Chemotherapeutic resistant cholangiocarcinoma displayed distinct intratumoral microbial composition and metabolic profiles

Patient characteristics and tissue sample collection

Thirty-six freshly frozen tissues were obtained from CCA patients who had undergone surgery at Srinagarind Hospital, Khon Kaen University during January 2017 until May 2019 and patient data have been previously described (Suksawat et al., 2019). The protocol of the specimen collection and study were approved by the Ethic Committee for Human Research, Khon Kaen University (HE601149). In addition, written informed consent was obtained from each patient prior to surgery. Fresh tumor tissues were obtained from the resection of the primary tumor and stored in Hank’s balanced salt solution (HBSS) with antibiotic (Ciproflaxin, Cefazolin and Amphotericin B) at −80 °C. In the present study, we further explored the tumor tissues based on the HDRA result from the study of Suksawat et al. (2019) which divided patients into subgroups based on chemotherapeutic response patterns. The chemotherapeutic response characteristics of CCA patients whose the intratumoral microbiota profile were analyzed using 16S rRNA sequencing and whose metabolic signature were analyzed using NMR spectroscopy are shown in Table 1.

Histoculture drug response assay (HDRA)

Fresh tumor tissues were obtained from the resection of the primary tumor and storage in Hank’s Balanced Salt Solution (HBSS) at 4 °C. Then, the tumor tissues were minced into small pieces of approximately 9–12 mg and placed onto sponge in 24 well plates. Each well of the 24 well plates contained RPMI-1640 medium and a varying concentration of the gemcitabine and cisplatin drugs. The medium was supplemented with 20% fetal craft serum (FCS), 100 U/mL penicillin and 100 mg/mL streptomycin. After that, the tumor tissues were incubated at 37 C in 5% CO₂ for 4 days. Then, 100 µL of HBSS containing 0.1 mg/mL of collagenase type I and 100 µL of MTT solution were added into each well and further incubated for 4 h. The cell viability was then measured using an MTT assay. After that, the MTT formazan products are dissolved in DMSO and subjected to absorbance measurement at 540 nm (TECAN sunrise ELISA Reader, Triad Scientific, Manasquan, NJ, USA). Finally, the percent cell growth inhibition rate was calculated as previously described (Suksawat et al., 2019). The criteria for classification sample into sensitive and resistant were previously reported (Suksawat et al., 2019). A total of thirty-six CCA tumor tissues were treated with chemotherapy in five conditions , including low dose gemcitabine (LDGem) at 1,000 ug/mL, high dose gemcitabine (HDGem) at 1,500 ug/mL, low dose cisplatin (LDCis) at 20 ug/mL, high dose cisplatin (HDCis) at 25 ug/mL and combined treatment composed of 1000 ug/mL of gemcitabine and 20 ug/mL cisplatin, and evaluated using HDRA. Tissues were then sub-classified into sensitive (S) and resistant (R) groups to a particular chemotherapeutic condition.

DNA extraction and 16s rRNA sequencing

Total DNA was isolated from approximately 50 mg fresh frozen tumor tissues following the manufacture’s protocol (QIAGEN, Hilden, Germany). For quantification of the DNA extracted a spectrophotometer (Nanodrop) was used and with 1.5% agarose gel electrophoresis for visualization. Amplification and sequencing of the V1-V2 region were conducted. Briefly, 7.5 µL of genomic DNA from tissues were amplified using the 16 rRNA gene at the variable region V1-V2 incorporating Illumina adapters and a barcode sequence amplified (forward primer:5′-TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGGAGTTTGATCMTGGCTCAG-3′ and reverse primer:5′GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGGCTGCCTCCCGTAGGAGT-3′) using polymerase chain reaction (PCR) (T100TM Thermal Cycler, Bio-Rad, Hercules, CA, USA) with the specific primer using Hotstar Master Mix (QIAGEN, Hilden, Germany). The PCR cycling conditions used were: initial denaturation at 95 °C for 3 min; 25 cycles of denaturation at 95 °C for 30 s, annealing at 55 °C for 30 s, and extension at 72 °C for 30 s; and the final extension step at 72 °C for 5 min. The negative control (DNase free water) was applied in DNA extraction and 16S amplification steps. The absent band of the negative control was observed. Sequencing was performed on the Illumina MiSeq platform (Illumina^®, Macrogen, Korea), with read length of 301 base pair, paired-end.

16S rRNA data processing

Following standard quality control and demultiplexing, the reads were processed using the QIIME2 (version 2021.11) pipeline (Hall & Beiko, 2018). First, paired-end reads were joined and size selected to reduce non-specific amplification. These reads were then grouped into operational taxonomic units (OTUs) based on sequence similarity using the SILVA database (version 132) (Quast et al., 2013) and classified at ≥ 99% identity of reads. Data were rarefied to the minimum library size using total sum scaling (TSS). The alpha diversity and richness of CCA tissues between resistant and sensitive groups were calculated by using Chao1 and the Shannon and Simpson diversity indices. In addition, the edgeR algorithm was applied in order to compare and classify of differential abundance between resistant and sensitive groups to chemotherapeutic treatments. To evaluate the intratumoral microbial community between resistant and sensitive groups, we used the abundance data and calculated the differential microbial composition using Bray-Curtis dissimilarity and visualized by non-metric multidimensional scaling (NMDS) on projection in MicrobiomeAnalyst (Chong et al., 2020; Dhariwal et al., 2017).

Metabolite extraction and metabolomics analysis

Approximately 100 mg of each fresh frozen tumor tissue was used for metabolite extraction. The tumor tissues were then homogenized using a Dounce homogenizer and extracted by adding 400 µL of methanol and 85 µL of HPLC grade water, followed vortex mixing. Then, 200 µL of chloroform and 200 µL of HPLC grade water were added followed by vortex mixed. Next, the tissue extracted solutions are transferred into 15 mL tubes and sonicated 3 times using the following parameters: sonicate on 30 s and sonicate off 10 s at amplitude 40% and temperature of 4 °C. After that, the 15 mL tubes were subjected to centrifugation at 1,000 g at 4 °C for 15 min. The aqueous phase was subjected to nuclear magnetic resonance (NMR) spectroscopy or global profiling analysis. The NMR spectra data acquisition from NMR used peak alignment, normalization with probablistic quotient normalization and scaling using matrix laboratory software (MATLAB) (MathWorks Inc., US). The significant metabolites were identified using statistical total correlation spectroscopy (STOCSY), human metabolome database (HMDB) (Wishart et al., 2018; Wishart et al., 2013; Wishart et al., 2009; Wishart et al., 2007) and the Chenomx NMR suite (Chenomx Inc., Canada). The pairwise comparison of the log2 transformed data of metabolites between the resistant and sensitive groups was conducted with a paired non-parametric test (Mann–Whitney U test) and adjusted p value was calculated with a Benjamini–Hochberg procedure. The data was illustrated using Graph Pad prism 5 (GraphPad Software, Inc., CA, US). The network analysis was performed using Metscape (Gao et al., 2010) for visualizing metabolic pathways.

Correlation analysis

The correlation analysis was performed with Spearman’s correlation coefficient at the genus level and metabolites using the M²IA pipeline (Ni et al., 2020) for the integrated microbiome and metabolome dataset.

Difference of intratumoral microbiota composition between resistant and sensitive group of chemotherapeutic treatment in cholangiocarcinoma patients

Out of 36 tumor tissues, amplification for V1–V2 regions was successful for 18 samples. These samples were sequenced and a total read of 3,504,888 were acquired for microbial profiling. Following quality trimming and merging of overlapping paired-end reads, total read counts of 540,202 counts were retained from 18 samples, average counts per sample 30,011 counts. These reads could be assigned into a total of 890 bacterial OTUs. Overall, the intratumoral microbiome profile revealed a common pattern with the phyla Proteobacteria, Actinobacteria and Firmicutes dominating in both the resistant and sensitive groups in all conditions of chemotherapeutic treatment (Figs. 1A and 1D). The top three most abundant classes were Gammaproteobacteria, Actinobacteria and Alphaproteobacteria (Figs. 1B and 1E). The intratumoral microbiome profile in genera were shown in Figs. 1C and 1F. We then compared the alpha diversity between the resistant and sensitive groups. The Shannon and Simpson indexes revealed that tumor tissues treated with LDGem and HDGem had significant differences in microbial diversity between the resistant and sensitive groups. In contrast, Chao1 index demonstrated no difference in species richness between the resistant and sensitive groups (Fig. 2). A comparison of taxonomic profiles at the phylum level revealed that LDGem resistant group, HDGem resistant group, LDCis resistant group and HDCis resistant group showed higher abundance of Proteobacteria. A comparison of the taxonomic profiles at the class level demonstrated that tumor tissues which were resistant to LDGem, HDGem and LDCis exhibited higher abundances of Gammaproteobacteria, whereas the abundances of Actinobacteria was found to be lower in LDGem resistant group and HDGem resistant group (Fig. 3).

To explore whether the intratumoral microbial composition of CCA patients was different between the resistant and sensitive groups, non-metric multidimensional scaling (NMDS) was performed. NMDS is based on Euclidean distance and can reveal a shift of centroid (indicated by arcs) and variation in the microbiota community profiles of each chemotherapeutic drug treatment condition (circled area). The NMDS analysis at the class level demonstrated the overlap of the circle areas in each plot between the sensitive and resistant groups, showing some similar bacterial communities between the sensitive and resistant groups in all chemotherapeutic treatment conditions except, the resistant group of HDCis showed the smallest variance in the bacterial community (Fig. 4).

Metabolic alteration associated with chemotherapeutic responses

¹H NMR metabolic signatures from the CCA tissues are represented in Table 2. The metabolic differences between resistant and sensitive groups of CCA patients can be distinguished on univariate analysis (Mann–Whitney U test) using a log₂ transformation of maximum intensity. Significantly higher levels of acetylcholine, adenine, carnitine and inosine were observed in the LDGem resistant group. For the LDCis treatment, the levels of acetylcholine, alpha-D-glucose and carnitine were significantly increased in the resistant group compared to the sensitive group (Fig. 5). Towards the understanding of host-bacterial altered metabolic profiles, we performed metabolic pathway analysis executed on Metscape using KEGG (Kyoto Encyclopedia of Genes and Genomes) pathways, to investigate the most relevant pathways triggered by the chemotherapeutic response conditions. In addition to the upregulated acetylcholine metabolism and carnitine metabolism in both LDGem and LDCis groups, LDGem group exhibited the enhanced inosine and adenine metabolism and glucose metabolism (Fig. 6). Therefore, adenine and inosine involved in nucleotide metabolism also promote cancer cell proliferation (Newman & Maddocks, 2017). In addition, carnitine indicated cancer development and progression (Kawai et al., 2017). In term of glucose, glucose serve as inducer of progression of CCA (Saengboonmee et al., 2016). Furthermore, acetylcholine can promote cancer stem cell proliferation (Nguyen et al., 2018).

Correlation of metabolic profile and intratumoral microbiota composition

To examine the overall correlation between tissue microbial and metabolic profiles and to identify the accountable microbiota and metabolite(s), we performed a Spearman-rank correlation analysis between the genus-level relative abundances of tissues microbiota and the log₂ transformed relative concentrations of metabolites. In LDGem, Deinococcus was negatively correlated with homocarnosine and L-methionine, and Escherichia-Shigella was negatively correlated with homocarnosine (Fig. 7A). In HDGem, Deinococcus and Pseudomonas were negatively correlated with acetic acid and L-methionine; Atopostipes and Paracoccus were negatively correlated with acetic acid; and Streptococcus was negatively correlated with L-methionine (Fig. 7B). Finally, in HDCis, Cutibacterium was found to be positively correlated with L-leucine and L-isoleucine (Fig. 7C). There was no observable correlation between microbiome and metabolites in the LDCis and combined groups.