Oxaliplatin and Calcium Folinate Injection were both obtained from Aosaikang (Jiangsu, China), and 5-FU Injection was obtained from SunRise (Shanghai, China). Oxaliplatin and FOLFOX Injection were prepared according to clinical guidelines as well as existing studies (Iida et al., 2013; Yang et al., 2016). Chemicals including O-Methoxyamine hydrochloride, N-methyl-N-trifluoroacetamide (MSTFA) and cortisone acetate were purchased from Sigma–Aldrich (St.Louis, MO, United States). Vancomycin (MB1260), Ampicillin (MB1507), Neomycin sulfate (MB1716) and Metronidazole (MB2200) were purchased from Meilunbio (Dalian, China).

Mice colon cancer cell line CT-26 was purchased from the Cell Bank of the Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences (Shanghai, China). The cells were cultured in RPMI-1640 (Gibco, Grand Island, United States) supplied with 10% Fetal Bovine Serum (Gibco, Grand Island, United States) in a humidified atmosphere with 5% CO₂ at 37°C.

Five to six-week-old male BALB/c mice were provided by Beijing Vital River Laboratory Animal Technology Co. Ltd. (License No. SCXK 2019-0001). The mice were housed in a temperature-controlled environment (24 ± 2°C) under a 12/12 h-dark/light cycle. The study was conducted in accordance with the Guide for the Care and Use of Laboratory Animals and approved by the Animals Ethics Committee of China Pharmaceutical University (License No: SYXK 2018-0019).

After acclimation for 1 week, approximately 1×10⁶ CT-26 cells were injected subcutaneously into the flank of mice. When the tumors reached to about 100 mm³, mice were randomly allocated to one of the groups (day 0), i.e., model for oxaliplatin (MO, n = 8), oxaliplatin treatment (Oxaliplatin, n = 27), model for FOLFOX (MF, n = 6), FOLFOX treatment (FOLFOX, n = 33). Based on previous studies, oxaliplatin (10 mg/kg) was intraperitoneally administrated twice a week (Yang et al., 2016), and FOLFOX (oxaliplatin 6mg/kg, 2 h after 5-FU 50 mg/kg and Calcium Folinate 90m g/kg treatment) was intraperitoneally administrated once a week (Robinson et al., 2013; Limani et al., 2016). MO and MF individuals were treated with corresponding vehicles. Individuals in the control group (n = 7) received neither tumor cell injection nor drug treatment. Tumor volume was monitored by a Vernier Caliper throughout the whole experimental period. All the mice were sacrificed at the end of the experiment (day 12), fecal samples were collected for 16S rDNA gene sequencing analysis and non-target metabolomics analysis.

Tumor volume (TV), Relative tumor volume (RTV) and inhibition rate were calculated by the following formulas: TV  ( m m 3 )  =  A 2 × B 2 , where A represents the longest diameter of tumor, and B represents the shortest diameter; RTV =  V t V 0 , where V₀ represents the tumor volume of day 0 (the day of first oxaliplatin administration), V_t represents the tumor volume of day t; Inhibition rate by Relative Tumor Volume (%) = (1−RTV_t /RTV_m) × 100%, where RTV_m represents the RTV of model group, and RTV_t represents the RTV of treatment group.

Total bacterial DNA was isolated from fecal with Stool Genomic DNA Kit (CWBIO, Beijing, China) according to the manufacturer’s instructions. DNA quantification was conducted by a NanoDrop 2,000 (Thermo Fisher Scientific, Waltham, United States).

The DNA integrity was checked by 1% agarose gels electrophoresis. PCR amplification was performed spanning the V3-V4 hypervariable regions of the bacterial 16S ribosomal RNA gene with the conventional barcoded universal bacterial primers 338F (5′-ACT​CCT​ACG​GGA​GGC​AGC​AG-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3′), and sequenced with an Illumina Hiseq 2500 platform (Illumina, San Diego, United States) (Holm et al., 2019; Wan et al., 2019; Chen X. et al., 2020). Raw fastq files were filtered by the Quantitative Insights in the Microbial Ecology software. High-quality sequences were clustered into Operational Taxonomic Units (OTUs) with the similarity threshold of 97% by USEARCH UPARSE (Feng et al., 2019). Then, OTUs were classified into kingdom, phylum, class, order, family, and genus levels referring to the Greengenes database (Shikany et al., 2019), and eventually an OTU table was created. The parameter α-Diversity (Chao1/Shannon/Simpson) was used to reflect the bacterial gene diversity. Wilcoxon rank-sum test was applied to identify differential taxa between groups (taxa with p < 0.1 was screened). Bacteria that existed in less than 50% samples were excluded (Feng et al., 2019). Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PICRUSt) was applied to predict the potential function of microbial communities based on the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway database as previously reported (Park et al., 2020). Data concerning the samples included in this study are deposited in the NCBI BioProject database under BioProject accession number PRJNA706146.

Akkermansia muciniphila (A. muciniphila, ATCC BAA-835) was purchased from American Type Culture Collection (Manassas, VA, United States), the bacterium was cultured in Brain Heart Infusion Agar at 37°C under anaerobic condition.

A broad-spectrum antibiotics mixture namely ABX consisting of Vancomycin (100 mg/kg), Neomycin sulfate (200 mg/kg), Metronidazole (200 mg/kg) and Ampicillin (200 mg/kg) was intragastrically administrated to mice every day for five consecutive days (day −14 to −9) to deplete gut microbiota and decrease its α-Diversity (Gong et al., 2019). The mice were then treated with A. muciniphila by gavage at 1 × 10⁸ colony forming unit (cfu)/mouse every other day until the end of the experiment (day −7 to 12). Fecal samples were collected at day −7 and 0 for transplantation efficiency verification. CT-26 cells (approximately 1 × 10⁶) were injected subcutaneously into the flank of mice at day −7. When the tumor volumes reached about 100 mm³, mice were randomly allocated to one of the following groups and the day was marked as day 0: Model (n = 10), FOLFOX (n = 10), ABX (n = 10), ABX-FOLFOX (n = 10), ABX-Akk (n = 10), ABX-Akk-FOLFOX (n = 10). FOLFOX was administrated once a week (day 2 and 9). Tumor volume was monitored by a Vernier Caliper throughout the experiment. All the mice were sacrificed at the end of the experiment (day 12), tumors were removed and prepared for immunohistochemistry analysis.

Relative levels of A. muciniphila were quantified by qPCR (Tsoi et al., 2017). Total bacterial genome DNA isolation and quantification were conducted as mentioned above. Then, qPCR was performed using SYBR Green Ι Master (Roche Diagnostics, Basel, Switzerland) on a LightCycler 480 instrument (Roche) following the manufacturer’s instructions. The levels of A. muciniphila were calculated according to the 2^–ΔΔCT method (Wang et al., 2021). Information on PCR primers was provided in Supplementary Table S1.

Tumors were fixed in formalin and embedded in paraffin. Sections were then subjected for hematoxylin and eosin (HE) staining and Ki67 immunohistochemistry detection as previously reported (Hou et al., 2018).

Methods for fecal sample non-target metabolomics analysis were presented in our previous studies (Zhang et al., 2017a; Gao et al., 2019). Briefly, gas chromatography-mass spectrometry (GC-MS) analysis was performed on a GC-MS-QP2010 Ultra (Shimadzu Inc., Kyoto, Japan) with an Rtx-5MS capillary column (30.0 m × 0.25 mm × 0.25 μm). Liquid chromatography-mass spectrometry (LC-MS) detection was carried out on an ultra-flow liquid chromatography system coupled with ion trap/time-off light hybrid mass spectrometry (UFLC-IT/TOF-MS, Shimadzu Inc., Kyoto, Japan) and compounds were separated by a Phenomenex Kinetex C18 column (100 × 2.1 mm, 2.6 μm). Details on sample preparation, instrument parameters, metabolite annotation, quality control, and data analysis were provided in the supporting information.

Spearman’s correlation analysis was applied to test the correlation between fecal bacterial abundance and pharmacodynamic indices (IR, RTV, TV, and body weight) or metabolite intensities. Data analysis and graph plotting were performed by GraphPad Prism 8 software (GraphPad Software Inc., La Jolla, CA, United States). The results were presented as mean ± SD. Independent unpaired two-tailed Student’s t test was performed to evaluate the differences between two groups, unless elsewhere specified.

In this study, a colon cancer xenograft model was constructed to evaluate the anti-cancer effect of oxaliplatin and FOLFOX (Figure 1A). Consistent with clinical practice and existing literature (Cvitkovic and Bekradda, 1999; Hoff et al., 2012), oxaliplatin caused more severe adverse effects than FOLFOX, manifested by a larger body weight reduction (Figure 1B). Moreover, the tumor development was more significantly inhibited with FOLFOX treatment compared to oxaliplatin throughout the whole experiment (Figures 1C,D). At the end of the experiment (day 12), significant difference in tumor inhibition rate (IR) (p < 0.01) was observed between the FOLFOX and oxaliplatin treated individuals (Figure 1E). Taken together, FOLFOX exhibited a superior therapeutic effect than oxaliplatin alone in colon cancer xenograft mice.

Many studies have illustrated important roles of gut microbiota in mediating chemotherapy efficacy (Roy and Trinchieri, 2017), primarily through affecting drug biotransformation directly and interacting with the host indirectly (Vivarelli et al., 2019; Zimmermann et al., 2019). To investigate the underlying mechanism for the superior therapeutic effect of FOLFOX, fecal samples were collected at the end of the experiment and all the samples were subjected to 16S rDNA gene sequencing analysis.

As is shown in Figures 2A,B, there was no significant difference in the α-Diversity for the model, FOLFOX, or oxaliplatin individuals compared to the controls indicated by Chao1, Shannon, and Simpson. This suggested that no dramatic changes in the overall microbial community richness were induced by the tumor model construction or the anti-cancer treatment. Notably, both FOLFOX, and oxaliplatin could reverse the disorder of gut microbiota induced by tumor at phylum, class, order, family, or genus level (Figures 2C,D and Supplementary Figure S1). At phylum level (Figure 2C), the control group was dominated by Bacteroidetes (55.27%), and Firmicutes represented the dominant bacteria in the MO (Model for oxaliplatin, 58.36%) and MF (Model for FOLFOX, 49.09%) groups, while the abundance of Bacteroidetes was reversed in the Oxaliplatin (48.00%) and FOLFOX (51.44%) groups. At the genus level, 21 significantly changed bacterial genera were obtained from MO vs. Oxaliplatin group (Supplementary Figure S2), and 19 bacterial genera from MF vs. FOLFOX group (Supplementary Figure S3). Moreover, according to LEfSe analysis based on KEGG pathways, oxaliplatin could increase the carbohydrate and nucleotide metabolism of gut microbiota (Supplementary Figure S4A), while carbohydrate and lipid metabolism was elevated in the FOLFOX group (Supplementary Figure S4B) (LDA score > 2.0 with p < 0.05). Taken together, the 16S rDNA gene sequencing analysis suggested an altered gut microbiota composition and function after oxaliplatin and FOLFOX treatment, implicating potential roles of the gut bacterial community in chemotherapy outcome.

Here, changes in the distribution and function of gut microbiota in tumor-bearing mice were observed after oxaliplatin and FOLFOX treatment, but the actual role of gut bacteria in the improved efficacy of FOLFOX requires further exploration. For this purpose, we listed the significantly changed bacteria from the comparisons of MO vs. Oxaliplatin and MF vs. FOLFOX, respectively. As is shown in the Venn diagram (Figure 3A), in total there were 29 bacterial genera, among which 11 were shared by the both treatments, 10 and 8 were unique to oxaliplatin and FOLFOX treatment, respectively. Meanwhile, Spearman correlation analysis was performed to reveal the correlation between differential bacteria and chemotherapeutic efficacy (IR, RTV, TV, and Weight) after FOLFOX treatment (Figure 3B and Supplementary Figure S5). The abundance of RF32 and A. muciniphila was positively correlated with the IR of FOLFOX (p < 0.05). According to existing literature, A. muciniphila is known to associate with the improved prognosis of cancer patients and shows beneficial effects on metabolic disorders as well (Everard et al., 2013; Routy et al., 2018). Considering A. muciniphila had the highest fold change (33.55) between the MF and FOLFOX groups among all the bacteria (Figure 3C and Supplementary Figure S3), we conducted further experiments to validate its function.

To verify the influence of A. muciniphila on FOLFOX efficacy, A. muciniphila colonization combined with CT-26 colon cancer xenograft model was constructed (Supplementary Figure S6). The A. muciniphila transplantation was established referring to the published studies and our previous exploration (Gong et al., 2019). Three mice (labeled with Akk1, Akk2, Akk3) were randomly selected to evaluate the bacterial transplantation efficiency. As is shown in Figure 4A, the relative abundance of A. muciniphila remarkably increased after the colonization, indicating the success of model construction. First of all, FOLFOX treatment caused decreased body weight of tumor bearing mice, while A. muciniphila colonization did not influence it (Figure 4B). After pharmacodynamic evaluation, we found that the anti-cancer effect of FOLFOX was increased from 36 to 48% by ABX pretreatment, suggesting the involvement of gut microbiota in FOLFOX efficacy (Figure 4C). More importantly, the inhibition rate of FOLFOX was significantly enhanced (from 48 to 76%) with A. muciniphila colonization (p < 0.05) (Figure 4C). In addition, immunohistochemistry of Ki67 in tumor tissues further supported the above conclusion (Figure 4D). To summarize, our study confirmed that A. muciniphila transplantation could improve the efficacy of FOLFOX on colon cancer.

It was well known that gut microbiota derived metabolites are important functional readouts of the gut microbiome and play essential roles in the action of chemotherapeutic drugs (Zhang et al., 2017b; Jia et al., 2018). Therefore, fecal samples from MF and FOLFOX groups were further subjected for non-target metabolomics analysis referring to our previous studies (Zhang et al., 2017a; Gao et al., 2019). In metabolomics, LC-MS is the most commonly applied platform. However, GC-MS with chemical derivatization is advantageous in acquiring polar metabolites such as carbohydrates and organic acids that are usually not well retained on a reverse phase column in LC-MS (Beale et al., 2018). Therefore, both GC-MS and LC-MS were utilized in the current study to achieve a wide coverage of metabolites. A tight clustering of the QCs in the PCA score plots was observed (Figures 5A–C), indicating good reproducibility of the methods. OPLS-DA models based on GC-MS and LC-MS were established to identify differential features between the two groups (Figures 5D–F). Permutation tests with 500 iterations were performed to confirm the OPLS-DA models were not overfitting (Figures 5G–I). As a result, 45 significantly changed metabolites were annotated with VIP > 1 and p < 0.05 (Supplementary Table S2). Spearman correlation analysis was performed to correlate the abundance of differential bacteria and annotated metabolites. As is shown in Figure 6, the relative level of stearoylethanolamide, arachidonic acid, and docosahexaenoic acid was positively correlated with the abundance of A. muciniphila, whereas phenylalanyl-valine, leucyl-glutamate, isoleucyl-alanine, linoleic acid, octadecanedioic acid, and lysoPE (18:2) had negative correlations with A. muciniphila (p < 0.05). Interestingly, we observed three branched-chain amino acid (BCAA) containing dipeptides (i.e., phenylalanyl-valine, leucyl-glutamate, isoleucyl-alanine) which might be potentially important mediators involving in the superior efficacy of FOLFOX.