Patients sampling and follow-up
This was a retrospective study on OV-associated cholangiocarcinoma (CCA) patients who underwent surgery at Srinagarind Hospital, Khon Kaen University, Khon Kaen, Thailand between 2007 and 2016. Pre-operative blood samples were collected from CCA patients and allowed to clot at room temperature before being centrifuged at 1,000 g at 4 °C for 10 mins. Then, the serum was carefully collected into 1.5 mL tubes and stored at -80 °C until analysis. CCA tissues were obtained from patients after surgery and kept in the BioBank of the Cholangiocarcinoma Research Institute. The patients were excluded if they received either radiotherapy or chemotherapy before surgery.
Patients were followed-up every 3 months in first year after surgery, then every 6 months thereafter. Computed tomography (CT)/magnetic resonance imaging (MRI) were performed to confirm post-operative recurrence in patients who have symptoms or signs of cancer recurrence. Recurrence-free survival (RFS) was measured from the date of surgery to recurrence or until the last of follow-up in patients without recurrence. The study was approved by the Human Research Ethics Committee, Khon Kaen University, Thailand (HE611412).
Sample preparation and acquisition for 1H-NMR spectroscopy
Prior to the metabolomics analysis, the frozen serum samples were defrosted at 4 °C and mixed. Then, the samples were centrifuged and 300 μl of supernatant was gently mixed with 300 μl of serum buffer (0.075 M Na2HPO4 pH 7.4 in D2O, 4.6 mM TSP, 0.004% NaN3). This was followed by centrifugation at 10,000 g, 4 °C for 10 mins. The mixed samples, 550 μl, were transferred into 5 mm NMR tubes (DWK Life Sciences, Germany). These were kept at 4 °C until analysis.
1H-NMR spectra were acquired at 298 K using an NMR spectrometer at 400 MHz (Bruker, USA). The Carr-Purcell-Meiboom-Gill (CPMG) pulse sequence was employed to obtain spectra (recycle delay-90°-t1-90°-tm-90°-acquisition) in 64 scans.
NMR spectral processing and statistical analyses
Data processing was performed using an in-house MATLAB script. Phase and baseline correction were performed in all NMR spectra and the TSP peak was set as 0 ppm. After peak alignment, the water region was excluded (d 4.2-d 5.2). Pseudo two-dimensional spectra were drawn in order to identify all metabolites by statistical total correlation spectroscopy (STOCSY), which confirmed the correlation of each resonance. Additionally, the resonances were searched against the Human Metabolome Database (HMDB), the ChenomxNMR Suite and available literature. The integral area under the peak was obtained using the in-house MATLAB script. The concentration of peak of interest was presented by comparison with TSP that added was as an internal standard. The concentrations of interested peaks were represented as median with interquartile range (IQR).
Sample preparation and UPLC-MS analysis
The sample preparation and UPLC-MS analysis were performed as previously published [27, 28]. In brief, the frozen serum samples were sorted to the set of 80 in a rack. Then, the samples were defrosted at 4°C overnight and transferred into 96-deep-well polypropylene plates (2 mL, Eppendorf). The plates were sealed and centrifuged at 3,486 g at 4°C for 10 mins. After centrifugation, any solid debris were removed using a clean pipette tip. The supernatant (50 μl) was aliquoted into individual 96-well polypropylene plates and four parts of isopropanol were added for protein precipitation. The 96-well polypropylene plates were sealed and then mixed at 1,400 rpm at 4°C for 2 h. After mixing, the plates were centrifuged at 3,486 g at 4°C for 10 mins. 125 µL of supernatant was aspirated into a new 96-well polypropylene plate. The supernatant of each sample was also pooled to create the study reference (SR) sample in order to perform quality control (QC), which was performed throughout the analysis in every 10 study samples. In addition, SR samples were diluted through seven dilution series and acquired at the beginning and end of the run.
The prepared samples were examined using reversed-phase ultra-performance liquid chromatography (RP-UPLC). A 2.1×100 mm BEH C8 column (Waters Corp., UK) was used for analysis and the column temperature set at 55 °C. Mobile phase A was a mixture of water, acetonitrile (ACN), isopropanol (IPA) in the proportion of 50:25:25 with 5 mm ammonium acetate, 0.05% acetic acid and 20µM phosphoric acid. Solvent B was a mixture of CAN, IPA in the proportion of 50:50 with 5 mm ammonium acetate and 0.05% acetic acid. The RP-UPLC was coupled with Xevo G2-S QTOF MS (Waters Corp., UK) via a Z-spray electrospray ionization (ESI) source for lipidomics analysis. The samples were acquired in both positive and negative ion modes in order to create the result in both positive and negative datasets, respectively.
LC/MS data processing and statistical analyses
After data acquisition, XCMS was used for feature extraction . In addition, the potential run-order effect elimination and feature filtering were performed using in-house and open-source scripts . In order to gain only the features with high accuracy and high precision, features with the coefficient of variance (CV) in SR samples less than 20% and features correlated to SR dilution which showed a Pearson correlation coefficient greater than 0.8 were retained. After that, the data matrix was normalized using median fold change normalization. The data file was subjected for multivariate analysis using SIMCA 14 software (Umetricas, Sweden). After orthogonal partial least squares discriminate analysis (OPLS-DA) was applied, the variables with relevance to the discrimination between recurrence (R) and non-recurrence (NR) based on a p(corr) cut-off of ê0.5êtogether with variables important in the projection (VIP) score above 1.0 were selected. The data was analyzed using the Mann-Whitney U-test in MetaboAnalyst 4.0 software; variables with a false discovery rate (FDR) adjusted P-value less than 0.05 were selected for further analysis
Subsequently, the significant features were identified using m/z by matching with online databases (Metline and HMDB). Then, the structure of the lipids was investigated using MS/MS fragmentation patterns. The level of assignment was grouped based on the previously published criteria : (1) m/z matched to database, (2) m/z matched to database and MS/MS fragment matched to in silico fragmentation pattern, (3) MS/MS fragment matched to database or literature review, (4) retention time matched to standard compound, (5) MS/MS fragment matched to standard compound.
The antibodies used in this study were: mouse monoclonal anti-CD44 (1:100; #ab516728); mouse monoclonal anti-CD44v6 (1:50; #ab78960); rabbit polyclonal anti-EpCAM (1:100; #ab71916); rabbit monoclonal anti-CD36 (1:25; #ab133625); rabbit monoclonal anti-ATP citrate lyase (1:200; #ab40793); rabbit monoclonal anti-SCD1 (1:100; #ab236868) and HRP conjugated rabbit anti-rat (1:50; #ab6734) antibodies (Abcam, CA), rat monoclonal anti-CD44v8-10 antibody (1:50; #LKG-M001) (Cosmo Bio, JP).
Immunohistochemistry (IHC) and scoring
Two independent punctures from paraffin-embedded tissues of each patient were used to produce tissue microarrays (TMA). Tissue sections were de-paraffinized and rehydrated stepwise of xylene, 100%, 90%, 80% and 70% ethanol, respectively. Microwave cooking was used for antigen retrieval with 10 mM sodium citrate, pH 6, 0.05% Tween20 for CD36, CD44 and CD44v6, whereas the Tris-EDTA, pH 9 was used for ATP-citrate lyase, SCD1, EpCAM and CD44v8-10. Endogenous hydrogen peroxide activity and nonspecific binding were blocked with 0.3% hydrogen peroxide and 10% skim milk for 30 mins. Primary antibody was added and incubated at room temperature for 1 h, then at 4 °C overnight. After washing, secondary antibody (Dako EnVision) was added for 1 h, except for CD44v8-10. HRP conjugated anti-rat was added and left for 3 h. 3, 3’diaminobenzidine tetrahydrochloride (DAB) substrate kit (Vector Laboratories, Inc., Burlingame, CA) was used for signal development. Sections were then counterstained with Mayer’s haematoxylin. Dehydration was performed stepwise of 70%, 80%, 90%, 100% ethanol and xylene, respectively and mounted with permount. Stained sections were viewed under a light microscope.
The IHC score of each patient was calculated as the average score from two independent punctures. Staining frequency and intensity were used for scoring. The percentage of positive cancer cells was defined as frequency with 0%=negative, 1–25%=+1, 26–50%=+2, and>50%=+3. Intensity was scored as three levels, weak=1, moderate=2, and strong=3. The range of final scores was 0-9, determined by multiplying the intensity with the frequency. IHC score was calculated as a median value and used as a cut-off point. Patients were classified as low or high expression groups if the grading score was lower or equal to or higher than the median, respectively. For protein having a median value equal to zero, patients were classified into negative or positive expression groups if the grading score was equal to or higher than zero, respectively.
The results from global metabolomics and lipidomics were analyzed using SPSS statistical package version 25 and SIMCA software 14 together with MetaboAnalyst 4.0 software. The differential metabolites were further analyzed using hierarchical clustering and correlation heatmap analysis, metabolic pathway analysis and also the receiver operator characteristic (ROC) curve using MetaboAnalyst. For IHC results, the correlation between proteins was analyzed using correlation heatmap analysis, MetaboAnalyst. The association between metabolic levels, protein levels and RFS was analyzed by Kaplan-Meier survival analysis and the log-rank test using SPSS. A P-value less than 0.05 was considered as statistical significantly.