Designing an effective drug delivery system for CRC therapy
Fig. 1a depicts the design principle of a synthetic probiotic that employs the P8 therapeutic protein to treat or prevent CRC. To design an anti-CRC therapeutic probiotic with enhanced stability and efficacy, we first adopted the alr complementation system that can prevent curing of the P8 expression vector, pCBT24-2 [11], in the absence of an antibiotic to maintain the plasmid. Alanine racemase is a pyridoxal 5’-phosphate-dependent enzyme involved in the interconversion of d-alanine (d-Ala) and l-alanine. d-Ala is involved in the cross-linking of the cell wall peptidoglycan layer and exists in extremely low amounts in nature. Thus, this component is essential for bacterial growth and deletion of the alr gene leads to cell death. To generate a d-Ala auxotroph of P. pentosaceus SL4(-7) that is a derivative of SL4 lacking all seven native plasmids, we performed knockout mutagenesis to remove alr from the chromosome using homolog recombination with a construct that has an in-frame deletion of the alr gene and 1 kb of its upstream or downstream flanking sequences (Additional file 1, Fig. S1a,b). The resulting auxotrophic mutant was either grown in a medium supplemented with d-Ala or complemented with a plasmid that expresses alr. Genotyping with specific primers confirmed the replacement of the intact gene (Additional file 1, Fig. S1c, Additional file 2, Table S1). This alr auxotroph complemented with plasmid-borne alr was designated as PP*.
In order to develop an effective gene expression system that can maximize the productivity of P8, four kinds of constitutive promoters involved in central glycolytic pathway: pyruvate kinase (PK), choline ABC transporter permease and substrate binding protein, glucose kinase, and l-lactate dehydrogenase. Using these promoters, we constructed five sets of dual expression systems that have two chimeric genes, each encoding the P8 peptide fused to the Usp45 secretion signal at its N terminus, which was cloned into the vector that contained alr (Fig. 1b). We then measured the concentrations of secreted P8 for each PP* clone with the dual expression module in the alr vector using ELISA to validate the PK-PK promotor system with the best stability and productivity (Fig. 1c). To further exclude the possibility that the host genotype could affect the performance of P8 secretion, we checked the concentrations of P8 secreted from the wild type P. pentosaceus SL4(-7) with the PK-PK promotor system in pCBT24-2 (PP-P8) and the Δalr mutant with the PK-PK promotor system in the alr vector (PP*-P8) and found no difference between the SL4 wild type and Δalr mutant (Additional file 1, Fig. S1d).
Anti-tumor efficacy of PP*-P8 in the DLD-1 xenograft mouse model
To determine whether PP*-P8 had anticancer activity in vivo, we assessed its efficacy using the DLD-1 xenograft mouse model. Athymic BALB/c nude mice with subcutaneous DLD-1 xenografts were treated with the commercial chemotherapy drug gemcitabine, PP* or PP*-P8 (see Materials and Methods for dose and dosage regimen), and the tumor sizes were monitored for 6 weeks before sacrifice (Fig. 2, Additional file 2, Supplementary Tables 2 and 3). Tumor growth rate was much faster in the untreated control group and the PP*-treated group than in those treated with gemcitabine or PP*-P8 (Fig. 2a). At the end of the experiment, the mean tumor volumes were 2,680.9±419.7 mm3 in the control group and 2,671.1±651.2 mm3 in the PP* group, while they were 498.6±192.7 mm3 and 1,371±349.8 mm3 in the gemcitabine and PP*-P8 treatment groups, respectively (Fig. 2a,b; control vs. PP*-P8, P = 4.9×10-5). Tumor weights were 2.13±0.31 g in the control and 2.35±0.32 mm3 in PP*, as compared to 0.39±0.16 g in gemcitabine and 0.97±0.30 g in PP*-P8 (Additional file 1, Fig. S2a; control vs. PP*-P8, P < 1×10-6). Inhibition ratios of tumor growth relative to the control were 84.1% and 50.8% in gemcitabine and PP*-P8, respectively (Fig. 2c; control vs. PP*-P8, P = 5.3×10-5). These results demonstrate that our synthetic probiotic PP*-P8 sufficiently suppressed tumor growth similar to that of an anticancer drug.
Next, we asked whether the growth inhibition of the CRC xenograft induced by PP*-P8 is due to cell cycle arrest. Western blot analysis revealed that expression of cell cycle regulatory factors Cyclin B1 and Cdk1 in tumor tissue decreased significantly in response to treatment with PP*-P8 (Fig. 2d, Additional file 1, Fig. S2b). Moreover, expression of p21, which suppresses Cyclin B1/Cdk1, increased after PP*-P8 treatment. In addition, expression of p53 also increased in the PP*-P8-treated group. Overall, the data suggest that the anticancer therapeutic protein P8 inhibits the p53-p21 signaling pathway, resulting in G2 arrest of DLD-1 cells.
PP*-P8 attenuates tumorigenesis associated with AOM/DSS-induced colitis
We also used the well-established AOM/DSS-inducible murine model for colitis-associated colon carcinogenesis to examine the anticancer effect of the synthetic probiotic PP*-P8 in situ. During the whole experimental period of 68 days, AOM was intraperitoneally injected into C57BL/6 mice on day 1, followed by three treatments of DSS administered in the animal drinking water. The mice were divided into five groups: untreated control (AOM/DSS only), gemcitabine, wild type P. pentosaceus SL4 (PP WT), PP*, and PP*-P8 (Fig. 3a; see Materials and Methods for dose and dosage regimen). Analysis of the relative abundance of the Pediococcus bacteria in the three groups indicated that bacterial populations were sustained at 0.01~0.03% (Fig. 3b; see Materials and Methods for microbial community analysis). Although the population of Pediococcus in PP WT increased during stage 1 as compared to PP* and PP*-P8, the three groups showed similar relative abundances in the subsequent two stages until the end of the experiment.
Drastic changes in the average bleeding score were observed before and after each episode of DSS administration (Fig. 3c; P = 3.12×10-2 between day 5 and 10, P = 6.40×10-7 between day 26 and 31, and P = 1.90×10-6 between day 47 and 52). The gemcitabine and PP*-P8 groups showed significantly reduced bleeding after the administration of DSS compared to the untreated (P = 8×10-6), PP WT (P = 8.68×10-2) and PP* (P = 2.42×10-4) control groups (Additional file 2, Supplementary Tables 4 and 5). Severe bleeding and bleeding around the anus were often noticeable in the controls, whereas only occult blood or slight bleeding was detected for PP*-P8. Fig. 4a as well as Supplementary Tables 4 and 5 in Additional file 2 show that DSS treatment had negative effects on weight gain in the gemcitabine and control groups, while body weight of mice in the PP*-P8 group increased until the end of the experiment. Kaplan–Meier survival curves similarly showed that, with no fatalities, PP*-P8 treatment increased the survival of AOM/DSS-treated mice during the experiment, although this increase was not statistically significant compared to the control groups (Additional file 1, Fig. S3). Colon length is one of the markers for evaluating colonic inflammation severity, and was measured after animals were euthanized to reveal that gemcitabine and the three controls had significantly decreased colon lengths in comparison to PP*-P8 (P < 1×10-6, P < 1×10-6, P = 1×10-6 for untreated, PP WT, PP*, respectively), which was indicative of severe inflammation (Fig. 4b). In comparison to the colon length of untreated control, which was administered with AOM/DSS only, the colon length of PP*-P8 exhibited close to that of the healthy mouse group, indicating that PP*-P8 treatment prevents the colon from being shortened due to the presence of AOM/DSS (Additional file 1, Table S4).
The number of nodular polypoid tumors located in the middle and distal colon in the PP*-P8 treatment group was lower than those in untreated control (P = 2.32×10-3) and PP* (P = 1.24 ×10-3) groups, while there was no significant change in PP WT (P = 0.27) (Fig. 4c,d). Taken together, these results from the AOM/DSS-induced colitis-associated cancer model indicate that the orally administered PP*-P8 probiotic effectively inhibited inflammation-associated carcinogenesis and tumor development in the colon.
PP*-P8 modulates gut microbiota to alleviate AOM/DSS-induced dysbiosis
We further explored the possible impacts of the synthetic probiotic PP*-P8 on gut microbiota in the AOM/DSS murine model for colitis-associated colon cancer. C57BL/6 mice were subjected to a dose regimen and a fecal sampling schedule that was divided into untreated control, fluorouracil, PP WT, PP*, and PP*-P8 treatment groups (Fig. 3a). Using DNA from the fecal samples, amplicon sequencing of the V3–V4 region of the 16S ribosomal RNA gene was performed to monitor microbial community structure. Processed reads were clustered into operational taxonomical units (OTUs) with a 97% threshold for sequence identity using QIIME [20] to calculate relative abundance (Additional file 2, Supplementary Tables 6 and 7).
Species richness and evenness were measured by the number of OTUs and the inverse Simpson index, respectively, to evaluate microbial diversity, which was severely disturbed by AOM/DSS treatment (Fig. 5a). As expected, all the experimental groups lost alpha diversity, which reduced the number of OTUs during each DSS administration; however, the OTUs partially recovered until the next administration. Interestingly, the PP*-P8 group seemed to restore taxonomic diversity in stage 3 better than the fluorouracil and control groups toward the end of the experiment (red lines in Fig. 5a). Principal coordinates analysis (PCoA) based on Bray-Curtis dissimilarity [21] illustrated the dissimilarities of fecal microbiota between each treatment group and pre-treated samples on day 0 and day 5 increased as stages of treatment progressed (Additional file 1, Fig. S4). The differences between the controls, fluorouracil, and PP*-P8 treatments were not obvious during stage 1 (Fig. 5b, Additional file 1, Fig. S4); however, beta diversity increased over time and permutational multivariate analysis of microbial variance resulted in significant statistical differences among the groups in stages 2 and 3 (P = 0.034 and P = 0.001, respectively). The PCoA plots also show that the three control groups became more dispersed in stages 2 and 3 than PP*-P8. It is noteworthy that fluorouracil and PP*-P8 appeared similar in stage 3 (bottom panel of Fig. 5b).
Distribution and abundance of microbial taxa for each group in each stage were examined and the results indicated that bacteria in the Bacteroidetes and Firmicutes phyla dominated the mouse gut microbiota (Additional file 2, Table S7). Relative abundance at the family level illustrated that on day 0 Muribaculaceae, Lachnospiraceae, Ruminococcaceae, and Lactobacillaceae were the main families, while during DSS administration Akkermansiaceae, Bacteroidaceae, and Erysipelotrichaceae, as well as Muribaculaceae, Lachnospiraceae, Lactobacillaceae, and Ruminococcaceae were the primary bacteria (Fig. 5c). The relative abundance of each family fluctuated and depended on the stage of the DSS treatment. When compared to the control, the most distinguishable beta diversity pattern was observed at stage 3 (Fig. 5b; P = 0.001), and the fluorouracil-treated group was enriched with Akkermansiaceae, Lachnospiraceae, and Ruminococcaceae, but depleted of Erysipelotrichaceae and Lactobacillaceae. In the PP*-P8 treatment group, Akkermansiaceae, Lachnospiraceae, and Lactobacillaceae increased, while Erysipelotrichaceae decreased compared to the controls. Our data from the AOM/DSS mouse model demonstrate that the PP*-P8 probiotic contributes to alleviating dysbiosis induced by AOM/DSS by modulating gut microbiota structure with respect to alpha and beta diversity, and the proportion of potentially beneficial taxa.
Specific bacterial taxa are associated with eubiosis maintained by PP*-P8
To determine which bacteria are most likely responsible for the differences between the treatment groups, we applied the linear discriminant analysis (LDA) and effect size (LEfSe) [22] method to calculate the LDA scores for days 56, 63, and 68 in stage 3 when the mice are recovering from the last DSS administration. The lists of taxonomic clades, ranked according to the effect size, that are differential among groups with statistical and biological significance are shown in Fig. 6a. They indicated that, between the fluorouracil and control groups, most discriminative (log10 LDA ≥ 4.0) in fluorouracil included Actinobacteria (phylum), Coriobacteriales, Bifidobacteriales, Actinobacteria (class), Bifidobacteriaceae, Bifidobacterium, and Coriobacteriaceae UCG_002, while one in the control was Turicibacter. Between PP*-P8 and the control, an uncultured Ruminococcaceae (OTU 330333), Akkermansia, Verrucomicrobiae, Verrucomicrobia, Verrucomicrobiales, Akkermansiaceae, GCA_900066575 (Lachnospiraceae), Oscillibacter, Pediococcus, Tannerellaceae, and Parabacteroides were most discriminative in PP*-P8, whereas Turicibacter, Erysipelotrichia, Erysipelotrichaceae, Erysipelotrichales, and Firmicutes were in the control. Similarly, Akkermansia, Verrucomicrobiae, Akkermansiaceae, Verrucomicrobia, Verrucomicrobiales, Oscillibacter, and an uncultured Ruminococcaceae (OTU 330333) were most differential in PP*-P8, whereas an uncultured Muribaculaceae (OTU 182112), Muribaculaceae, Bacteroidetes, Turicibacter, Erysipelotrichia, Erysipelotrichaceae, Erysipelotrichales, Bacteroidales, Bacteroidia, an unassigned Rhodospirillales, Dubosiella, and Catenibacterium were in PP*. Between PP*-P8 and fluorouracil, Bacilli at various taxonomic ranks down to Lactobacillus, Tannerellaceae, and Parabacteroides were most distinctive in PP*-P8, and an uncultured Muribaculaceae (OTU 182112), Muribaculaceae, and Actinobacteria at various ranks down to Bifidobacterium were in fluorouracil (Additional file 1, Fig. S5). LEfSe plots of OTUs between the groups at stage 3 showed a similar tendency (Additional file 1, Fig. S6) in that Akkermansia was the most discriminative genus in PP*-P8 (log10 LDA = 4.72), while Turicibacter was for the control (log10 LDA = 4.99).
Overall, the LEfSe results after the last DSS administration showed that Akkermansia and Verrucomicrobia at various ranks, to which Akkermansia belongs, followed by an uncultured Ruminococcaceae (OTU 330333) and Oscillibacter, were most characteristic of PP*-P8, and Turicibacter and Erysipelotrichia at various ranks, to which Turicibacter belongs, were characteristic of the control (Fig. 6b). Akkermansia was higher in PP*-P8 and fluorouracil than in the controls, and the uncultured Ruminococcaceae and Oscillibacter were abundant in PP*-P8 and PP WT. Turicibacter was highly enriched in the control and dramatically reduced in the other treatment groups, which was most noticeable in the PP*-P8 and fluorouracil groups. To identify the interactions between members of gut microbiota, a pairwise Spearman's rank correlation coefficient was calculated for the last three samples on days 56, 63, and 68 in stage 3 and visualized as a heat map for systematic analysis (Fig. 6c). Akkermansia, the signature taxon of the PP*-P8 group’s microbial profile, had a highly negative correlation with Turicibacter, which is a biomarker for control, and an uncultured member of Muribaculaceae (OTU 1107458). Also, two members of Muribaculaceae (OTU 270451, OTU 259609) had strong negative correlations with a member of Lactobacillus (OTU 463794) and two members of Bacteroides (OTU 4226929, OTU 513445). Another member of Muribaculaceae (OTU 322372) had a similar relationship with Lactobacillus and Bacteroides. These results suggest that specific bacterial taxa such as Akkermansia and Turicibacter are associated with eubiosis or dysbiosis, respectively, and positive or negative relationships among microbial members shape the community structure.