Identification of novel biomarkers of hepatocellular carcinoma by high‐definition mass spectrometry: Ultrahigh‐performance liquid chromatography quadrupole time‐of‐flight mass spectrometry and desorption electrospray ionization mass spectrometry imaging

Rationale Hepatocellular carcinoma (HCC) is a highly malignant disease for which the development of prospective or prognostic biomarkers is urgently required. Although metabolomics is widely used for biomarker discovery, there are some bottlenecks regarding the comprehensiveness of detected features, reproducibility of methods, and identification of metabolites. In addition, information on localization of metabolites in tumor tissue is needed for functional analysis. Here, we developed a wide‐polarity global metabolomics (G‐Met) method, identified HCC biomarkers in human liver samples by high‐definition mass spectrometry (HDMS), and demonstrated localization in cryosections using desorption electrospray ionization MS imaging (DESI‐MSI) analysis. Methods Metabolic profiling of tumor (n = 38) and nontumor (n = 72) regions in human livers of HCC was performed by an ultrahigh‐performance liquid chromatography quadrupole time‐of‐flight MS (UHPLC/QTOFMS) instrument equipped with a mixed‐mode column. The HCC biomarker candidates were extracted by multivariate analyses and identified by matching values of the collision cross section and their fragment ions on the mass spectra obtained by HDMS. Cryosections of HCC livers, which included both tumor and nontumor regions, were analyzed by DESI‐MSI. Results From the multivariate analysis, m/z 904.83 and m/z 874.79 were significantly high and low, respectively, in tumor samples and were identified as triglyceride (TG) 16:0/18:1(9Z)/20:1(11Z) and TG 16:0/18:1(9Z)/18:2(9Z,12Z) using the synthetic compounds. The TGs were clearly localized in the tumor or nontumor areas of the cryosection. Conclusions Novel biomarkers for HCC were identified by a comprehensive and reproducible G‐Met method with HDMS using a mixed‐mode column. The combination analysis of UHPLC/QTOFMS and DESI‐MSI revealed that the different molecular species of TGs were associated with tumor distribution and were useful for characterizing the progression of tumor cells and discovering prospective biomarkers.


| INTRODUCTION
Hepatocellular carcinoma (HCC), the primary type of liver cancer, is highly malignant and is continuing to increase worldwide. The fiveyear survival rate is less than 15%, and HCC is the second leading cause of death in East Asia. 1 Early diagnosis of HCC is necessary to reduce the number of deaths and maintain quality of life with cancer. α-Fetoprotein (AFP) and a protein induced by vitamin K absence or antagonist-2 (PIVKA-II) are conventionally used as diagnostic markers for HCC. However, the number of new cases of HCC is currently equal to the number of deaths every year, and these diagnostic markers might not be reflected in the early stage of HCC. 2,3 Therefore, it is necessary to develop prospective or prognostic biomarkers to characterize the progression of HCC in relation to the phenotype of tumor cells.
The phenotypic changes in HCC are regulated by the genome, epigenome, transcriptome, proteome, environmental factors, and microenvironment, such as growth factors and cytokines, [4][5][6] and are extraordinarily heterogeneous among individuals. 7 Metabolomics is the field of omics studies that examines a whole set of small molecules in biological samples and is able to detect subtle changes in metabolic pathways and deviations from homeostasis before the manifestation of phenotype. Metabolomics is therefore widely used for biomarker discovery and is a promising approach for the identification of disease-related small molecules. 8,9 Therefore, metabolomics, also called metabolic phenotyping, is an approach with the potential to discover biomarker candidates for HCC by analyzing tumor expression.

Metabolomics is classified into targeted metabolomics (T-Met)
and untargeted metabolomics. 10 Although the quantitative values of known features of a metabolic pathway can be acquired and analyzed with other omics layers by T-Met, 11,12 it is possible to miss the metabolite most associated with a disease. On the other hand, untargeted metabolomics can profile the phenotype based on a number of features and unbiased analysis to identify the diseaserelated metabolites in the process of biomarker discovery 13 and can be useful in searching for disease markers and elucidating biological function. 14,15 Although the value of untargeted metabolomics has been recognized in the clinical field, 16 some bottlenecks remain to be considered.
Multiple analytical platforms were conventionally prepared for untargeted metabolomics to acquire a wide polarity range of metabolites as for global metabolomics/metabolic profiling (G-Met) which were time-consuming for acquisition and data analysis. 17 Although G-Met has already been applied for large-scale analyses such as cohort studies, 18,19 the median intensities of detected metabolites decreased during the assays along with the injections required for normalization. 17 Therefore, a comprehensive and reproducible method is needed for G-Met in one assay. Recently, a mixed-mode column including both an ion-exchange phase and a reverse phase was developed, allowing a wide polarity range of metabolites to be separated in a short time in one assay. 20,21 The metabolites were generally identified by the matching of mass spectra and fragment ions with high accuracy to theoretical values in a database and matching of the retention times of the chromatogram to chemical standards, 22,23 which defined the quality of identification. 24,25 However, there were still isomeric and isobaric species, such as lipids. High-definition mass spectrometry (HDMS) is a technology that relies on the mobility of ions, including small molecules, according to the charge, shape, and size in the gas phase inside the MS instrument. 26,27 The collision cross section (CCS) of each metabolite can be calculated with the drift time acquired in the mobility cell to provide specific information on each structure. In addition, the background noise levels were reduced using HDMS, and clear mass spectra could be obtained for the identification of metabolites. [28][29][30] In the study reported here, we developed a wide-polarity G-Met method using a mixed-mode column and identified biomarker candidates for HCC in human liver samples using HDMS. Then, we

| Samples
Institute of Cancer Research mouse liver tissue was purchased from KAC Co. (Kyoto, Japan). Human liver was obtained from HCC patients at the Hepatobiliary Pancreatic Surgery Division, Department of Surgery, at the University of Tokyo Hospital between January 2013 and October 2014. Patient information was described in a previous report. 4 The research ethics committee of the Faculty of Medicine, University of Tokyo approved the present study, which was conducted in accordance with the ethical guidelines of the 1975 Declaration of Helsinki. All patients provided written informed consent for the use of clinical samples.

| UHPLC/QTOFMS analysis
, and methanol (C). Features were separated by gradient conditions; the initial condition was 0% B with 0.4 mL/min, followed by a linear gradient to 100% B from 1 to 4 min; 100% B was maintained for 2.5 min; and the mobile phase was returned to the initial conditions and maintained for 2.5 min until the end of the run. The total run time was 10 min, and the column oven temperature was 45°C. The flow line of the mobile phase (C) was connected to the line from the column to the ESI source as the postcolumn infusion.

| Data processing
The data procession of G-Met was followed as in a previous report. 17 All data obtained using UHPLC/QTOFMS were imported to Progenesis QI 3.0.3.0 (Nonlinear Dynamic, Newcastle, UK) for peak picking, alignment, and normalization to produce peak intensities for retention time (t R ) and m/z data pairs. The relative intensities of features were analyzed by principal component analysis (PCA) using SIMCA13.0.0 (Umetrics, Umeå, Sweden), and normalized with Quantbolome software.

| Human liver analysis by UHPLC/QTOFMS
Tumor (n = 39) and nontumor (n = 79) human liver samples were prepared as described in section 2.3. The supernatant of the centrifuged sample was transferred to a 96-well sample collection plate and diluted (×2) on the plate with water containing 0.1% formic acid. The preparation of SQC and dQC and the run order of samples for G-Met analysis were performed as described in a previous report. 17 Then, 4-μL samples were injected into the UHPLC/QTOFMS system. The data processing was performed as described above. Then, the normalized intensities of features were used for further multivariate analysis, such as PCA and orthogonal partial least squares discriminant analysis (OPLS-DA).

| Identification of metabolites by HDMS
The features extracted by OPLS-DA were first identified with the Chemspider database, human metabolome database, and Lipidmaps database. Then, the features were identified with the Synapt G2-Si (traveling-wave ion mobility mass spectrometer) system in HDMS mode. The source parameters for ESI were the same as described in  FIGURE 1 PCA score plot based on the intensity of features detected in mouse liver A, after and B, before normalization. Four-, six-, and eight-week-old male mice are represented by blue circles, squares, and triangles, and female mice are represented by red circles, squares, and triangles, respectively. The time-dependent drifts of the sample analyses are shown by black arrows Definition Imaging (HDI) software (Waters Corporation) to process the mass spectral data and to construct two-dimensional ion images.

| UHPLC/QTOFMS analysis
We first evaluated the utility of G-Met by UHPLC/QTOFMS with a mixed-mode column for the analysis of multiple biological liver samples. The six groups of mice were clearly separated on the PCA score plot based on the detected chemical features from mouse livers after normalization by software ( Figure 1A). The median intensities of ions decreased with the analysis of multiple injections, 32 as we could see the time-dependent drift line on the score plot of PCA before normalization ( Figure 1B, black arrows).
The results indicated that the present method can be utilized for the G-Met analysis of liver samples with the normalization procedure.

| Human liver analysis by UHPLC/QTOFMS
We then applied the G-Met method to human tumor and nontumor liver samples. A total of 2447 or 1601 features were originally detected in positive or negative ion mode, and 1394 or 728 metabolites were identified with the databases within 5 ppm mass accuracy and classified into 13 groups, which included candidates based on the biological function of metabolites or lipid species, as shown in Figure 2. Although G-Met has been performed with a hydrophilic interaction liquid chromatography column and/or a reversed-phase liquid chromatography C18 column, each method has a limited capacity to retain hydrophobic and/or hydrophilic metabolites, respectively, and decreased reproducibility because of ion suppression for those metabolites. 33 Therefore, combination analyses by both columns have been conventionally used for G-Met.
The mixed-mode column combines ionic phases with the C18 phase and improves the retention of both hydrophilic and hydrophobic metabolites. 20 The present G-Met method that we developed can expand the polarity range of detected metabolites in one assay and has enough coverage of biological molecules for biomarker discovery.
We next compared the intensity of 2447 features originally detected in positive ion mode from the tumor and nontumor regions by using multivariate analyses. The tumor regions were slightly shifted from the nontumor regions and overlapped on the PCA score plot ( Figure 3A). We then extracted the features that contributed to the separation on the S-plot of OPLS-DA, as shown in Figure 3B.   Recently, the identification of tumor and nontumor regions was achieved using direct MS technology in the tissue of cancer patients during surgical operation. 51 The profiling of FA species in TG has the potential to distinguish tumor, nontumor, and boundary regions by means of this technology. However, the quality of biomarker candidates generally must be evaluated by plasma analysis to be used in clinical diagnosis. AFP is conventionally used for the diagnosis of HCC and shows 60%-80% sensitivity as a biomarker.

| Identification of metabolites by HDMS
However, the sensitivity decreases to approximately 40% for the detection of small tumors. 52 Therefore, the combination of several molecules is necessary to improve the sensitivity and specificity for the early diagnosis of HCC. 38 In this study, we demonstrated the difference in TG species in HCC tissue. However, it is still necessary to evaluate the correlation with the plasma concentration of TGs and reproducibility as a candidate biomarker with increased sample numbers.

| CONCLUSIONS
The present G-Met method with a mixed-mode column was comprehensive and reproducible and could detect a wide polarity range of metabolites in biological samples. Novel biomarkers for HCC were identified with the G-Met method, and the difference in