BL520A, Biosharp) and visualized by the GelDoc XR+ imaging system (Bio-Rad, China).

Seven-week-old female BALB/c and C57BL/6 mice (≈20 g) were purchased from Dashuo Biotechnology Co., Ltd. (Chengdu, China). Animal care, experiments, and euthanasia were performed following the principles stated in the Guide for the Care and Use of Laboratory Animals. All efforts were made to minimize animal suffering during the experiments. All mice were acclimated for 7 days after arrival before experiments started.

The CT26 and B16F10 cell lines were used to establish the colon carcinoma model in BALB/c mice and the melanoma model in C57BL/6 mice, respectively. Briefly, CT26 (5 × 10⁵) or B16F10 cells (2 × 10⁵) suspended in 100 μl of PBS was injected subcutaneously into the right back of each mouse.

The virulence of Salmonella bacterial strains were tested in BALB/c mice as previously described (Kong et al., 2011c). Freshly grown bacteria were harvested, washed, and diluted to the required inoculum density in phosphate-buffered saline containing 0.01% gelatin (BSG). Groups of four mice each were infected orally with various doses of bacterial strains in a volume of 20 μl, ranging from 1 × 10³ to 1 × 10⁹ CFU. Then, mice were monitored for 4 weeks post infection and deaths were recorded daily. Experiments were repeated twice. The LD₅₀s of different strains were calculated with the software SPSS.

In the colonization experiment, subcutaneous colon carcinoma model was established in BALB/c mice. When tumor volume reached about 200 mm³, each mouse was infected intraperitoneally (i.p) with 100 μl of PBS or bacterial suspension containing 5 × 10⁶ CFU (Day 0). Tumor-bearing mice were then euthanized at the indicated dpi (1, 3, 7, 14, and 21) and tissues of spleen, liver, and tumor were taken to determinate bacterial burden. Tissue samples were homogenized and diluted in PBS, and dilutions of 10¹–10⁷ (depending on the tissues) were plated onto LB plates to count the CFU of viable bacteria. Bacterial burden within normal and tumor tissues was expressed as CFU g⁻¹ tissue. The experiment was repeated twice.

The potential therapeutic anti-tumor effects of different S. Typhimurium bacterial strains and recombinant bacteria D2-asd carrying empty or different expression plasmids were evaluated in mouse models of colon carcinoma and melanoma. When tumor masses were visible, tumor-bearing mice were randomly divided into groups (termed as “PBS,” “ST/3342,” “ST/ES,” “ST/TRAIL,” etc.) of eight mice each and injected with 100 μl of bacterial suspension (5 × 10⁶ CFU) or PBS intraperitoneally. Tumor volume (mm³) was measured with a caliper every 2–3 days and estimated using the formula (L × W² × π)/6, where L and W represent the length and width of the tumor, respectively. Mice with CT26 tumors exceeding 2,500 mm³ were scheduled for euthanasia. For the melanoma model, tumor volume and the survival were recorded until mice died. Meanwhile, body weight of mice was measured as an indicator for general health status. Experiments were performed twice.

During animal experiments, three mice of each group were euthanatized at 14 dpi, and tumor and normal tissues were taken and fixed immediately with 4% paraformaldehyde for IHC and pathological studies. Standard hematoxylin and eosin (H. E) staining of paraffin-embedded tissues was performed for pathological examination. For the IHC staining, heat-induced antigen retrieval was performed at temperature > 95°C in 10 mM sodium citrate buffer (pH 6.0). Endogenous peroxidase activity was quenched by incubating the sections with 3% hydrogen peroxide for 10 min. After that, the sections were blocked with blocking buffer containing 0.1% Triton X-100, 3% BSA, and 2% normal donkey serum for 1 h, and incubated with following primary antibodies: anti-Salmonella (ab156656, Abcam), anti-endostatin (MA1-40230, Themo fisher), anti-TRAIL (ab2219, Abcam), anti-CD34 (ab198395, Abcam), and anti-caspase 3 (ab13847, Abcam; which recognizes a cleaved form of caspase-3 produced after apoptosis has been induced). After washing with PBS, HRP-conjugated secondary antibodies were added. Then, the sections were stained with a freshly prepared 3, 3′-diaminobenzidine (DAB) chromogen and counterstained with hematoxylin. Photos were taken in five random fields of each sample. The integrated optical density (IOD) of positive staining was analyzed by the software Image-Pro Plus 6.0.

Numerical data were expressed as means ± SEM if not stated. One-way or two-way ANOVA analysis followed by Tukey’s multiple comparisons test was used to evaluate the difference significance of bacterial motility and invasion, cell viability, tumor volume, tumor and spleen weight, and the mean IOD of IHC staining among groups. The Kaplan–Meier survival curve was used for monitoring mouse survival and difference significance was analyzed by the log-rank test. Data analysis was performed using GraphPad Prism 8.0. p < 0.05 was considered as statistically significant (*, #, or †); p < 0.01 as very significant (**, ##, or ††); and p < 0.001 as extremely significant (***, ###, or †††).

In this study, S. Typhimurium strains 14028, SL1344, LT-2, UK-1, and SL7207 were indicated by A, B, C, D and E for short, respectively. First, we compared different S. Typhimurium bacterial strains in their in vitro phenotypes. The growth of these bacterial strains was tested in LB broth (37°C, 180 rpm) and no obvious differences were observed (Supplementary Figure S1). All strains grew to the stationary phase after 10 h, showing similar OD₆₀₀ values (≈1.5) of their cultures. Surprisingly, the strain LT-2 (C) showed decreased motility by about one-third when compared to other wild type strains, which was even weaker than the auxotrophic strain SL7207 (E), as indicated by the swimming diameter of bacteria on semi-solid agar plates (Figure 1A). OMPs purified from these strains were stained on SDS-PAGE gel using coomassie brilliant blue and similar bands were shown except that one more protein of about 60 kDa was present in LT-2 (or this protein was expressed in higher abundance; Figure 1B). There were no obvious differences among the LPS profile of these strains detected by silver staining (Supplementary Figure S1B). The interaction between Salmonella bacteria and cancer cells of CT26 and B16F10 cell lines was also imitated in vitro. It was shown that the strains 14028, SL1344, and UK-1 had comparable ability to invade cancer cells while the invasion of LT-2 was significantly weaker than the above three (Figure 1C). The immunofluorescence assay further confirmed the invasion of Salmonella bacteria into cancer cells (Figure 1D). Consistently, LT-2 showed the lowest toxicity to cancer cells and other wild-type strains decreased the viability of cancer cells to the similar extent (Figure 1E).

The virulence of different bacterial strains was tested in BALB/c mice (Table 2). Consistent with previous studies, the virulence of LT-2, which is a major strain for cellular and molecular biology research in Salmonella (Swords et al., 1997) is low with the oral LD₅₀ of about 1.0 × 10⁷ CFU, which was at least 100 times those of other three wild type strains (A, B, and D; Luo et al., 2012). Besides, the oral LD₅₀ of the auxotrophic strain SL7207 (E) was larger than 1.0 × 10⁹ CFU, indicating that mutation of aroA was effective in attenuation of bacterial virulence.

Colonization experiments were conducted to test the tumor-targeting ability of Salmonella bacteria in the subcutaneous model of colon carcinoma. For the attenuation of bacterial virulence, we introduced the ∆aroA mutation into different wild type strains, respectively, and generated mutants auxotrophic for aromatic amino acids (AAA) including tyrosine, phenylalanine, and tryptophan. It was also reported that the deletion of aroA increased the immunostimulatory capacity of Salmonella, an important part of its inherent anti-tumor activity (Felgner et al., 2016a). In addition, the gene purM responsible for purine synthesis was also deleted to avoid the possibility of virulence recovery. Finally, constructed ∆aroA ∆purM mutants (A1, B1, C1, and D1) could grow in LB medium but could not grow in minimal salt medium without the supply of AAA and adenine (Supplementary Figure S2A). On the 1st, 3rd, 7th, 14th, and 21st day after tumor-bearing mice were intraperitoneally injected with Salmonella mutants, tissues including tumor, liver, and spleen were picked up to determine bacterial burden. As shown in Figure 2, at dpi 1, 1 × 10⁵ and 1× 10⁶ CFU/g of Salmonella bacteria were detected in liver and spleen tissues, respectively, and about 1 × 10⁸ CFU/g in the tumor. Then, the number of colonized bacteria was gradually reduced in normal tissues but remained relatively stable in the tumor (Figures 2A–C). At least within 4 weeks after infection, all these auxotrophic strains preferentially accumulated in the tumor relative to normal tissues, with tumor-to-normal tissue ratios ranging from 100: 1 to over 100,000: 1 (Figures 2D,E). Especially, C1 (LT-2-∆aroA ∆purM) was cleared from normal tissues more quickly than other auxotrophic mutants including A1, B1, D1, and E, showing significantly lower number of colonized bacteria 2 weeks post infection (Figures 2A,B).

We next evaluated the native anti-tumor efficacy of these auxotrophic strains in mice (Figure 3). It was shown that auxotrophic strains of 14028, SL1344, and UK-1 (A1, B1, and D1) possessed similar anti-tumor ability, which is significantly stronger than LT-2-∆aroA ∆purM (C1) in terms of inhibition of tumor growth (p < 0.5 or 0.01). Moreover, 14028-∆aroA ∆purM (A1), UK-1-∆aroA ∆purM (D1) and SL7207 (E) obviously improved the survival of melenoma-bearing mice when compared to PBS treatment (33.5, 36.5 and 34, vs. 20 days). The median survival time of C1-treated mice was about 22 days, which was almost the same as PBS group.

Considering the importance of bacterial motility, invasion ability, virulence, and native anti-tumor activity to Salmonella-mediated cancer therapy, we finally chose UK-1 from four wild-type strains mentioned above for further studies.

In this study, we attempted to utilize live attenuated Salmonella bacteria to deliver therapeutic anti-tumor agents. Therefore, we further optimized selected auxotrophic strain D1 (UK-1-∆aroA ∆purM) through introduction of the three mutations ∆pagP ∆pagL ∆lpxR and generated a new mutant strain named D2 (UK-1-∆aroA ∆purM ∆pagP ∆pagL ∆lpxR). D2 was expected to synthesize homogeneous hexa-acylated lipid A, which has been considered to possess the most potent immunomodulatory activity, and to reverse the immunosuppression in the tumor microenvironment (Teghanemt et al., 2005; Liang et al., 2019). The phenotypes of auxotrophic strain D1 and its derived lipid A-modified strain D2 were identified together with their wild-type strain UK-1 (D; Supplementary Figures S2, S3). It showed that ∆aroA and ∆purM mutations resulted in the inability of Salmonella to synthesize AAA and adenine (Supplementary Figure S2A) and that generated auxotrophic strains D1 and D2 grew more slowly than the wild-type strain UK-1 in LB broth (Supplementary Figure S2B). The motility of D1 and D2 was decreased as well, according to their swimming on semi-solid agar plates (5.0, 5.0, vs. 6.0 cm; both p < 0.05; Supplementary Figure S3A). Related gene mutations did not seem to obviously influence the OMP profile of Salmonella bacteria (Supplementary Figure S3B). The LPS staining indicated that the structure modification of lipid A was achieved by ∆pagP ∆pagL ∆lpxR mutations as well. As shown, the homogeneous ladder of LPS synthesized by the strain D2 was obviously different from other strains (Supplementary Figure S3C). The ability of auxotrophic strains (D1 and D2) to invade and to kill cancer cells was both significantly weaker than D (Supplementary Figures S3D,E). Compared to the wild type strain, D2 was significantly attenuated in mice. After oral administration to mice, D2 was avirulent even at a dose higher than 1.0 × 10⁹ CFU. Upon intraperitoneal infection, the LD₅₀ of D2 was more than 1.0 × 10⁷ CFU, while that of UK-1 was only about 1.0 × 10² CFU (Table 3).

Then, we constructed several plasmids (asd⁺) for expressing anti-tumor molecules including potent angiogenesis inhibitor endostatin (ES) and/or apoptosis inducer TRAIL (Table 1), and transformed them and the empty plasmid, respectively, into the ∆asd derivative of D2 (D2-asd, named as “ST” for short when plasmids were carried), where plasmids and bacterial strains constitute a balanced-lethal system to ensure the plasmid stability (Wang et al., 2013). In this study, the β-lactamase signal sequence (bla ss) was added to the N-terminus of anti-tumor molecules for the secretion of them by Salmonella bacteria. Besides, to increase the specificity of endostatin’s anti-angiogenic anti-tumor activity, endostatin was linked to short RGD (arginine-glycine-aspartate) peptides or anti-prostate specific membrane antigen (PSMA) single-chain variable fragment (scFv), respectively, both of which could bind to overexpressed surface proteins of proliferating endothelial cells with high affinity and selectivity, i.e., the αvβ3 integrin and PSMA (Liu et al., 1997; Weis and Cheresh, 2011; Danhier et al., 2012; Afshar-Oromieh et al., 2016). The expression and secretion of ES and TRAIL by Salmonella bacteria was confirmed by western blotting (Figures 4A–C; Supplementary Figure S4C). Subsequently, the toxicity of recombinant Salmonella to the cancer cells or HUVECs was detected by the CCK-8 assay. ES-expressing Salmonella was shown to cause lower viability of HUVECs when compared with bacteria carrying the empty plasmid, so did their filtered culture medium (Figure 4D). Salmonella expressing TRAIL exerted higher toxicity to cancer cells of CT26 and B16F10 cell lines (Figure 4E). Moreover, when treated with Salmonella carrying ES-TRAIL co-expression plasmids, both HUVECs and cancer cells could be killed (Figure 4F).

Animal experiments were performed to study the potential therapeutic effects elicited by our engineered Salmonella bacteria (D2-asd) and its delivered anti-tumor molecules (ES and TRAIL) mentioned above. When tumor masses were visible, tumor-bearing mice were grouped randomly and intraperitoneally received 5 × 10⁶ CFU of ST/3342, ST/ES, ST/RGD4C-ES, ST/RGD10-ES or ST/PSMA-ES, or 100 μl of PBS. In the CT26 colon carcinoma model, infection of Salmonella bacteria carrying different ES expression plasmids significantly delayed tumor growth compared to the treatment with PBS (Figure 5A). At 14 dpi, the mean tumor volume of PBS group was about 2,400 mm³, whereas those of other groups were about 1900 mm³ (ST/3342, ns, vs. PBS group), 1,500 mm³ (ST/ES, p < 0.01), 900 mm³ (ST/RGD4C-ES, p < 0.001), 1,100 mm³ (ST/RGD10-ES, p < 0.001), and 900 mm³ (ST/PSMA-ES, p < 0.001), respectively (Figure 5A). Engineered attenuated Salmonella carrying expression plasmids of endostatin linked with RGD peptides or anti-PSMA scFv (ST/RGD4C-ES, RGD10-ES, and PSMA-ES) showed significantly stronger suppressive effects on tumor growth compared to ST/3342 (p < 0.05 or 0.01, data not shown). These recombinant strains seemed to cause stronger suppression of tumor growth than ST/ES, although the differences were not significant statistically. Tumor tissue was also weighed after mice were sacrificed. The comparison of tumor weight change among different groups was similar with that of tumor volume suppression (Supplementary Figure S5A).

Then, we established the aggressive B16F10 melanoma model to further evaluate the anti-tumor efficacy of our engineered Salmonella bacteria. Similarly, mice infected by Salmonella carrying ES expression plasmids showed significantly superior therapeutic benefits compared to those treated with PBS or ST/3342 at dpi 14, 17, and 20 (p < 0.001; Figure 5B). After 2 weeks, the mean tumor volume of PBS group was about 6,600 mm³, whereas those of Salmonella infection groups were about 3,600 mm³ (ST/3342), 2,400 mm³ (ST/ES), 1,100 mm³ (ST/RGD4C-ES), 1,300 mm³ (ST/RGD10-ES), and 1,300 mm³ (ST/PSMA-ES), respectively. Moreover, the life of melanoma-bearing mice was significantly prolonged by treatment with Salmonella expressing endostatin linked with different targeting peptides, as shown in Figure 5. In detail, tumor-bearing mice treated with D2-asd carrying RGD4C-ES, RGD10-ES, or PSMA-ES survived for over 30 days, followed by mice treated with ST/ES for 23 days, whereas the median survival time of ST/3342- and PBS-treated mice was 20 and 17 days, respectively (Figure 5D).

Previous studies have shown that one single anti-tumor agent may not be enough to cause sustained tumor suppression. Targeting different signal pathways may synergistically produce stronger anti-tumor effects. Therefore, we attempted to use the optimized Salmonella vector to deliver the apoptosis inducer TRAIL, of which the anti-tumor efficacy was also tested in this study (Supplementary Figures S4A,B), in addition to anti-angiogenic endostatin and expected that both cancer cells and proliferating endothelial cells of tumor vasculature could be targeted. The co-expression plasmid of TRAIL and endostatin linked with RGD4C peptide was constructed and transformed into the strain D2-asd. The anti-tumor efficacy of newly generated recombinant Salmonella strain ST/TRAIL-RGD4C-ES was then evaluated in mouse models of colon carcinoma and melanoma. The results showed that significant inhibition of tumor growth and prolongation of mouse survival were caused by treatment of ST/TRAIL-RGD4C-ES or ST/TRAIL, compared to PBS and ST/3342 (Figures 6A–D; Supplementary Figure S5C). Surprisingly, significantly different anti-tumor effects were not observed between ST/TRAIL-RGD4C-ES and ST/TRAIL. The mean tumor volume of melanoma-bearing mice treated with PBS or ST/3342 was more than 3,000 mm³, while those of ST/TRAIL and ST/TRAIL-RGD4C-ES treated mice were only 2,100 mm³ and 1,500 mm³, respectively. The median survival time of mice was increased to 19, 27, and 29 days, respectively, by treatment with Salmonella carrying empty, TRAIL-expressing and TRAIL and RGD4C-ES co-expression plasmids, compared to that of PBS group (14 days).

During animal experiments, tissue samples were also taken for studying the in vivo anti-tumor mechanisms and adverse toxic-side effects of our engineered Salmonella through IHC and H.E staining. Consistent with the results of colonization experiments, attenuated Salmonella carrying empty or different expression plasmids could accumulate inside tumor tissue after systemic infection, which was indicated by staining for Salmonella O-antigen. The anti-tumor molecules released by Salmonella were also detected there (Figure 7; Supplementary Figures S6, S7).

Since we aimed to utilize optimized live Salmonella to mediate cancer therapy combining suppression of angiogenesis and induction of cell apoptosis, Salmonella was equipped with plasmid-encoded endostatin and/or TRAIL. To test whether these two molecules and Salmonella itself exerted their anti-tumor activities, staining was conducted on tumor tissue for CD34 and cleaved caspase-3, two indicators, respectively, for angiogenesis and cell apoptosis (Fina et al., 1990; Porter and Jänicke, 1999). Compared with ST/3342 and PBS treatment groups, the ST/TRAIL and ST/TRAIL-RGD4C-ES groups showed significantly higher level of cell apoptosis within tumor tissue, which was indicated by the increase of activated caspase-3 (Figure 7; Supplementary Figure S7B). The expression level of CD34, a surface marker of proliferating vascular endothelial cells, was significantly decreased inside tumor tissue infected by ST/TRAIL-RGD4C-ES compared to other treatment groups (Figure 7; Supplementary Figure S7B). Similar results were also observed when mice were infected by Salmonella expressing different forms of endostatin (Supplementary Figures S6, S7A). In detail, bacterial delivery of endostatin linked with RGD or PSMA scFv targeting peptides was shown to significantly decrease CD34 expression within the tumor, compared to treatment with PBS or Salmonella only. Besides, it showed that Salmonella possessed native anti-tumor activities targeting angiogenesis and cancer cells even though it carried the empty plasmid. Taken together, these results demonstrated that the accumulation of Salmonella within tumor tissue and local release of TRAIL and endostatin led to apoptosis induction and angiogenesis inhibition, which at least partly explained the superior therapeutic benefits elicited by Salmonella co-expressing TRAIL and RGD4C-ES.

During animal experiments, we also monitored the general health of tumor-bearing mice through recording their body weight. The body weight of mice was obviously decreased by about 10–15% in the initial several days after i.p infection of Salmonella and gradually recovered 1 week later (Figures 3, 5, 6). It seems plausible that our engineered attenuated Salmonella (i.e., D2-asd carrying different asd⁺ plasmids) indeed elicited toxic-side effects on tumor-bearing mice at the early infection stage, while the administration dose adopted in this study (5 × 10⁶ CFU for each mouse) was tolerable for mice. Meanwhile, we also took susceptible normal tissues including liver and spleen for pathological studies. The swelling and degeneration of hepatocytes in the liver and the inflammation in the spleen were observed after H.E staining, indicating that i.p infection of Salmonella bacteria caused pathological changes on these tissues (Figure 8). Besides, splenomegaly was obvious in mice infected by Salmonella (Supplementary Figures S5B,D).