2.1 Chemicals and reagents

Analytical grade solvents were used for LC-MS analysis. Solvents and reagents included methanol, formic acid (Daejung chemicals and metals, Siheung, Korea), acetonitrile (Tedia, Tedia way, Fairfield, USA), N-Fmoc-L-alanine (Chem-Impex International, Wood Dale, IL, USA), N-Fmoc-L-serine (OtBu) (GL Biochem, Shanghai, China). Throughout the study, deionized water (Milli-Q) 18.2 MΩ cm, obtained from Millipore assembly (Billerica, MA, USA), was utilized.

2.2 Patient selection

Hepatitis C patients, male/female, with HCC or without HCC were recruited from the out-patients department (OPD) of the National Institute of Liver and GI Diseases (NILGID), and Blood Collection Point, Dow Diagnostic Reference and Research Laboratory (DDRRL), Dow University of Health Sciences, Ojha Campus, Karachi. The control group consisted of normal healthy individuals. Written informed consent was obtained from all the 147 participants with the approval of the Institutional Review Board of Dow University (Protocol No. IRB-489/DUHS/-14) before their enrollment in the study. Independent Ethics Committee of International Center for Chemical and Biological Sciences had approved the experimental protocol (ICCBS/IEC-040-HB-2018/Protocol/1.0). Clinical history of the patient including complete blood count (CBC), prothrombin time, Hepatitis Viral Load, α-fetoprotein (AFP), Liver function tests, and Albumin were determined. Exclusion criteria included: Anti-HCV negative, HBsAg positive, history of alcoholic abuse and cirrhosis, HCV co-infection, and patients having evidence of any other liver inflammatory or some additional tumor at the same time with HCC.

2.3 Sample collection

Sample collection was done after proper evaluation, medical checkup, and checking previous records/history under the supervisionof a medical consultant. None of the participants had received any kind of treatment prior to sample collection. About 5 mL of blood was taken from patients and healthy controls by venipuncture and transferred to gel-based BD vacutainer tubes (BD Franklin Lakes NJ, USA, REF: 367381), interior coated with silicone for clot activation, then centrifuged for 5 min at 3500 rpm to separate the serum. The serum portion was immediately stored at −80℃ until further analysis.

2.4 Sample preparation

The samples were prepared as reported earlier with some modifications (Chen et al., 2011b). The frozen serum samples were thawed on ice bath and an aliquot of 100 µL of serum sample was added with 100 µL of water (containing 1.0 mg/ml each of N-Fmoc-L-alanine and N-Fmoc-L-serine (OtBu) as internal standards), followed by the addition of 400 µL of a mixture of methanol and acetonitrile (5:3). Samples were vortexed for 2 min and left at room temperature for 10 min. After centrifugation at 12,000 rpm for 20 min, the supernatant was separated and then dried completely under vacuum at room temperature and reconstituted in 200 µL of milli-Q deionized water. After vortexing for 2 min, the mixture was centrifuged at 12,000 rpm for 20 min and the supernatant was transferred to the autosampler vial for analysis. A total of 16 blind samples were also prepared for validation in the similar way as mentioned above.

A separate pool referred to as the quality control (QC) was made by mixing 10 µL aliquot of serum from each sample by following the same method and was subjected to analysis periodically after every 6 samples to provide robust quality assurance for each metabolic feature detected.

2.5 UPLC-QTOF acquisition

LC-MS analysis was performed on Bruker maXis II HR-QTOF mass spectrometer (Bremen, Germany) consisting of Thermo Scientific Dionex Ultimate 3000 liquid chromatography system coupled with QTOF mass spectrometer. A 5 µL aliquot of the sample was injected at a random order into a reverse phase, 2.0 mm × 100-mm, 1.8 µm particle, NUCLEODUR C18 Gravity column with a compatible guard column (Macherey-Nagel, Germany) held at 40 °C. For chromatographic separation mobile phase, A was water 0.1% formic acid and B was methanol with 0.1% formic acid. The column was kept at 5% B for a minute then the sample was eluted with a linear gradient of 5–95% B over 1–9 min, 95% B over 9–10.5 min, 5% B over 10.7 min, 5% B over 10.7–13 min. The flow rate was 0.5 mL/min. 0.1 M Na-formate calibrant solution was injected prior to each injection at the rate of 0.17 mL/hr using automatic switch valves. The sample temperature of 4 °C was maintained during the analysis.

The parameters of mass spectrometry for MS and MS/MS data were adjusted as follow, drying gas temp of 270 ℃, drying gas flow of 12.0 L/min, mass range of 100–1200 m/z, capillary voltage of 4500 V, and nebulizer pressure 3.1 bar. Data was recorded using Compass HyStar 4.1 acquisition software (Bremen, Germany). To condition the column, assess instrument stability and to ensure data quality, 10 QC’s were subjected at the start of each analytical batch and then one QC is injected after every 06 samples and 05 blanks were randomly injected to assess sample carryover. A column check was used to check the stability of the column before and at the end of each batch.

2.6 Data processing and statistical analysis

First the calibration of data was performed using Bruker’s Compass Data Analysis (version 4.4). After calibration features were extracted using Bruker’s MetaboScape (version 3.0) software with following parameters: 3000 counts intensity threshold, 12 spectra minimum peak length, 9 spectra minimum peak length (recursive), feature signal intensity, minimum numbers of features for extraction and presence of features is 30% of analyses, 0.3–0.8 EIC correlation, 13 min retention-time range, 100–1500 m/z mass range, and [M + H]⁺primary ion. N-Fmoc-L-serine (OtBu) and N-Fmoc-L-alanine as internal standards were used for checking chromatographic performance and quality control, while the intensity of N-Fmoc-L-serine (OtBu) was used for data normalization. The extracted features were exported to the .csv format and processed on Mass Profiler Professional (MPP) software 12.5. Data filtering involve 5,000 counts minimum absolute abundance with 3 number of ions minimum. Z transform base line option was selected to treat all the compounds equally irrespective of their intensity, and healthy serum set as control parameter against HCV infected, HCV induced CLD, and HCV induced HCC. All noteworthy obtained features were also checked individually in all sample files to ensure the correctness of the results.

The obtained resulting data from MetaboScape were examined by univariate student T-test (volcano plot) and multivariate statistical tools. Import of the data to SIMCA (version 15.0, Umetrics, Umeå, Sweden) was carried out for the distribution of distinct groups and detection of outlier by principal component analysis (PCA) whereas overview of after mean centering and Pareto scaling was performed by orthogonal partial least squares discriminant analysis (OPLS-DA)/ orthogonal partial least squares discriminant analysis (OPLS-DA) was performed after mean centering and Pareto scaling to show metabolic alteration in the development of HCV to HCC. A permutation randomization test was also performed in which the distribution of the test statistic under the null hypothesis is obtained. Metabolite identification was done by comparing MS/MS data of significantly distinctive features observed from the volcano plot with the NIST tandem mass spectral library and HMDB.

3.1 Metabolic profiling and identification

MetaboScape software generated 2599 features in all samples. These features on filtering, followed by ANOVA, results in 129 features showing significant up/down-fold changes in HCC, 96 in CLD, and 52 in HCV infected in contradiction of healthy control generated with Benjamini Hochberg FDR at<0.05p-value and fold change > 1.5. Volcano plot also showed notable features between HS vs HCV, HS vs CLD, and HS vs HCC. After combining the two lists, 13 features have been commonly identified by comparing the MS/MS spectra with those available in the NIST tandem mass spectral library and 05 features are identified by HMDB, based on their MS/MS spectra while 03 features were annotated on the basis of exact mass only, shown in Table 2.

3.2 Chemometric analysis

Various uni/multivariate statistical analyses were performed for the interpretation of data. The univariate analysis that is volcano plot were plotted on the features observed from MetaboScape software between HS vs HCV (Fig. 1A), HS vs CLD (Fig. 1B), and HS vs HCC (Fig. 1C). The plots presenting orange dots are significantly different from healthy controls by log fold change limit > 1.5 or p-value < 0.05 shown in Supplementary Table 1. Besides the univariate analysis of data, hierarchical clustering was also generated for the clustering of study groups using Mass Profiler Professional (MPP) software 12.5. A dendrogram was generated for the 21 significantly different metabolites (Fig. 2) using their normalized and averaged intensities by applying Harmonic distance metric and complete linkage parameters. Among these 18 were identified and 3 were annotated. The four subject groups were categories into two levels; in level 1, HCC and HCV groups were clustered together with 1.8 × 10^-4 difference, while in level 2, CLD and HS were at dissimilarity level of 5.9 × 10^-3. On the whole, these two levels were dissimilar from each other with a value of 1.3 × 10^-2.

Close

Fig. 1

Volcano plots showing significantly distinctive features. (A) HS vs HCV, (B) HS vs CLD, and (C) HS vs HCC.

Export to PPT

Close

Fig. 2

Comparison of four groups; HS, HCV, CLD, and HCC by hierarchical clustering for the 18 identified and 3 annotated features that were found significantly different.

Export to PPT

For the multivariate statistical analysis, SIMCA software was used. The PCA scores plot (Supplementary Fig. 2) shows the maximum variability within the groups based on disease category, gender, and age; and doesn’t show any trend of separation between groups. However, the OPLS-DA model (Fig. 3) reveal the evident separation between the all four disease groups after excluding the outliers observed in the PCA by hotelling’s T2 plot (Supplementary Fig. 3). To check the stability of instrument and analysis reproducibility, PCA was generated with QC data (Supplementary Fig. 4) showing all QC pools were lie between the samples.

Close

Fig. 3

OPLS-DA score plot showing evident separation between all four groups, HS (square), HCV (circle), CLD (triangle), HCC (4-point star).

Export to PPT

Internal validation by the ROC curve for OPLS-DA was plotted between the true positive rate (TPR, sensitivity) and false positive rate (FPR, 1-specificity). Our study shows 100% sensitivity; while HCV, CLD, and HCC showed 98.5%, 100%, and 90.9%, specificity respectively. The area under curve values for HS, HCV, CLD, and HCC were found to be 0.99, 1, 1, and 0.89, respectively (Supplementary Fig. 5). The permutation randomization test (500 times) of the OPLS-DA model including correlation coefficient between the original Y and the permuted Y versus the R2 (cum) and Q2 (cum) of HS, HCV, CLD, and HCC with the regression line shown in Supplementary Fig. 6 (A-D). The indication for the validity of the original model is that R2 original point to the right is higher than the permuted values to the left. The intercept (R2 and Q2 when correlation coefficient is zero) for HS, HCV, CLD, and HCC are R2 = 0.496, 0.491, 0.5, 0.509 and Q2 = -0.347, −0.317, −0.312, −0.321 that is very small which shows the model is satisfactory. External validation of the method was done using 16 blind samples and validated using above mentioned internally validated OPLS-DA model (Supplementary Fig. 7). The misclassification table for OPLS-DA (Supplementary Table 2) shows that all samples were correctly classified in their own group except 1 blind sample from the HCV group which classified as HCC samples, this might due to the metabolic profile for the sample which start to shift towards carcinoma.

3.3 Differential metabolites in diseased and healthy control

It has been reported previously that HCC is advanced from CLD (liver cirrhosis) which primarily progresses from HCV infection (Axley et al., 2018), however, it is very hard to identify HCC at an initial stage. Considerable changes were observed in the concentrations of several metabolites in 03 stages of liver disease compared with healthy controls (Fig. 4 and Supplementary Fig. 8) that seems differently in the development of the disease. A group of metabolites 3-indolepropionic acid, LPC 14:0 (1-myristoyl-sn-glycero-3-phosphocholine), LPC 16:0 (1-palmitoyl-sn-glycero-3-phosphocholine), LPC 18:2 (1-(9Z,12Z-octadecadienoyl)-sn-glycero-3-phosphocholine), LPC 17:0 (lysophosphatidylcholine 17:0), PAF C-16, L-tryptophan were found considerably down regulated in CLD as compare to healthy controls. Whereas, 1-methylguanine, glutamyltyrosine, glycocholic acid, unknown (carbon number 11) (ID # 1370), 1,20-eicosatetraenedioic acid shows up-regulation in CLD as compared to HS. N-fructosyl tyrosine and hydroxyindoleacetic acid showed an increase in level from HCV to HCC whereas L-aspartyl-L-phenylalanine and thyroxine showed a decrease in level from HCV to HCC in comparison with HS. L-phenylalanine is up-regulated in HCC and C-17-sphinganine was found down-regulated in HCC as compared to HS. However, glycerol-3-phosphocholine, Thr-Cys-Arg, and dioleoylphosphatidylcholine do not show any clear alteration in their levels.

Close

Fig. 4

Box-and-Whisker plots of identified metabolites showing the progression of the disease. (A) Hydroxyindoleacetic acid, (B) N-fructosyl tyrosine, (C) L-aspartyl-L-phenylalanine, and (D) Thyroxine.

Export to PPT

3.4 Pathway analysis

MetaboAnalyst 4.0 online software (www.metaboanalyst.ca/) was used to identify the biological pathways that are involved in the metabolism and progression of HCV to HCC. About 10 pathways (Fig. 5) were complemented according to the p values, impact factor and FDR value from the pathway enrichment analysis by KEGG metabolic pathway database and HMDB. Consequently, 6 out of 10 identified pathways (glycerophospholipid metabolism, tryptophan metabolism, phenylalanine, tyrosine, tryptophan biosynthesis, aminoacyl-tRNA biosynthesis, phenylalanine metabolism, and tyrosine metabolism) (Chen et al., 2011b, Fitian et al., 2014, Gao et al., 2015) had been reported earlier, however, we have observed alterations only in glycerophospholipid metabolism (Supplementary Fig. 9), phenylalanine, tyrosine, tryptophan biosynthesis (Supplementary Fig. 10), phenylalanine metabolism (Supplementary Fig. 11) and tryptophan metabolism (Supplementary Fig. 12) out of 6 reported pathways with FDR value <1 and p-value < 0.05.

Close

Fig. 5

Summary of the biological pathways involved in the liver disease, analyzed by MetaboAnalyst 4.0 Software.

Export to PPT