Characterisation of the Serum Metabolic Signature of Cholangiocarcinoma in a United Kingdom Cohort

Abstract

Background

A distinct serum metabonomic pattern has been previously revealed to be associated with various forms of liver disease. Here, we aimed to apply mass spectrometry to obtain serum metabolomic profiles from individuals with cholangiocarcinoma and benign hepatobiliary diseases to gain an insight into pathogenesis and search for potential early-disease biomarkers.

Methods

Serum samples were profiled using a hydrophilic interaction liquid chromatography platform, coupled to a mass spectrometer. A total of 47 serum specimens from 8 cholangiocarcinoma cases, 20 healthy controls, 8 benign disease controls (bile duct strictures) and 11 patients with hepatocellular carcinoma (as malignant disease controls) were included. Data analysis was performed using univariate and multivariate statistics.

Results

The serum metabolome disparities between the metabolite profiles from healthy controls and patients with hepatobiliary disease were predominantly related to changes in lipid and lipid-derived compounds (phospholipids, bile acids and steroids) and amino acid metabolites (phenylalanine). A metabolic pattern indicative of inflammatory response due to cirrhosis and cholestasis was associated with the disease groups. The abundance of phospholipid metabolites was altered in individuals with liver disease, particularly cholangiocarcinoma, but no significant difference was seen between profiles from patients with benign biliary strictures and cholangiocarcinoma.

Conclusion

The serum metabolome in cholangiocarcinoma exhibited changes in metabolites related to inflammation, altered energy production and phospholipid metabolism. This study serves to highlight future avenues for biomarker research in large-scale studies.

The molecular phenotyping of blood matrices (serum/plasma) has been widely used to generate biomarker fingerprints for cancer diagnostic and screening applications. The measurement of chemical constituents in multiple body fluids provides complementary and more comprehensive snapshots of metabolic phenotypes in healthy and diseased states. The blood metabolome is under tight homoeostatic control and can provide chemical information on a wide variety of molecular species, including small-molecular-weight metabolites and lipids.¹

Dysregulation in phospholipid homoeostasis was a prominent finding in previous studies examining the serum metabolome from patients with cholangiocarcinoma (CCA).² In nonmalignant primary biliary disease, nuclear magnetic resonance (NMR), liquid chromatography (LC)–mass spectroscopy (MS) and gas chromatography–MS metabolic profiling revealed differential serum metabolite signatures from healthy controls, collectively indicative of amino acid perturbation, glycolysis, oxidative stress, lipid peroxidation, cholestasis and inflammation.^{3, 4} A panel of 22 metabolite markers, predominantly free amino acids (such as phenylalanine, serine and aspartate) and several dipeptides, was found to exhibit a high classification power (95% predictive accuracy) for distinguishing between patients with primary biliary cirrhosis (PBC) and primary sclerosing cholangitis (PSC).³ More recently, Hao et al. identified a potential biomarker panel for discriminating patients with PBC from patients with chronic hepatitis and healthy controls using ¹H NMR spectroscopy of serum. The PBC-specific panel, consisting of four metabolites, 4-hydroxyproline, 3-hydroxyisovalerate, citraconate and pyruvate, achieved an area under the curve (AUC) value of 0.81 (95% confidence interval [CI] = 0.87–0.98) in the training set (n = 100), and the findings were reproducible in an independent validation set (n = 37), generating an AUC value of 0.89 (95% CI = 0.74–0.97).⁴

Clinically, primary biliary duct diseases are distinct entities but show some similar features, such as chronic cholestasis, idiopathic nature and high risk of malignant transformation.⁵ Moreover, the cholangiographic distinction between PBC, PSC and malignant biliary strictures is often complicated.⁵ Thus, the differential metabolic phenotypes observed in serum among patients with malignant and benign cholestatic biliary diseases may offer clinical value alongside available diagnostic modalities.

Given the potential for metabolomic profiling to be used in biomarker discovery, we aimed in a pilot study to describe the serum metabolite profile associated with CCA and to compare it for the first time with both benign disease controls (bile duct strictures) and malignant disease controls (hepatocellular carcinoma [HCC]) using hydrophilic interaction liquid chromatography (HILIC) coupled to a mass spectrometer. These MS techniques allow metabolites of interest to be assayed effectively, with an a priori hypothesis that lipids, bile acids, high-energy metabolites and amino acids would be useful to target as potential biomarkers in biliary cancers.

Methodology

Patient and Healthy Volunteer Recruitment

Owing to the rarity of CCA in the UK, serum samples were collected from participating UK liver centres all around England, in London, Manchester, Newcastle, Nottingham, Plymouth and Southampton, and frozen and transported to the Hepatology Biobank at St. Mary's Hospital, London, UK. Potential participants were identified and recruited by their clinician from the inpatient or outpatient population at the various participating institutions.

Healthy volunteers were sought from among the visitors to the hospital, staff and students. After the participants provided consent, they were assessed at baseline for demographic data, medical history, drug history and dietary history. Ethical approval was obtained from Imperial College London, London, UK (REC reference 09/H0712/82).

A total of 47 serum specimens from 20 healthy controls, 8 cases of benign bile duct strictures, 8 patients with CCA and 11 patients with HCC were included. Plain no additive serum (4 mL) was drawn per patient and allowed to clot for 30 min before centrifugation. The clotted serum samples were centrifuged at 4°C at 1000 g for 10 min before 4 aliquots of 250 μL were transferred to 4 Eppendorf tubes (Eppendorf Ltd, Stevenage, UK); of which, one was used for the present study. The samples were deep frozen at −80°C until chromatographic analysis.

Sample Preparation

Before MS analysis, the frozen serum samples were left to thaw overnight at 4°C. Hundred microlitres of the thawed sample was mixed with 300 μL of acetonitrile and then vortexed for a minute and refrigerated for 2 h. The mixture was centrifuged at 4°C at 16089 g for 5 min before a 100-μL aliquot of the supernatant was transferred into 96-well 350-μL plates with cap mats (Waters Corporation, Milford, MA, USA). A sample pool was generated from 20 μL of each sample for sample conditioning before analysis and quality control throughout the analysis. All samples were kept refrigerated at 0–4°C and then at 4°C in the autosampler during the MS analysis.

Instrumentation and Chromatographic Conditions

Ultraperformance liquid chromatography (UPLC) separation was performed on a Waters ACQUITY™ UPLC system (Waters Corporation). An ACQUITY™ UPLC BEH HILIC 2.1 × 100 mm, 1.7 μm column (Waters Corporation) was used to retain and separate polar compounds in serum samples. An Xevo™ G2 QTof (Waters MS Technologies, Manchester, UK) mass spectrometer with an electrospray ionisation (ESI) interface was used to acquire the spectral data in both positive and negative polarities. The data were collected in a centroid mode from 50 to 1200 m/z, a scan time of 0.20 s and an interscan time of 0.014 s. Leucine enkephalin was used for lock mass correction with a scan time of 0.15 s and a scan frequency of 30 s. MS conditions in the positive ion mode were as follows: capillary voltage: 1.5 kV, sample cone voltage: 30 VV, desolvation temperature: 600°C, source temperature: 120°C, desolvation gas flow: 1000 L/hr and cone gas flow: 150.0 L/hr. Conditions in the negative ion mode were as follows: capillary voltage: 1.5 kV, sample cone voltage: 30 V, desolvation temperature: 600°C, source temperature: 120°C, desolvation gas flow: 1000 L/hr and cone gas flow 48 L/hr. MS/MS experiments, using data-dependent acquisition, were carried out with an collision energy ramping from 20 to 40 eV.

The conditions were initially optimised using the quality control (QC) samples in terms of peak shape, reproducibility and retention time. A QC sample was then injected for every 10 samples to monitor the analytical stability throughout the run. An injection of solvent blank samples was used to detect solvent impurities and filter features resulting from contaminants, if present.

Data Processing and Chemometric Analysis

The raw spectral data files were converted to the comma-separated values (CSV) format by MassLynx^TM version 4.1 application manager (Waters Corporation). Peak picking and filtering, correction for retention time drift, peak matching across samples and peak filling of the data were carried out using R Project version 3.1.0 (R Foundation for Statistical Computing, 2014) using XCMS package version 2.14 (Bioconductor: www.bioconductor.org).

Statistical Analysis

SIMCA-P+ version 13.0.2 (Umetrics, Umeå, Sweden) was used for multivariate statistical analysis of the processed data. Initial analysis was performed using unsupervised principal component analysis (PCA) to explore variation in the data set and examine clustering patterns or trends in the data set, based on metabolic profile similarities or differences. After PCA, orthogonal projections to latent structures discriminant analysis (OPLS-DA) was performed to maximise separation between predefined sample classes, to view discriminatory features. Feature selection was based on the variable importance in projection (VIP) coefficients, which allow the X variables to be classified according to their explanatory power of Y (class information). Features with a high VIP value, higher than 1, were found to be the most relevant for explaining Y class information. The top 30 features were selected and identified for each model.

Validating multivariate models is essential to avoid overfitting the data. The model statistics, R²X, Q²Y, permutation test and cross-validated residuals (CV)–analysis of variance (ANOVA) P-value were used to evaluate the model's robustness. Permutation testing (with 100 permutations) was carried out for every OPLS-DA model using SIMCA-P+ version 13.0.2 (Umetrics). Significant variables (metabolites) differentiating between groups were screened, based on a threshold of VIP value (VIP > 2.5) from a 7-fold crossvalidated OPLS or OPLS-DA model. These differential metabolites were subsequently assessed by univariate statistics using one-way ANOVA with post hoc tests (Tukey's honestly significant difference (HSD)) and presented graphically as box-and-whisker plots.

Correlation With Hierarchical Clustering Order

R Project version 3.1.0 (R Foundation for Statistical Computing, 2014) using corrplot package version 0.77 (CRAN) was used to perform hierarchical cluster analysis of Spearman's correlation coefficient matrix. The cluster analysis was used to investigate correlations among the identified biochemical components. The correlation matrix was represented as a heat map with rows and columns ordered according to hierarchical clustering analysis. Hierarchical clustering is an unsupervised method. No class information was given to calculate the model, which was found to be suitable for exploratory data analysis.

Positively and negatively correlated analytes were displayed in blue and red colours, respectively. A circle was used to represent correlations between pairs of compounds. The circle diameter and colour intensity were proportional to the correlation coefficients and indicate statistically significant correlations (<0.05). The circle diameter and colour intensity were proportional to the correlation coefficients. Nonsignificant correlations were represented as X.

Metabolite Identification

Accurate mass was first matched against online spectral libraries such as HMDB,⁶ METLIN⁷ and Lipid Maps⁸ for the tentative assignment of discriminant metabolites. Evaluation of the analyte mass accuracy (within the range of 0–10 ppm), elution time, isotopic pattern and tandem MS fragmentation pattern was used for further structural elucidation.

The identification of lipid classes was based on characteristic and diagnostic fragmentation patterns in tandem MS. Fragment patterns characteristic of the polar head groups typically allow the differentiation between lipid classes. Diagnostic fragments and neutral loss of the fatty acyl chain provide information that permits the structural elucidation of the specific lipid subclass.⁹ Identification of lipid structures based only on mass accuracy is insufficient. As an example, a database search of [M + H]⁺ at an m/z of 834.601 generated 30 matching hits with 0-ppm mass accuracy, including phosphatidylcholines (PCs) and phosphatidylethanolamines.

Structural assignment of lipids was only provided if both characteristic and diagnostic fragments were present in a MS/MS spectrum with an adequate signal. Nevertheless, accurate identification of phospholipid species remains challenging owing to the presence of structural isomers and the formation of different ions with various adducts. Internal standards of lipid species were required for accurate annotation.

Results

Serum HILIC–UPLC–MS Chromatogram

A nontargeted HILIC–UPLC–MS metabonomic approach was applied to characterise the serum metabolome of patients with malignant and benign hepatobiliary conditions in comparison with disease-free controls. A total of 1200 and 688 features were detected in the positive and negative ESI modes, respectively. Figure 1 shows an example of an HILIC chromatogram in the positive (a) and negative (b) ionisation mode. A possible contaminant, trifluoroacetic acid, m/z 112.9 eluting at 5.63 min, was identified in the ESI⁻ run.¹⁰ This contamination was also found in system blanks and was later discarded from the data set.

Figure 1.

Example of metabolites detected in nontargeted hydrophilic interaction ultraperformance liquid chromatography. A chromatogram acquired in the (A) positive and (B) negative electrospray ionisation mode.

Demographics, Clinical Data and Cohort Description

The study included a total of 47 adult participants: 20 healthy individuals, 8 benign biliary stricture cases, 8 patients with CCA and 11 patients with HCC. The age and sex distribution of the enrolled participants was difficult to match between the groups, given the limited number of samples available (Table 1).

Table 1.

Demographic and Biochemical Characteristics of the Study Population.

Characteristic	Healthy	Benign biliary stricture	Cholangiocarcinoma	Hepatocellular carcinoma
Participants, n	20	8	8	11
Age, mean (range)	34.1 (24–58)	64.7 (50–74)	72.7 (57–83)	64.5 (52–76)
Male/female, n	10/10	4/4	1/5	8/3
Serum biochemistry, mean (range)
Bilirubin, μmol/L		19.4 (10–28)	90.1 (55–135)	32.2 (6–104)
Alanine aminotransferase, U/L		124.0 (23–479)	231.2 (78–578)	49.4 (25–81)
Alkaline phosphatase, U/L		195.0 (60–454)	398.7 (159–1012)	179.5 (14–550)
Albumin, g/L		37.3 (31–41)	28.6 (18–42)	37.1 (28–44)
Urea, mmol/L		5.48 (3.9–9.3)	3.6 (2.6–6)	5.8 (3–9)
Creatinine, μmol/L		66.3 (33–93)	65.0 (53–83)	86.3 (65–111)

An initial OPLS-DA was computed using the peak intensity matrix of cases versus controls. Features related to the intake of paracetamol, omeprazole, metronidazole, gliclazide, metformin and warfarin were predominantly found in the serum metabolite profiles from participants with hepatobiliary disease, based on the OPLS-DA VIP projection. Drug peaks and their coeluting ions and adducts were excluded from all subsequent analyses.

Discrimination of Serum Metabolic Profiles Between the Study Groups

Discriminant analysis identified serum metabolite profile differences between the healthy metabotypes, compared with patients with liver pathologies. The metabolic phenotypes of the two malignant groups (HCC vs. CCA) were not statistically distinct from one another (CV-ANOVA P-value > 0.05).

PCA and OPLS-DA score plots and permutation test plots were generated for the comparisons between healthy and disease classes (Figure 2, Figure 3, Figure 4). Biomarker candidates (VIP > 2.5) were subsequently tested for significance using one-way ANOVA analysis and multiple testing correction of the P-values for positive (Table 2) and negative ion mode data (Table 3).

Figure 2.

Healthy controls vs. benign biliary strictures: principal component analysis (PCA) score plots for (A) positive and (B) negative ion mode data of healthy controls and patients with benign biliary strictures. Orthogonal projections to latent structures discriminant analysis (OPLS-DA) score plots showing group separation for both (C) positive and (D) negative ion mode data. The corresponding permutation tests for (E) positive and (F) negative ion mode data. ANOVA, analysis of variance.

Figure 3.

Healthy controls vs. patients with cholangiocarcinoma: principal component analysis (PCA) score plots for (A) positive and (B) negative ion mode data of healthy controls and patients with cholangiocarcinoma. Orthogonal projections to latent structures discriminant analysis (OPLS-DA) score plots showing group separation for both (C) positive and (D) negative ion mode data. The corresponding permutation tests for (E) positive and (F) negative ion mode data. ANOVA, analysis of variance.

Figure 4.

Healthy controls vs. patients with hepatocellular carcinoma: principal component analysis (PCA) score plots for (A) positive and (B) negative ion mode data of healthy controls and patients with hepatocellular carcinoma. Orthogonal projections to latent structures discriminant analysis (OPLS-DA) score plots showing group separation for both (C) positive and (D) negative ion mode data. The corresponding permutation tests for (E) positive and (F) negative ion mode data. ANOVA, analysis of variance.

Table 2.

Discriminant Serum Metabolites Between Normal and Pathological Samples: Positive Ion Mode.

m/z	RT	Tentative assessment	Adduct	Mass error	Chemical formula	Fragments	FDRa	Post hoc testsb
195.086	0.83	Caffeine	M + H	8	C8H10N4O2	138.0, 110.0	0.041	HCC-CCA; healthy-HCC; STRI-HCC
177.101	1.15	Cotinine/serotonin	M + H	6	C10H12N2O		0.012	HCC-CCA; healthy-HCC; STRI-HCC
185.127	1.58	Unknown				126.0, 98.0	0.084
137.045	2.25	Hypoxanthine	M + H	5	C5H4N4O	119.0, 110.0, 94.0	0.036	Healthy-HCC
114.065	3.22	Creatinine	M + H	10	C4H7N3O	72.0	0.029	HCC-CCA; STRI-HCC
247.143	5.18	Hypaphorine	M + H	4	C14H18N2O2	188.0, 146.0, 118.0	0.013	Healthy-CCA; healthy-HCC
166.085	5.94	Phenylalanine	M + H	7	C9H11NO2	120.0, 103.0	0.007	Healthy-CCA; healthy-HCC
118.085	6.40	l-Valine	M + H	10	C5H11NO2	72.0	0.433
144.101	6.48	Proline betaine	M + H	6	C7H13NO2	84.0	0.509
204.122	7.21	l-Acetylcarnitine	M + H	5	C9H17NO4	85.0, 145.0	0.304
162.111	7.58	l-Carnitine	M + H	9	C7H15NO3	85.0, 103.0	0.203
Phospholipid species
834.601	4.81	Phosphatidylcholine	M + H	0	C48H84NO8P	184.0, 146.9	0.0003	HCC-CCA; healthy-CCA; STRI-HCC; STRI-healthy
806.571	4.84	Phospholipid	M + H	1	C46H80NO8P		0.007	HCC-CCA; STRI-HCC
808.585	4.84	Phospholipid	M + H	0	C46H82NO8P		0.184
786.601	4.89	Phospholipid	M + H	0	C44H84NO8P		0.051
784.584	4.89	Phospholipid	M + H	1	C44H82NO8P		0.026	HCC-CCA; healthy-CCA
782.57	4.90	Phospholipid	M + H	0	C40H78NO8P		0.026	HCC-CCA; healthy-CCA
762.591	4.91	Phospholipid	M + H	12	C42H84NO8P		2.75E-05	Healthy-CCA; STRI-CCA; healthy-HCC; STRI-healthy
760.584	4.91	Phospholipid	M + H	1	C42H82NO8P		1.43E-05	HCC-CCA; healthy-CCA; STRI-CCA; healthy-HCC; STRI-healthy
758.57	4.92	Phosphatidylcholine	M + H	0	C42H80NO8P	184.0	0.103
756.555	4.93	Phospholipid	M + H	1	C42H78NO8P		0.046	Healthy-CCA
732.555	4.95	Phospholipid	M + H	1	C10H78NO8P		0.0003	Healthy-CCA; healthy-HCC; STRI-healthy
524.371	5.69	Lysophosphatidylcholines (18:0)	M + H	0	C26H54NO7P	184.0, 104.1	0.0007	Healthy-CCA; STRI-CCA; healthy-HCC; STRI-HCC
542.322	6.06	Lysophosphatidylcholines (18:2)	M + Na	0	C26H50NO7P	184.0, 104.1, 86.0, 520.3	0.0006	Healthy-CCA; healthy-HCC; STRI-healthy

ANOVA, analysis of variance; CCA, cholangiocarcinoma; FDR = false discovery rate; HCC, hepatocellular carcinoma; RT = retention time; STRI, benign biliary stricture group.

^aFDR-adjusted P-value (or q-value) calculated using one-way ANOVA and Tukey's post hoc test.

^bGroups showing statistically significant differences (FDR-adjusted P-value < 0.05); increasing trends are highlighted in bold.

Table 3.

Discriminant Serum Metabolites Between Normal and Pathological Samples: Negative Ion Mode.

m/z	RT	Tentative assessment	Adduct	Mass error	Chemical formula	Fragments	FDRa	Post hoc testb
327.232	0.68	Docosahexaenoic acid	M-H	3	C22H32O2		0.436
367.158	0.71	Sulphated steroid hormone	M-H	1	C19H28O5S		9.28E-06	Healthy-CCA; healthy-HCC; STRI-healthy
369.173	0.73	Sulphated steroid hormone	M-H	3	C19H30O5S		0.0006	Healthy-CCA; healthy-HCC
151.026	1.23	Xanthine	M-H	0	C5H4N4O2		0.423
103.040	1.99	Hydroxyisobutyrate	M-H	0	C4H8O3	83.0, 59.0	0.933
135.031	2.23	Hypoxanthine	M-H	2	C5H4N4O		0.022	Healthy-HCC
Phospholipid species
790.535	4.42	Phospholipid	M-H	5	C45H78NO8P		5.01E-05	HCC-CCA; healthy-CCA; STRI-CCA
766.537	4.44	Phospholipid	M-H	2	C43H78NO8P		0.030	HCC-CCA; healthy-CCA; STRI-CCA
762.506	4.45	Phospholipid	M-H	2	C43H74NO8P		0.002	HCC-CCA; healthy-CCA; STRI-CCA
878.587	4.80	Phospholipid	M-H	5	C49H86NO10P		0.002	HCC-CCA; healthy-CCA; STRI-HCC; STRI-healthy
850.558	4.83	Phospholipid	M-H	2	C47H82NO10P		0.021	STRI-HCC
826.558	4.85	Phospholipid	M-H	2	C45H82NO10P		0.142
828.572	4.88	Phospholipid	M-H	4	C45H84NO10P		0.208
802.558	4.91	Glycerophosphocholine (16:0/18:2)	M-H	2	C43H82NO10P	742.5, 279.2, 255.2	0.373
804.572	4.91	Phospholipid	M-H	4	C43H84NO10P		8.82E-05	Healthy-CCA; healthy-HCC; STRI-healthy
776.542	4.94	Phospholipid	M-H	3	C41H80NO10P		0.031	Healthy-CCA; healthy-HCC; STRI-healthy
564.330	6.12	Lysophosphatidylcholines (18:2)	M-H	0	C26H50NO7P		0.016	Healthy-CCA; healthy-HCC
Bile acid species
514.284	4.69	Taurocholic acid	M-H	0	C26H45NO7S		0.411
448.306	4.93	Glycoursodeoxycholic acid	M-H	1	C26H43NO5		0.118
464.301	5.79	Glycocholic acid	M-H	1	C26H43NO6		0.356

ANOVA, analysis of variance; CCA, cholangiocarcinoma; FDR = false discovery rate; HCC, hepatocellular carcinoma; RT = retention time; STRI, benign biliary stricture group.

^aFDR-adjusted P-value (or q-value) calculated using one-way ANOVA and Tukey's post hoc test.

^bGroups showing statistically significant differences (FDR-adjusted P-value < 0.05); increasing trends are highlighted in bold.

Correlation Between Serum Metabolite Concentrations

Spearman's rank-order correlation was performed to evaluate the interaction among discriminant metabolites detected in positive (Figure 5) and negative (Figure 6) ionisation modes.

Figure 5.

A heat map of Spearman's correlations among discriminant serum metabolites detected in the ESI⁺ mode. The blue circles represent positive correlations, whereas the red ones show the negative correlations. The larger the circle diameter, the higher the correlation (X indicates no significant correlation, P > 0.05). ESI, electrospray ionisation.

Figure 6.

A heat map of Spearman's correlations among discriminant serum metabolites detected in the ESI⁻ mode. The blue circles represent positive correlations, whereas the red ones show the negative correlations. The larger the circle diameter, the higher the correlation (X indicates no significant correlation, P > 0.05). ESI, electrospray ionisation.

Discussion

A distinct serum metabolic phenotype was found to be associated with hepatobiliary disease, globally indicative of dysregulation in lipid metabolism (phospholipid, steroid and bile acid) in the diseased participant groups. However, no absolute separation was observed in serum metabolite profiles between CCA (n = 8) compared with benign biliary stricture (n = 8) and CCA (n = 8) compared with HCC (n = 11), in part owing to the small sample size. Small study numbers reflected the rarity of CCA in the UK, but the study itself is the first of its type and serves as a template for larger, future investigations.

The most prominent difference between the groups was found in the serum concentrations of metabolites that are chiefly markers of lipid metabolism perturbation, including phospholipids, bile acids and steroids. Several lipid species were predominantly elevated in individuals with CCA, relative to HCC and biliary stricture cases, whereas some phospholipids were downregulated in HCC compared with individuals with CCA or with benign biliary strictures. A decrease in lysophosphatidylcholine (lysoPC) and steroid metabolites in serum was implicated in all the three disease groups. LysoPC is a major lysophospholipid generated through phospholipase A2–mediated hydrolysis of PC.¹¹ It has been proposed that lysoPC exhibits proinflammatory activities and is an important mediator of cardiovascular disease and cancer.¹¹

Perturbation of steroid metabolism was observed in CCA, whereas serum phenylalanine levels were elevated in participants with both HCC and CCA. Upregulation of amino acids, including phenylalanine, in liver cancer is possibly related to the higher protein turnover in cancerous tissue, together with altered amino acid uptake and synthesis in liver injury.^{2, 12}

Findings from metabonomic studies in human and animal models suggest that the dysregulation of bile acid and phospholipid homoeostasis is triggered by hepatic inflammatory signalling pathways, rather than being indicative of the tumour status.^{2, 13, 14}

Tanaka et al¹⁴ investigated the serum metabolome and gene expression profile in a diet-induced animal model of nonalcoholic steatohepatitis. The serum metabolite profile of steatotic mice was associated with a significant depletion in lysoPC species (including 16:0, 18:0 and 18:0) and a marked elevation in bile acids, compared with normal mice. The gene expression analysis revealed significant upregulation in hepatic mRNAs encoding enzymes and proteins involved in lysoPC degradation and bile acid excretion, with a marked suppression in genes involved in bile acid uptake into hepatocytes.

In addition, the hepatic mRNA expression of proinflammatory cytokines, such as tumour necrosis factor-α and transforming growth factor-β1, interleukin (IL)-1 and IL-6, was upregulated in mice with steatohepatitis and was found to be closely associated with the changes observed in lipid and bile acid species in serum. Interestingly, the differential serum metabolites and related gene expression profile were independent to dietary variation between animals (choline deficiency) or hepatic steatosis, yet closely associated with steatohepatitis and inflammation. This steatohepatitis-specific phenotype is most likely induced by enhancement of hepatic inflammatory signalling and may apply not only to nonalcoholic steatohepatitis but also to other inflammatory liver diseaises.^{2, 14}

In this study and previous literature, the metabolic pattern observed in cholestatic biliary disease and CCA was in close agreement with the ‘core metabotype’ observed in steatotic and cirrhotic liver disease. Previous bile MS-based metabolic profiling has revealed that biliary lysoPC concentrations were similar among patients with PSC and patients with CCA, potentially indicating that shared metabolic pathways underpin both disorders.¹⁵ Although our investigation supports this suggestion, we cannot be more definitive owing to the size of this study. Bell et al³ identified aberrant lipid and bile acid metabolism in the global serum metabolome of patients with sclerosing cholangitis conditions. Serum markers suggestive of systemic inflammation (such as kynurenine), endogenous oxidative stress and lipid peroxidation (including dimethylarginine and 7β-hydroxycholesterol) were reflected in the serum metabolome of patients with sclerosing cholangitis.³ The pathogenesis of chronic cholestatic biliary disease is idiopathic. However, its manifestation is known to be associated with progressive inflammation of the biliary epithelium along with destruction of intrahepatic and/or extrahepatic bile ducts.¹⁶

Attenuation of circulating lysoPC appears to be influenced to some degree by other solid tumour entities. Taylor et al¹⁷ examined lysoPC patterns and their relationship with parameters of inflammation, including C-reactive protein (CRP) and whole blood hydrogen peroxides (HPOs), and nutritional status, such as weight loss, body composition and body mass index (BMI), in a group of 59 patients with cancer (including cancer of the breast, lung and gastrointestinal tract). The ranges of lysoPC concentration in patients with cancer were found to be lower or near the lower value of the reference interval reported for a healthy population. Measures of oxidative stress (HPO) and inflammatory activation (CRP) were found to be inversely correlated with lysoPC concentrations. The relationship between lysoPC levels and inflammatory markers were found to be stronger in a number of subgroups, suggesting that cachexia, systemic inflammation and oxidative stress are codependent and closely interrelated processes with lipid peroxidation in malignant disease.

These results suggest that the attenuation of lysoPC species in blood may be investigated further as a marker pertaining to the severity of malignancy in general. It is unclear whether lysoPC displays shared or distinct patterns between liver tumours and other tumour entities. Nevertheless, the liver is the major site of lysoPC biogenesis, and phospholipid catabolism has been found to be impaired in the early stages of liver disease progression.^{2, 14} Overall, the literature is inconclusive, and comprehensive studies using a targeted quantitative lipidomic approach are required to validate cancer-specific metabolite abundance patterns.

In the literature, marked elevations in the relative abundance of a number of phospholipid molecular species have been associated with cholestatic liver disease, particularly in malignant conditions. Yet downregulation of phospholipids/PCs in bile was elucidated in the early CCA metabonomic studies.^{18, 19, 20, 21} It is postulated that genetic variations in biliary transporter proteins, such as multidrug-resistant protein 3 (MDR3), mediate the reduction in biliary phospholipid secretion in cholangiopathic disorders.^{20, 22}

ABCB4, the gene that encodes for MDR3, is an ATP-binding cassette (ABC) transporter in the canalicular membrane. In the liver, ABCB4 is responsible for the secretion of phospholipids into bile; it facilitates the translocation of phospholipids, preferentially PC, from the inner to the outer leaflet of the hepatocanalicular membrane.²³ Biliary phospholipids preserve the integrity of the biliary epithelium against the cytotoxic detergent effect of bile salts (as monomers and as simple micelles) by facilitating the formation of mixed micelles (bile salt–cholesterol–phospholipid mixtures), which strongly attenuates the toxicity of the bile salt pool.²³

Further support for the involvement of the ABCB4 gene in the pathogenesis of cholestatic liver disease was presented in recent studies. A mutant ABCB4 genotype was detectable in 18 of 90 patients with cholangiopathy (20.0%; 95% CI = 12.3–29.8%). The ABCB4 mutation–related PSC phenotype was found to be more frequent in patients with a personal or family history of other cholestatic liver disease (28.6 vs. 8.7%) and was associated with earlier ages at diagnosis.²⁴ Emerging direct evidence revealed that nonsynonymous polymorphic variants of ABCB4 have deleterious effects at the protein level (e.g., low expression levels, maturation defect and retention in the endoplasmic reticulum).²⁵ The mutants compromised the ABCB4-mediated efflux of phospholipids, resulting in a significant reduction in PC secretion from cells.

The ABCB4 genotype is likely to underlie the pathophysiology of cholestatic liver disease in a specific subpopulation of patients. A low or absent ABCB4 expression may therefore favour the formation of ‘toxic bile’, as a result of decreased biliary phospholipid secretion, and aggravate inflammatory bile duct injury. Chronic cholestasis and cholangitis may further drive secondary sclerosis or cirrhosis and predispose to hepatobiliary cancer. This proposed pathogenetic mechanism implies that future therapeutic strategies that address the underlying cholestasis by targeting specific and potent nuclear receptor ligands may provide a window of opportunity for cholestatic liver disease management and treatment.

A serum metabolic panel, including metabolites involved in lipid metabolism (phospholipids, bile acids and steroids) and amino acid metabolism, distinguished between healthy controls and individuals with hepatobiliary diseases. Markers related to inflammation (lysoPC species and bile acids), hepatobiliary injury (steroid-related metabolites) and high protein turnover (phenylalanine) were characteristic of liver disease, particularly in CCA. This pilot study did not find any robust multivariate differentiation to be observed between CCA and either benign biliary strictures (as a nonmalignant control) or HCC (as a malignant control). Yet the abundance of a number of phospholipid metabolites indicated metabolic abnormalities across the disease groups. The results of this small pilot study may not have unearthed any useful biomarker but do point to underlying metabolic disturbances, which may provide further insight into the homoeostatic imbalances that occur in hepatobiliary cancer development.

Further studies are needed to gain better understanding of these metabolic alterations throughout the oncogenic processes in liver disease. A targeted metabonomic approach would be ideal as it provides both qualitative and quantitative analysis of the metabolites of interest.

Conflicts of interest

The authors have none to declare.