Plasma proteome atlas for differentiating tumor stage and post-surgical prognosis of hepatocellular carcinoma and cholangiocarcinoma

Although mass spectrometry-based plasma proteomics enables sensitive and large-scale discovery and validation of biomarkers for various diseases, its integrative application to hepatocellular carcinoma (HCC) and cholangiocarcinoma (CCA) is not well investigated. Therefore, we analyzed albumin- and immunoglobulin G-depleted plasma samples from 148 and 60 patients with HCC and CCA, respectively, using liquid chromatography-tandem mass spectrometry. The algorithm used to measure the content of each protein was the percentage of exponentially modified protein abundance index. From 5320 proteins assayed in plasma, 53 and 25 biomarker candidates were identified for HCC and CCA, respectively. The abundance of six and two HCC markers particularly protruded in stage II and III, respectively, whereas plasma serine protease inhibitor was the sole marker the level of which steadily decreased with CCA progression. From a prognostic facet, we showed candidate markers and their cutoff levels for evaluating probability of tumor recurrence and patient survival period. Combination Kaplan-Meier models showed that HCC stage III or IV and both the content of alpha-2-HS-glycoprotein and apolipoprotein CIII <0.2% exhibited the poorest post-surgical recurrence-free and overall survivals. Furthermore, the content of afamin ≥0.2% played a significant role on the poor prognosis in patients with CCA. Our findings, taken together, characterized novel plasma biomarker signatures in dissecting tumor stages and post-surgical outcomes of HCC and CCA.


Introduction
Hepatocellular carcinoma (HCC) and cholangiocarcinoma (CCA; also known as bile duct cancer) are the two most common types of hepatobiliary malignancies, arising from neoplasms of hepatocytes and cholangiocytes, respectively. Hepatobiliary cancer is the fifth most common type of cancer and the third most common cause of cancer-related deaths worldwide. Approximately 700,000 new cases of HCC are reported annually on a global scale, of which hepatitis B or C virus-infected cases account for more than 75% [1]. CCA incidence has risen over the last two decades, particularly in Southeast Asia, with an estimated 130,000 new cases per year. Depending on the tumor stage, hepatic functional reserve, and performance status of patients, the treatment options for HCC include radiofrequency ablation, transarterial chemoembolization, radioembolization, multikinase inhibitors, hepatic resection, and liver transplantation [2]. The 5-year survival rates of HCC at early and late diagnoses are approximately 30% and lower than 15%, respectively. Unlike HCC, therapeutic choices for CCA are limited because it is strongly resistant to chemotherapy. The curative options are surgery and liver transplantation in its early stage. However, CCA is difficult to diagnose, aggressive, and heterogenous; therefore, less than one-third of such cancers are unresectable or metastasized at diagnosis. With this, the remnant treatments available are systemic or palliative therapy, leading to a 5-year mortality rate of higher than 90% [3,4]. Mass spectrometry-based proteomics in liquid biopsies is one of the most powerful platforms to noninvasively determine protein biomarkers of various diseases because it is easy to access, has high sensitivity to identify targets at a very low abundance within complex mixtures, and can detect negligible differences in expression levels. The hepatobiliary system is the major contributor to the plasma protein pool; hence, plasma proteomics should be leveraged in the diagnosis, elucidation of the oncological processes, and prognosis of HCC and CCA. Early diagnosis of tumor greatly improves the curative frequency and medical outcomes of patients, and plasma proteome analysis for HCC and CCA will not only solve technical problems on tumor staging, such as insufficient resolution in cancer screening and sampling bias in liver biopsy, but also facilitate longitudinal tracking of the disease status [5]. For this purpose, we performed a label-free, quantitative proteomics study to search plasma protein markers specific to HCC and CCA, addressed their expression patterns in different tumor stages, and evaluated their applicability to tumor recurrence and patients' survival.

Patients and study design
The Institutional Review Board of National Cheng Kung University Hospital (NCKUH) approved our study (approval numbers: B-ER-103-133 and B-ER-105-098), which was conducted following the guidelines of the Declaration of Helsinki. We collected the plasma samples, clinical data, and laboratory data of patients with HCC (n = 148) and CCA (n = 60) who underwent surgical resection from the Tissue Bank, Research Center of Clinical Medicine, NCKUH. Medical records of the patients were accessed from December 2002 to March 2014, and all data were fully anonymized before obtaining access. All the patients could not be identified. We also enrolled 95 control participants who were negative for hepatobiliary diseases from the Health Examination Center of NCKUH and obtained written informed consent from them. All plasma samples were stored at -80˚C until use. We followed the TNM classification guidelines given in the eighth edition of the American Joint Committee on Cancer Staging Manual for staging HCC or CCA.
using 10% sodium dodecyl sulfate plus 10 mM dithiothreitol at 95˚C for 10 minutes, and alkylated with 10 mM iodoacetamide at 37˚C in dark for 1 hour. The proteins were then cleaned and concentrated using Amicon Ultra-0.5 mL centrifugal filters (molecular weight cutoff: 3,000 Da) device (Merck Millipore, Darmstadt, Germany) and digested at 37˚C overnight using sequencing grade trypsin (Promega, Fitchburg, WI, USA) in 10 mM ammonium bicarbonate in an enzyme-to-substrate ratio of 1:50. We followed the procedure of liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis outlined in our previous study [6]. A rapid separation liquid chromatography system (Ultimate 3000; Dionex, Santa Clara, CA, USA) equipped with a C18 column (Acclaim PepMap RSLC, 75μm × 150 mm, 2 μm, 100 Å) was coupled to a Q Exactive Orbitrap mass spectrometer (Thermo Fisher Scientific). Mobile phase A and B consisted of 0.1% fluoroacetic acid and 0.1% fluoroacetic acid in 95% acetonitrile, respectively. The gradient comprised a linear increase from 1% to 25% B over 45 minutes, a linear increase from 25% to 60% B over 10 minutes, and finally, isocratic elution at 80% B for 10 minutes at 250 nL/minute for separation. A full MS spectrum (m/z 300-2000) with a mass resolution of 140,000 was acquired, followed by 10 sequential collision-induced dissociation-MS 2 scans using the mass spectrometer in data-dependent mode.

Protein identification and quantification
Raw data of LC-MS/MS were processed into peak lists by Proteome Discoverer 1.4 for Mascot database (version 2.4.1, Matrix Science Ltd., London, UK) searched against the Swiss-Prot_2015_07 database. Parameters were set as follows: enzyme, trypsin; missed cleavages, 1; peptide mass tolerance, 10 ppm; fragment mass tolerance, 0.05 Da; fixed modification, carbamidomethyl (C). The exponentially modified protein abundance index (emPAI) was used to calculate the number of sequenced peptides per protein [7]. The percentage of each emPAI from the summation of all the emPAI values for the identified proteins was leveraged as the algorithm to quantify the content of each protein [8]. All identified targets were included in the calculation of emPAI% of each protein to avoid creating a bias when measuring the content of each protein. However, biomarker candidates shown a protein score <30 were removed from the list of marker candidates to reduce a false identification. All analysts were blinded to any information about the subjects.

Statistical analysis
Statistical analyses were performed using PASW Statistics for Windows (version 18.0; SPSS Inc., Chicago, IL, USA). Differential proteins in HCC or CCA were identified using receiver operator characteristic curves (the area under the ROC curve > 0.7 and P < 0.00001). Continuous and nominal variables in different tumor stages were compared using Kruskal-Wallis tests and Pearson Chi-square tests, respectively.
Venn diagrams were obtained using InteractiVenn (http://bioinfogp.cnb.csic.es/tools/ venny/) [9]. The emPAI% values of protein markers in different tumor stages were compared using one-way analysis of variance with Scheffé posterior comparison. Protein levels between advanced-stage and non-advanced-stage HCC and CCA as well as between metastatic and non-metastatic HCC and CCA were compared using Mann-Whitney U tests. The significance of protein biomarkers on recurrence-free and overall survivals of the patients was assessed using Kaplan-Meier analyses in combination with log-rank tests. Proteins associated with post-surgical tumor recurrence and mortality were identified using stepwise Cox regression analyses. Both models with and without a post-univariate Bonferroni correction were shown. Significance was set as two-tailed P < 0.05.

Characteristics of patients at different HCC and CCA stages
We used the American Joint Committee on Cancer staging system to classify the cancer stage of the patients. Of the 148 patients with HCC, 58, 55, 20, 5, 6, and 4 were in stage I, II, IIIA, IIIB, IIIC, and IV, respectively. Although they were at different stages of HCC, the patients demonstrated similar demographic, biochemical, and hematological data ( Table 1). The median alanine aminotransferase and aspartate aminotransferase levels in the different tumor stages were below three and two times the upper limit of normal, respectively. Hepatitis B virus infection and liver cirrhosis were prevalent in more than 50% and more than 60% of patients with stage I-III HCC ( Table 1). The proportion of cirrhotic patients increased from 45% in stage IIIA HCC to 100% in stage IIIB HCC (S1 Table). Furthermore, post-surgical HCC recurrence was observed in over 70% of these patients in five years. The 5-year survival rates declined in stages III and IV HCC as expected, with a 100% mortality rate in stages IIIC and IV.
Among patients with CCA, 11, 22, 6, and 21 were in stages I, II, III, and IV, respectively (Table 1). No gender and age differences among patients in different tumor stages were found. Biochemical testing results and hepatitis B or C virus infection rate also did not alter the CCA progression. A slight increase in white blood cell count was seen in stage III CCA. A post-surgical recurrence rate of more than 30% was observed in each tumor stage of CCA. At the 5-year post-surgical follow-up, 100% and 95.2% mortality rates of patients in stages III and IV CCA, respectively, were seen.

Plasma proteome profiles and protein markers for HCC and CCA
A flowchart of this study was shown in Fig 1. A total of 5320 proteins were identified in the pre-operative, albumin-and IgG-depleted plasma samples of the subjects. The minimal protein score of identified protein was 13. Of 30 randomly selected plasma samples, the mean false discovery rate of protein identification was 1.35%. Our results demonstrated that 1319 proteins (24.8%) were common to all three groups, whereas 1211 (22.8%), 672 (12.6%), and 959 (18.0%) were unique to HCC, CCA, and controls, respectively (Fig 2A). The biomarker candidates for HCC (Table 2) and CCA (Table 3) were 53 (34 upregulated and 19 downregulated) and 25 (2 upregulated and 23 downregulated), respectively. Of these, 12 for HCC and 6 for CCA possessed both specificity and sensitivity of more than 70%. Comparisons of our HCC plasma biomarker candidates with 2 review articles regarding circulating or secretory HCC protein biomarkers were shown in S1 Fig. Few common HCC biomarkers were identified among each other. Our data showed that expressions of afamin, alpha-2-HS-glycoprotein, apolipoprotein B-100, clusterin, hepatocyte growth factor-like protein, and kininogen-1 were stimulated in HCC but repressed in CCA, while the expression of Ig lambda chain V region 4A was stimulated in CCA but repressed in HCC (Fig 2B). Levels of five biomarker candidates for HCC were higher in stage III and IV than in stage I and II and levels of fibrinogen gamma chain and selenoprotein P changed in metastatic HCC (S2 Fig). Moreover, plasma serine protease inhibitor was found to be downregulated in advanced and metastatic CCA.

Tumor markers with HCC and CCA progression
Regarding the relevance of these biomarker candidates in different tumor stages, we observed that the protein contents of 13 HCC markers, of which 11 were upregulated and two Ig kappa chain-associated proteins were downregulated, varied in different HCC stages (Table 4). We also observed different expression levels of six and two proteins between stages I and II and stages I and III, respectively.
Fold changes of the mean emPAI% values of these 13 HCC markers in comparison with the controls were shown in Fig 2C and S2 Table. Most markers peaked in stage II and dropped in stage III with or without a mild rebound in the end stage. Stage II HCC showed strikingly high expressions of 2-hydroxyacylsphingosine 1-beta-galactosyltransferase (5.61×) and Ig lambda chain VIV region Hil (19.83×). No marker was perfectly correlated with HCC progression.  White blood cell (10 3 /μL) 6.9 (4.2-9.1) 5.9 (3.6-12.  Plasma serine protease inhibitor was the sole protein for CCA staging (Table 4), showing reductions of 36%, 61%, 81%, and 85% in stages I, II, III, and the end stage, respectively ( Fig  2D and S2 Table).

Markers for HCC and CCA prognosis
Kaplan-Meier analyses revealed six and seven significant markers to evaluate post-surgical tumor recurrence and survival for HCC, respectively, and six and two markers were shown to assess post-surgical tumor recurrence and survival for CCA, respectively. The parameters predicting better recurrence-free and overall survivals in patients with HCC than their counterparts were the values of emPAI% of alpha-2-HS-glycoprotein �0.2%, apolipoprotein CIII �0.2%, Ig lambda chain VI region NEWM �1.0%, and serum amyloid P component �0.3%  regression analyses showed significant factors that were associated with high recurrence and mortality rates in five years after surgery in patients with HCC (Table 5) or CCA (Table 6). Combination models showed that tumor stage and two prognostic markers, alpha-2-HS-glycoprotein and apolipoprotein CIII, had a similar effect on the prognosis of HCC (Fig 3A). More specifically, HCC stage III or IV and both prognostic markers <0.2% exhibited the poorest post-surgical recurrence-free and overall survivals. By contrast, early-stage HCC and both prognostic markers �0.2% demonstrated the most favorable post-surgical outcomes. Furthermore, the level of afamin played a similar role, even more significant than the tumor stage, on the tumor recurrence and overall survival in CCA (Fig 3B).

Discussion
Current, HCC and CCA are screened and staged mainly using imaging (e.g. ultrasound, Xrays, computed tomography scan, magnetic resonance imaging), pathological tests, and laboratory tests. However, a portion of patients with HCC or CCA are misdiagnosed or receive a delayed diagnosis and are typically amenable to undergo surgical resection or liver transplantation. Moreover, annually, nearly 1 million liver cancer-related deaths continue to be reported worldwide. This information indicated the urgency of developing novel highly sensitive and specific screening platforms, by overcoming bottlenecks of traditional systems, for the determination of early onset of hepatobiliary cancers and manifestations of their deterioration. Label-free quantitative mass spectrometry is applicable to a wide variety of research fields, for which the following three approaches are the most commonly used: spectral counting, peptide chromatographic peak area, and emPAI. Spectral counting method measures the number of MS/MS spectra of a given protein. The peak area method involves calculating and comparing the mean intensity of peak areas for all peptides from each protein [10,11]. The emPAI quantification method is a modified form of spectral counting [7]. Dowle et al. compared the diagnostic accuracy of the three label-free methods by analyzing a solution of 18 exogenous proteins added to E. coli lysate and found that spectral counting and emPAI displayed satisfactory positive predictive values for Gross differences (1.5-to 1-fold difference), whereas peak area performed best for smaller-fold differences (1.1-to 1.75-fold) [12]. A study using complex mixtures of mouse neuro2A cells demonstrated that emPAI was strongly correlated with the actual protein amount in a wide dynamic range from 30 fmol/μL to 1.8 pmol/μL in the sample solution [7]. Nevertheless, this method is more suitable for interprotein and Table 3. Differential plasma proteins for cholangiocarcinoma (CCA).  [8]. We adopted emPAI% using a fixed amount of protein for each sample in spite of the fact that absolute protein levels have yet to be known to perform an LC-MS/MS-based intersample, comparative proteomic study. Our previous study enrolled patients with HCC, CCA, or combined hepatocellular-cholangiocarcinoma, in which a protein biomarker pool containing 57 entities to discriminate the patients from the controls was established and the clinical relevance of different glycosylation patterns on complement C3 in HCC was subsequently analyzed [6]. Although using the same dataset, we used a different strategy here to identify specific plasma biomarkers for HCC or CCA. In addition, only albumin-and IgG-depleted sample fraction rather than whole plasma proteins was used in the mass spectrometry analysis in this study to reduce the interference caused by albumin and IgG. Moreover, simplified sample processing may improve the feasibility of applying these tumor biomarkers in clinical testing.

Cell-cell interaction or adhesion
There have been numerous proteomic studies and diverse tumor biomarker candidates for liver cancer reported. Awan et al. concluded 38 liver-specific secreted or shed protein marker candidates for HCC from seven publicly accessible gene and protein databases [17]. Also, Kimhofer et al. reviewed 22 reports and selected 29 protein biomarkers for HCC [18]. Only one (apolipoprotein CIII) and four (afamin, apolipoprotein B-100, clusterin, and serum paraoxonase/arylesterase 1) common HCC biomarkers were identified when comparing our data to these reviews, respectively. Moreover, haptoglobin is the only common HCC biomarker between these two review papers. From our plasma proteomics, several Ig-related proteins or non-liver-specific proteins were linked to HCC or CCA. It is not surprising to see few common HCC biomarkers among different reports and identify many novel biomarkers in our study because of different sample types, subject enrollment criteria, sample collection,  processing procedures, analytic approaches, and biomarker selection criteria from different works of literature. Combined hepatocellular-cholangiocarcinoma was not included because the number of patients (2 in stage I, 5 in stage II, 3 in stage III, and 2 in stage IV) was too small to acquire precise statistical results. Moreover, we did not include complement C3 on the list of HCC markers in this analytic model. Intriguingly, no ideal markers that perfectly correlated with HCC progression were found. The identified 13 HCC biomarker candidates fluctuated in different tumor stages, and their quantities did not perfectly correlate with the disease severity. The result of the emPAI% calculation shows that these markers are not tumor-specific proteins but possess distinguishable levels during hepatocarcinogenesis. We observed instead that most  of them increased from stage I to II, reduced in stage III, and finished with or without a rebound in the end stage. A similar result was noted using a nonparametric statistical model. Although detecting enhanced levels of these biomarkers resulting from the growth of a solitary tumor from stage I to II with affordable liver functions is plausible, when transitioning to stage III, liver functions might be severely impaired because of expansions of multiple large tumors, thereby contributing to a detrimental effect of protein synthesis. Stage III might see limited production and secretion of biomarker proteins by deep cancerous cells because of central tumor necrosis, a typical feature of advanced HCC. In the final stage, the abundance of biomarker proteins is dependent on the number and size of tumor foci that metastasize to regional lymph nodes or distal organs. These circulating proteins may serve as markers may reflect not only the progression of liver cancer but also the physiological condition of the liver system in patients. One of the limitations of this study is that we could not enroll many patients with late-stage HCC or CCA since only a few patients with advanced-stage HCC or CCA are amenable to undergo surgery. Nonetheless, we found that some of the biomarker candidates decreased in late-stage HCC and identified some biomarker candidates that specifically changed in advanced or metastatic HCC or CCA. These findings, even though they were obtained from a relatively small number of patients, provide significant information as well about the practicability of using these circulating biomarkers to non-invasively evaluate the status and progression of HCC or CCA. Apolipoprotein C-III and serum amyloid P component are another two biomarkers associated with a better prognosis of HCC. Apolipoprotein C-III is a key component in the regulation of triacylglycerol-rich lipoproteins and high-density lipoproteins [19,20], whereas serum amyloid P component, of the pentraxin superfamily, is an acute-phase reactant produced in the liver that activates the classical complement pathway [21,22]. They reflect the capability, at least partially, of the liver on the lipid metabolism and immune activation. Stage-specific analyses revealed that these two proteins showed maximum expression in stage II HCC and continuously reduction afterward, suggesting a decline of liver and immune functions by late-stage HCC. Therefore, not surprising to see a correspondence of low levels of apolipoprotein C-III and serum amyloid P component with a high post-surgical recurrent rate and short survival period in patients with HCC.

Pvalue
Plasma serine protease inhibitor (SERPINA5), mainly synthesized in the liver, is a multifunctional tumor suppressor that reduces the metastatic property of hepatic cancer cell lines by disrupting the fibronectin-integrin signaling pathway [23][24][25]. Repression of this protein has been reported in several cancer types, such as prostate, kidney, ovary, and liver [25][26][27][28][29]. Our result also corroborated this finding and demonstrated a downregulation of plasma serine protease inhibitor in patients with CCA, with a higher level observed in non-metastatic than metastatic CCA. Unlike the trend of HCC biomarker proteins in different tumor stages, plasma serine protease inhibitor decreased gradually with CCA progression. Moreover, it gives impetus to a better 5-year survival of patients with CCA. In contrast to plasma serine protease inhibitor, afamin, a fellow of albumin, a-fetoprotein, and vitamin D-binding protein family [30], seems to be unconducive for the post-surgical outcome of CCA. A low afamin level in serum has been reported in ovarian cancer and CCA [31,32]. We currently do not know the physiological roles of plasma serine protease inhibitor and afamin in cholangiocytes. Future researches should focus on clarifying the molecular mechanisms and pathogenic effects of these two proteins.

Conclusion
Our large-scale label-free, quantitative plasma proteome study identified significant stage-specific and prognostic biomarkers for HCC and CCA. Our findings may provide new insight into clinical application startups using these biomarkers in the diagnosis and follow-up of hepatobiliary cancers. From a clinical perspective, we expect proteomics of liquid biopsy to be added to routine laboratory testing for clinical oncology soon.
Supporting information S1 Table. Characteristics of patients with stage IIIA to IIIC hepatocellular carcinoma. Writing -original draft: Cheng-Hsun Ho.