Analytical and Bioanalytical Chemistry

, Volume 403, Issue 1, pp 203–213 | Cite as

Serum metabolomics reveals the deregulation of fatty acids metabolism in hepatocellular carcinoma and chronic liver diseases

  • Lina Zhou
  • Quancai Wang
  • Peiyuan Yin
  • Wenbin Xing
  • Zeming Wu
  • Shili Chen
  • Xin Lu
  • Yong Zhang
  • Xiaohui Lin
  • Guowang Xu
Original Paper

Abstract

Patients with chronic liver diseases (CLD) including chronic hepatitis B and hepatic cirrhosis (CIR) are the major high-risk population of hepatocellular carcinoma (HCC). The differential diagnosis between CLD and HCC is a challenge. This work aims to study the related metabolic deregulations in HCC and CLD to promote the discovery of the differential metabolites for distinguishing the different liver diseases. Serum metabolic profiling analysis from patients with CLD and HCC was performed using a liquid chromatography–mass spectrometry system. The acquired large amount of metabolic information was processed with the random forest–recursive feature elimination method to discover important metabolic changes. It was found that long-chain acylcarnitines accumulated, whereas free carnitine, medium and short-chain acylcarnitines decreased with the severity of the non-malignant liver diseases, accompanied with corresponding alterations of enzyme activities. However, the general changing extent was smaller in HCC than in CIR, possibly due to the special energy-consumption mechanism of tumor cells. These observations may help to understand the mechanism of HCC occurrence and progression on the metabolic level and provide information for the identification of early and differential metabolic markers for HCC.

Keywords

Hepatocellular carcinoma Metabolomic profiling Random forest–recursive feature elimination Acylcarnitine Fatty acid oxidation 

Abbreviations

Q-TOF

Quadrupole time-of-flight

AFP

α-fetoprotein

ALT

Alanine transaminase

AST

Aspartate transaminase

CEA

Carcinoma embryonic antigen

CHB

Chronic hepatitis B

CIR

Hepatic cirrhosis

CLD

Chronic liver diseases

CN

Acylcarnitine

CoA

Coenzyme A

CPT 1

Carnitine palmitoyl transferase 1

CPT 2

Carnitine palmitoyl transferase 2

FAO

Fatty acid oxidation

GCA

Glycocholic acid

GCDCA

Chenodeoxycholic acid glycine conjugate

HBsAg

Hepatitis B surface antigen

HBV

Hepatitis B virus

HCC

Hepatocellular carcinoma

IDO

Indoleamine 2,3-dioxygenase

LC-MS

Liquid chromatography–mass spectrometry system

N

Healthy controls

OOB

Out-of-bag

QC

Quality control

RF

Random forest

RFE

Recursive feature elimination

RF-RFE

Random forest–recursive feature elimination

RRLC

Rapid-resolution liquid chromatography

SCD

Stearoyl-CoA desaturase

T2DM

Type 2 diabetes

TCA

Tricarboxylic acid

γ-GT

γ-glutamyl transpeptidase

Introduction

Hepatocellular carcinoma (HCC) has been demonstrated to be strongly related to chronic hepatitis virus infection [1] and is a major cause of death among patients with cirrhosis (CIR) [2]. Especially in China, most of HCC patients have the disease background of chronic hepatitis B (CHB) or further CIR. The pathological changes of liver caused by hepatitis virus are complicated. It begins with the cycle of hepatic injury and regeneration, and then to more severe abnormal liver nodule formation (cirrhosis), and finally to the precancerous lesions of HCC [3]. And in the early stage of HCC, there are large scale of hepatic injury and cirrhosis around the malignant lesions. This makes it difficult to identify early HCC from chronic liver diseases (CLD).

Despite increasing studies of genes and proteins involved in the occurrence and development of HBV infection-related HCC [4, 5, 6, 7, 8, 9], the knowledge on pathogenesis and diagnosis of this malignant tumor remains scarce. As a major organ of metabolism in the body, liver takes part in many metabolic pathways and also has a great impact on the systemic state of the organism. And metabolic dysregulation in liver may influence the incidence and development of HCC [10]. For example, the activation of lipogenic pathway was found contributing to cell proliferation and prognosis [10] in HCC patients.

Through screening as many metabolites as possible, metabolic profiling or metabolomics can provide a large amount of endogenous metabolite information, facilitating a systematic view of metabolic deregulations [11, 12]. Metabolomics has already been used in the study of pathogenesis of liver diseases or malignant hepatic tumors and the discovery of novel diagnostic markers based on the comprehensive analysis of differentially expressed metabolites. It has been reported that there is an increase of total lipid profile and a decrease of lipid polyunsaturation in liver biopsies with increasing hepatic fibrosis in patients with chronic hepatitis C [13]. Differentially expressed metabolites were found in urine of hepatitis C virus infected Egyptian patients with HCC. Among the differential metabolites, glycine, citrate, carnitine, and trimethylamine-N-oxide are related to energy production and aberrant chromosomal methylation [14]. Recenly, γ-glutamyl dipeptides have also been reported as biomarkers to discriminate six different CLD and HCC [15]. And also, some important serum metabolites have been found altered in HCC and CIR involved in lipid metabolism, increased protein degradation, perturbations of the tricarboxylic acid (TCA) cycle [16]. Importantly, when comparing the serum metabolic changes of CIR and HCC in our previous study, we found not only several differential metabolites specific to CIR and HCC but also different metabolic correlation networks [17].

Generally, rich metabolome data can be collected by the metabolic profiling analysis based on liquid chromatography-mass spectrometry [18, 19, 20], gas chromatography–mass spectrometry [21] or NMR [11]. In order to mine important deregulation information and find out metabolic features, an effective multivariate analysis method is necessary. Random forest (RF) [22] is a very efficient multivariate analysis technique, and has been successfully applied in many fields such as text mining, genomics, and metabolomics [23, 24, 25]. It can efficiently process large number of variables and small samples, such as metabolome data, due to its two randomness [22]. Furthermore, it can deal with classification problems as well as feature selection problems. RF could measure the variable importance through “Gini importance” [26] or “permutation importance” [27]. Random forest–recursive feature elimination (RF-RFE) which combines RF with RFE [28] is a recursive backward feature elimination procedure. It begins with all the features. In each iteration, a random forest is constructed to measure the features’ importance and the feature which is the least important is removed. This procedure is repeated until there is no feature left. Finally, the features are ranked according to the deleted sequence, the top ranked feature is the last deleted and the most important.

In this study, the samples collected from CHB, CIR, and HCC patients and healthy controls were analyzed on a LC-MS. RF-RFE was adopted to select the informative variables from the serum metabolome data. The study aimed at the exploration of important metabolic deregulations related to the CLD and HCC in order to provide useful pathogenesis information and facilitate the discovery of differential metabolites for different liver diseases.

Experimental

Reagents and chemicals

Acetonitrile and formic acid (HPLC grade) was purchased from Merck (USA), and Tedia (USA), respectively. And home filtered Milli-Q water was used as solvent or for preparing mobile phases. The standard compounds were obtained from Sigma-Aldrich (St. Louis, MO, USA).

Sample collection

Fasting sera were collected from 30 healthy volunteers and 90 patients with liver diseases (30 CHB, 30 CIR, and 30 HCC) from the Sixth People’s Hospital, Dalian. The healthy controls were confirmed by normal liver function and negative for hepatitis B surface antigen (HBsAg). All of the enrolled HCC patients were accompanied with CIR, and the age and sex of each group were matched. The CIR and HCC patients were diagnosed on the basis of tumor markers (AFP and CEA), liver function tests, serum and urine biochemical tests, ultrasonography and CT, or MRI results. All of the CHB patients were positive with HBsAg and all the patients enrolled have taken liver function tests to investigate the serum content of alanine transaminase (ALT), aspartate transaminase (AST), and γ-GT, which can indicate the extent of liver damage. The detailed clinical parameters were listed in Table 1. About 1 ml serum from each person was collected and stored at −80 °C immediately until analysis.
Table 1

Baseline characteristics of patients with liver diseases

Characteristics

N (n = 30)

CHB (n = 30)

CIR (n = 30)

HCC (n = 30)

Normal range

Age

40 ± 2

46 ± 2

55 ± 2

59 ± 2

Sex

  Male

22

20

19

21

  Female

8

10

11

9

AFP

2.4 ± 0.16

42.2 ± 17.91

15.5 ± 6.21

246.9 ± 77.94

0–7

CEA

1.8 ± 0.22

31.3 ± 29.16

3.4 ± 0.39

8.4 ± 3.2

0–6.2

ALT

31.6 ± 4.91

167.3 ± 32.1

44.3 ± 4.9

56.7 ± 7.67

0–50

AST

25.6 ± 1.83

113.4 ± 22.88

59 ± 6.45

86.2 ± 11.16

0–40

ALP

90.2 ± 4.68

101 ± 6.15

128 ± 12.89

125.6 ± 13.69

40–150

r-GT

54 ± 16.15

93.1 ± 10.68

113.7 ± 36.88

107.7 ± 17.4

7–50

ALB

45.1 ± 1.19

39.9 ± 1.16

31.7 ± 1.03

34.6 ± 1.79

35–53

GLOB

27.5 ± 0.66

31.6 ± 1.26

31.7 ± 1.28

32.3 ± 2.14

21–33

TP

73.7 ± 0.68

68.9 ± 2.99

63.3 ± 1.65

60.3 ± 3.4

61–83

A/G

1.7 ± 0.05

1.9 ± 0.63

1.1 ± 0.06

8.6 ± 4.36

1.5–2.5

T-BIL

17.1 ± 1.23

27.9 ± 7.9

52.1 ± 10.7

134.4 ± 77.44

3.4–20

TBA

318.1 ± 283.92

63.7 ± 8.79

112.5 ± 44.82

0–20

CHE

8 ± 0.39

6,505.4 ± 579.74

2,334.7 ± 204.15

3,357.4 ± 411.24

3,930–11,500

ALP alkaline phosphatase, r-GT γ-glutamyl transpeptidase, ALB albumin, GLOB globulin, TP total protein, A/G the ratio of albumin to globulin, T-BIL total bilirubin, TBA total bile acids, CHE cholinesterase

Sample preparation for analysis

Four hundred-microliter acetonitrile was added to 100-μl serum after thawing at room temperature. Each mixture was votexed vigorously and then centrifuged at 15,000×g for 10 min at 4 °C. About 100-μl supernatant was moved to autosampler vial for analysis. Ten-microliter supernatant from each sample was mixed as the quality control (QC) sample [29]. The QC was run ten times to equilibrate the liquid chromatographic system before the batch analysis and was run every ten samples to monitor the stability of instrument during the batch analysis [29].

Liquid chromatographic separation

The supernatant from each sample was injected into an Agilent 1200 Rapid Resolution Liquid Chromatography (RRLC) system (Agilent, USA) for the metabolic profiling and fatty acid analysis. A reversed-phase HSS T3 column (100 × 2.1 mm, 1.7 μm; Waters, USA) was used for chromatographic separation. The mobile phase A was of 0.1% formic acid in water (v/v) and mobile phase B was of 0.1% formic acid in acetonitrile (v/v). The gradient started with 2% mobile phase B, increasing to 30% B at 3 min, then to 80% B at 22 min, and then to 100% B at 24 min. The total run time was 35 min including equilibration of 5 min. The flow rate was 0.3 ml/min, and the injection volume was 4 μl.

Mass spectrometry analysis

Data acquisition was performed on an Agilent 6510 quadrupole time-of-flight (Q-TOF) mass spectrometer (Agilent, USA) equipped with dual-electrospray source which was operated in the positive ion mode for the metabolic profiling and in the negative ion mode for the fatty acid analysis, separately. The ion source temperature was set at 300 °C and other parameters were set as our previous protocol [17]. The mass accuracy was calibrated before analysis and a real-time calibration of mass accuracy was implemented to ensure the stability of mass accuracy. Mass spectra data were collected in the centroid mode with acquisition rate of 2 spectra/second. The scan range was from m/z 100 to 1,000. And at the end of a sequence in each ion mode, data-dependent information acquisition was performed to acquire MS/MS fragments with the collision energies set at 10 and 30 V, respectively.

Data analysis

Data pretreatment

The raw total ion current spectra acquired from RRLC/Q-TOF MS were directly analyzed by the Molecular Features Extraction from Agilent to pick compounds. Then, the peak alignment was done on the export files by GeneSpring MS 1.2 (Agilent, USA). After these steps, an excel table including retention time, m/z, peak intensity was given. The ions in the table were kept if they exist in 80% samples of at least one group [30]. With this method, most of the missing values can be removed. And then the remaining variables were imported to SIMCA-P 11.0 (Umetrics AB, Umea, Sweden) for principal component analysis (PCA) to evaluate the reliability of the obtained profiling data.

Multivariate analysis and feature selection

RF-RFE was performed on the constricted excel table to select the informative ions. The algorithm was written in C++. Each decision tree in the random forest is constructed as a binary tree, its training set is produced by the bootstrapping technique and one sample can be repeated at most two times in each training set. To value the variable’s splitting effect, we adopt distinct class based splitting measure [31], which consists of two parts: one is related to the number of sample classes in each child partition and the other is related to each class’s proportion in the partition. To find meaningful metabolites which can discriminate liver diseases from the control, discriminate among HCC, CHB and CIR, further discriminate between each two different liver diseases, RF-RFE was conducted on five cases, labeled with RF-RFEi (i = 1, 2, 3, 4, and 5) in sequence. RF-RFE1 considers CHB, CIR, and HCC as one liver disease group, trying to select variables which have a significant difference between the disease group and the control group. RF-RFE2 compares three liver diseases, aiming at finding discriminative variables for three different liver diseases. And the remaining three cases were conducted as the binary classification problems among three liver diseases.

Each RF-RFEi (i = 1, 2, 3, 4, and 5) was run 50 times. For each run, one third of the data was randomly selected as an independent test set and the remaining two thirds was used as the training set to conduct the feature selection. The procedure starts from all the input variables on the training data. In each loop, a random forest is constructed based on the training data, the features’ scores are calculated by the permutation importance [27], and the feature having the lowest score is deleted. The procedure is repeated until there is no feature left. In each iteration, the performance of the random forest is tested by the out-of-bag (OOB) data, the maximal OOB accuracy rate and its corresponding random forest model are kept. When the iteration terminates, a feature rank is produced according to the deleting sequence of the features, the top-ranked features are the last deleted and the most important ones. The occurrence frequency of each variable in the top 50 ranked features of each run is calculated. The variables whose frequencies are bigger than 0.5 are selected. The feature subsets selected by RF-RFEi are denoted as FSSi (i = 1, 2, 3, 4, and 5), respectively. The final feature subset is a union of FSSi (i = 1, 2, 3, 4, and 5).

Univariate analysis: the informative ions were imported into SPSS 11.5 for Mann–Whitney U test analysis. A significant level of p < 0.05 was chosen.

Feature ion identification

Identification of the remaining differential expression ions was performed according to our former work [32] with the small modification. Firstly, quasi-molecular ions were picked out from their product and adduct ions. And then online MS/MS spectra databases such as Metlin (http://metlin.scripps.edu/) or HMDB (http://www.hmdb.ca/) were searched using the exact mass with a mass deviation of 0.1 Da. If an ion was not in the databases, accurate mass with mass deviation smaller than 3 ppm was measured to calculate its element composition, and multi-stage MS fragmentation was obtained to deduce the possible chemical structures. Also, the fragments of possible endogenous compounds searched from Chemspider (http://www.chemspider.com/) and KEGG (http://www.genome.jp/kegg/kegg2.html) were analyzed with the MassFrontier 6.0 software (HighChem, Slovakia). And then retention time and MS/MS spectra of authentic standards were checked to confirm the identification results.

Results and discussion

Serum metabolic profiling of HCC and CLD patients

Sera from patients with HCC and CLD including CHB and CIR, and the healthy controls were profiled to investigate metabolic variations among liver diseases. In order to evaluate the injury extent of liver mitochondria, the ratio of AST to ALT was calculated. It was found that the level of liver mitochondria injury increased with the liver disease severity from 0.8 ± 0.08 in CHB patients, 1.42 ± 0.08 in CIR to 1.63 ± 0.18 in HCC patients.

To simplify sample pretreatment procedure and reduce metabolic information loss during the refrigerated centrifuge and reconstitution, serum supernatant was injected directly without condensation after protein precipitation. No significant solvent effect was found as the inject volume was 4 μl (5 μl in the negative mode). A typical extracted ion chromatogram of the compounds found by Molecular Feature Extraction algorithm of Qualitative analysis software (Agilent, USA) is shown in Fig. 1. After aligning peaks, there were 1,622 ions with the peak area above 2,000. The peak area limit was set to 2,000 to get rid of noise interference. And 391 variables were obtained after further removing the missing values.
Fig. 1

A typical extracted chromatogram of the compounds found by the Molecular Feature Extraction software. Most of the identified compounds (in Table 2) are marked by corresponding colored compound ID

The system stability was monitored by the QC sample, which was run before and during the sequence. After removing missing values and performing unit variance scaling for the variables detected at the positive mode, PCA was performed. It is found that the QC samples cluster together tightly in the score plot of PCA with the eclipse indicating 95% confidence interval when all the real samples and QC samples were projected to the plane of principal components 1 and 2 (Fig. 2). This ensures the reliability of the metabolic profiling data [29].
Fig. 2

PCA score plot of all the real samples and QC injections performed on the unit variance scaled data. QC injections (black triangles); real samples/healthy control (white squares), CHB (plus sign), CIR (white circles), and HCC (asterisks). The eclipse indicates a confidence interval of 95%. PC1 and PC2 mean principal components 1 and 2, respectively

Significantly changed metabolites among liver diseases defined by RF-RFE

In order to explore more valuable information from the massive metabolic data, RF-RFE was applied. Since RF is not sensitive to data standardization and data normalization [33], the data pretreatment is not necessary except for the removal of the missing values with the 80% rule [30].

In RF, the size of the candidate splitting feature subset was set as its default value sqrt (N), where N was the number of the features; the tree number of each random forest was set as 100. Each RF-RFEi (i = 1, 2, 3, 4, and 5) was run 50 times; the average accuracy rates of RF-RFEi (i = 1, 2, 3, 4, and 5) on the independent test sets were 96.55 ± 3.50%, 66.07 ± 7.64%, 71.60 ± 9.34%, 84.40 ± 6.36%, and 78.40 ± 8.89%, respectively. This implies that RF-RFE can measure the features properly and produce a good model performance.

In RF-RFEi (i = 1, 2, 3, 4, and 5), each run could generate a feature ranking sequence. The top-ranked features were the most informative. The occurrence frequency of each variable in the top 50 ranked features of each run was calculated. The variables with the frequency bigger than 0.5 were selected. The final feature subset including 96 feature ions was the union of all selected features from RF-RFEi (i = 1, 2, 3, 4, and 5).

Important differential expression variables chosen by above RF-RFE were tested through the Mann–Whitney U test. Variables with p < 0.05 were kept as the feature ions. Nintety-one variables were finally defined. Their identification was performed based on our previous method [32] with the small modification, 18 identified feature metabolites were listed in Table 2 including tryptophan, cortisol, glycocholic acid (GCA), chenodeoxycholic acid glycine conjugate (GCDCA), some lysophosphatidylcholines and lysophosphatidylethanolamines, and several acylcarnitines.
Table 2

Identified differential expression compounds compared with healthy controls (N) selected by RF-RFE

Compound ID

Compounds identified

tR(min)

Mass

Mass error (ppm)

HCC/N

CIR/N

CHB/N

Mann–Whitney U test (p)

HCC vs. CIR

HCC vs. CHB

CIR vs. CHB

1

Tryptophan

3.98

158.09

8.25

0.28**

0.33**

0.32*

0.91

0.05

0.03

2

Cortisol

6.78

362.21

2.71

1.6*

1.07

1.84**

0.04

0.09

0.00

3

GCA

8.92

411.28

2.49

10.27**

18.36**

3.13**

0.91

0.02

0.05

4

GCDCA

11.24

431.3

1.26

12.96**

13.09**

4.79**

0.48

0.02

0.00

5

LysoPC (0:0/14:0)

13.09

467.3

2.19

0.62**

0.83*

1.36

0.36

0.00

0.01

6

LysoPC (14:0/0:0)

13.62

467.3

1.33

0.72**

0.85*

1.35

0.54

0.00

0.01

7

LysoPC (20:3)

16.18

545.35

0.47

0.6*

0.71

1.16

0.51

0.00

0.00

8

LysoPC (20:5)

14.3

541.32

0.68

0.64*

0.54**

1.63

0.72

0.00

0.00

9

LysoPC (22:6)

15.65

567.33

0.21

0.81

0.71*

1.29

0.58

0.01

0.00

10

LysoPC (15:0)

14.99

481.32

1.18

0.88

0.87

1.35*

0.66

0.00

0.00

11

LysoPE (20:4)

15.12

501.29

3.88

0.76

0.69*

0.99

0.36

0.03

0.00

12

LysoPE (22:6)

15.17

525.29

1.29

0.92

0.78

1.7**

0.79

0.00

0.00

13

LysoPE (22:6)

15.55

525.29

0.91

0.91

0.74

1.56**

0.46

0.00

0.00

14

C16:1-CN

14.72

397.32

2.73

2.72**

2.76**

1.48*

0.27

0.02

0.00

15

C10-CN

8.88

315.24

5.18

0.32**

0.24**

0.37**

0.34

0.76

0.16

16

C10:1-CN

8.23

313.23

5.07

0.43**

0.3**

0.46**

0.05

0.76

0.01

17

C8-CN

6.52

287.21

6.07

0.26**

0.2**

0.27**

0.35

0.91

0.38

18

C6-CN

4.96

259.18

6.77

0.32**

0.24**

0.22**

0.28

0.21

0.73

GCA glycocholic acid, GCDCA chenodeoxycholic acid glycine conjugate, CN acylcarnitine, LysoPC lysophosphatidylcholine, LysoPE lysophosphatidylethanolamine

*p < 0.05, significant difference level; **p < 0.01, significant difference level between corresponding two groups

Systemic metabolic deregulation in the HCC and CLD patients

The related metabolic changes may help to interpret the biological events in the different liver diseases. From Table 2, we can see a decrease in tryptophan, several acylcarnitines (CN) with medium and short fatty acid chain including C10-acylcarnitine (C10-CN), C10:1-CN, C8-CN, C6-CN, and some lysophospholipid molecules and an increasing level of cortisol, GCA, GCDCA, C16:1-CN, and other lysophospholipid molecules in the liver diseases.

It was found that tryptophan decreased in the sera of CHB, CIR and HCC patients. Tryptophan can be metabolized by rate-limiting enzyme IDO [34, 35] induced by interferons [36] or other inflammatory cytokines after the system immune activation to resist viral infection [37, 38] or tumor[39].

Cortisol increased in both CHB and HCC groups compared with the healthy control. Most of CHB patients in this study were with active hepatitis virus B replication. It was reported that serum cortisol significantly increased in patients with the chronic active hepatitis B while increased but not statistically significant in cases with HBeAg (+) and chronic persistent hepatitis B [40]. And elevated cortisol was also observed in plasma of 15 Southern African black patients with HCC in comparison to ten control subjects which was significantly correlated with alpha (2)-macroglobulin elastase binding capacity [41].

The formation of acylcarnitines is essential to the mitochondria shuttle of long-chain fatty acid for β-oxidation. And acylcarnitines with different acyl residues are usually used to screen or diagnose inborn fatty acid oxidation (FAO) errors [42]. And also, to examine FAO alteration in obesity and type 2 diabetes (T2DM), acylcarnitines in human plasma from individuals with obesity and T2DM during fasting and insulin-stimulated conditions have been profiled. Higher level of long-chain acylcarnitines has been found both in obesity and T2DM with accumulated shorter-chain species in T2DM as well [43]. In comparison, in this study of CLD and HCC cases, C16:1-CN as one of the long-chain acylcarnitines, increased with severity of chronic liver diseases, while acylcarnitines with medium and short-chain including C10-CN, C10:1-CN, C8-CN, and C6-CN decreased. So, it is important to further investigate serum free fatty acid level and the special alteration of FAO in different liver diseases.

Fatty acid profiling of HCC and CLD in the negative mode

It is known that long-chain acylcarnitines play a decisive role in free fatty acid (FFA) oxidation, responsible for the transportation of acyl-coenzyme A (CoA) into mitochondria. Thus, a further investigation was also performed for FFA at the same operation conditions as above but detected in the negative ion mode. The QC injections were used to monitor the consistency of instrumental performance, the peak area RSD of FFAs in QC injections was calculated: four of ten detected FFAs had a peak area RSD smaller than 15%, and eight of ten smaller than 20%, only FFA (C18:0) and FFA (22:5) had a peak area RSD of 29% and 20%, respectively. Table 3 shows the total increasing trend of FFA relative to the healthy with liver disease development. All of the FFAs accumulated progressively from controls to CHB, then to CIR and HCC, with scarce difference between CIR and HCC except FFA (C16:1). FFA (C16:1) decreased a lot in HCC compared with CIR, though not significantly.
Table 3

Alteration trends of fatty acids in CHB, CIR, and HCC, compared with healthy controls (N)

FFA

tR (min)

Mass

HCC/N

CIR/N

CHB/N

Mann–Whitney U test (p))

HCC vs. CIR

HCC vs. CHB

CIR vs. CHB

FFA (C16:1)

21.89

254.23

5.9**

8.4**

2.43*

0.26

0.00

0.00

FFA (C16:0)

24.04

256.24

2.49**

2.3**

1.40

0.87

0.01

0.01

FFA (C18:2)

22.84

280.24

1.79**

1.94**

1.31

0.52

0.06

0.02

FFA (C18:1)

24.53

282.26

3.16**

3.13**

1.56

0.72

0.00

0.00

FFA (C18:0)

25.58

284.27

2.61**

1.74*

1.12

0.26

0.01

0.11

FFA (C20:5)

21.04

302.23

2.86**

2.75**

2.46*

0.99

0.16

0.16

FFA (C20:4)

22.66

304.24

2.44**

2.2**

1.44

0.66

0.01

0.02

FFA (C20:2)

24.92

308.27

3.1**

2.62**

1.41

0.32

0.00

0.01

FFA (C22:6)

22.38

328.24

2.61**

2.87**

2.20

0.65

0.19

0.14

FFA (C22:5)

24.37

330.26

2.27**

2.32**

1.19

0.88

0.01

0.01

FFA free fatty acid

*p < 0.05, significant difference level; **p < 0.01, significant difference level between corresponding two groups

Ratio of saturated FFAs to monounsaturated FFAs (Fig. 3) was calculated to monitor stearoyl-CoA desaturase (SCD) enzyme activity. The ratio decreased significantly in CHB, CIR, and HCC compared with that in the healthy control, and decreased significantly in CIR and HCC compared with that in CHB. There was a decreasing trend with severity of liver diseases, but the alteration extent was smaller in HCC than in CIR.
Fig. 3

Relative signals of SCD activity in CLD and HCC compared with the control, reversely reflected by ratio of saturated fatty acids to monounsaturated fatty acids. Asterisks mean the signal in corresponding liver disease group was significantly different from that in the control (p < 0.05); plus sign means signal in CIR or HCC was significantly different from that in CHB

Alterations of serum acylcarnitine level among HCC and CLD

To further analyze the metabolic profile of acylcarnitines, the raw metabolic profiling data were further extracted, and the peak area RSD of acylcarnitines in QC samples was calculated: among the 18 acylcarnitines, 13 acylcarnitines had a peak area RSD smaller than 15%, 17 acylcarnitines smaller than 20%, and only C18:0-CN had a peak area RSD of 21%. The obtained small peak area RSD of acylcarnitines in QC injections ensured the reliability of data. The trend of all the acylcarnitines and free carnitine detected can be seen in Fig. 4a, b. The results demonstrated that free carnitine and acylcarnitines with medium and short-chain decreased from controls to CHB, and then to CIR (except C2-CN and C3-CN), while acylcarnitines with long chain reversely accumulated. The level of C14-CN also increased in CIR and HCC compared with that in the healthy control. And C14:1-CN, the first intermediate of C16-CN, was a key turning point. Smaller changing extent of acylcarnitines was observed in HCC than that in CIR, except C18-CN.
Fig. 4

Contents of acylcarnitines and free carnitine in CLD and HCC. A Acylcarnitines with long chain; B free carnitine and acylcarnitines with a medium or short chain; Asterisks means peak area in corresponding liver disease group was significantly different from that in the control (p < 0.05); plus sign means peak area in CIR or HCC was significantly different from that in CHB; number sign means peak area in HCC was significantly different from that in CIR

Ratio among acylcarnitines or free carnitine was calculated which can reflect related metabolic activities. From Fig. 5a–c, with severity of liver diseases, we can see a similar increasing trend of (C16 + C18)-CN/carnitine, (C16 + C18:1)-CN/C2-CN, C16-CN/C8-CN and C2-CN/carnitine from healthy control to disease group, having smaller alteration extent in HCC than that in CIR. And compared with the healthy control, (C16 + C18)-CN/carnitine, (C16 + C18:1)-CN/C2-CN, and C16-CN/C8-CN were significantly higher in CHB, CIR and HCC. Furthermore, (C16 + C18)-CN/carnitine and C16-CN/C8-CN and C2-CN/carnitine were significantly increased in CIR relative to CHB, C2-CN/carnitine also significantly increased in HCC relative to CHB. And the (C16 + C18)-CN/carnitine was significantly higher in HCC in comparison to CIR.
Fig. 5

Relative signals of corresponding enzyme activities in CLD and HCC compared with those in the control. A Ratio of sum of C16-CN and C18-CN to carnitine, reflecting carnitine palmitoyl transferase 1 (CPT 1) activity; B ratio of sum of C16-CN and C18:1-CN to C2-CN, reversely reflecting carnitine palmitoyl transferase 2 (CPT 2) activity; C ratio of C16-CN to C8-CN, reflecting long-chain acyl-CoA dehydrogenase activity (solid line), ratio of C2-CN to free carnitine as an indicator for beta-oxidation of even-numbered fatty acids (dashed line). Asterisks mean the signals in corresponding liver disease group were significantly different from those in the control (p < 0.05); plus sign means the signals in CIR or HCC significantly different from those in CHB; number sign means the signals in HCC were significantly different from those in CIR

Deregulation of fatty acid related metabolism in HCC and CLD

The significantly altered fatty acid related metabolic pathways in HCC and CLD are illustrated in Fig. 6. The reprogramming of energy metabolism has been considered as an important hallmark of tumors [44]. Cancer cell takes glycolysis as a major way for energy metabolism. In the mean time, this over consumption of glucose would also increase the lipolysis so that FFA could provide energy via FAO. Actually, no significant difference of FFAs has been found between HCC and CIR cases. This may partly attribute to the accompaniment of cirrhosis lesion in the HCC individuals. However, the changing trends of the FFAs could also be informative for the lipid deregulation. The increase of saturated FFAs may imply a rising level of lipolysis and oxidative stress. Of note, rising FFA (C16:1) is able to increase insulin sensitivity [45].
Fig. 6

Significantly altered fatty acid-related metabolic pathways in HCC and CLD. Free fatty acids are produced through lipolysis and then transferred to liver for oxidation or lipid synthesis. For beta-oxidation, fatty acids are firstly transferred into acyl-S-CoAs in cytoplasm. The thioesters have to be catalyzed by outer membrane CPT 1 to generate acyl-carnitines, which are transported into the mitochondria by a specific inner membrane CACT in exchange for internal free carnitines. The inner membrane CPT 2 catalyzes inner acyl-carnitines to generate acyl-S-CoAs, which are involved in the following mitochondria beta-oxidation. FFA free fatty acid, SCD1 stearoyl-CoA desaturase, Acy-CN acylcarnitine, CPT 1 carnitine palmitoyl transferase 1, CACT carnitine-acylcarnitine translocase, responsible for transporting acylcarnitines from cytoplasm into mitochondrial matrix, CPT 2 carnitine palmitoyl transferase 2

SCD, with its main isoform SCD1 expressed in liver [46], was activated in CHB, CIR and HCC, characterized by reversely decreasing ratio of saturated FFAs to monounsaturated FFAs (Fig. 3). The benefit of SCD1 deficiency to increase FAO and inhibit lipid synthesis has been reviewed [46], concluding that increased SCD activity in humans and animals is related to significantly accumulated lipids in liver [46]. And the significantly decreased ratio of saturated FFAs to monounsaturated FFAs in CIR and HCC compared with CHB indicated that the SCD activity increased with the severity of liver diseases.

The combustion of FFAs in different liver diseases can be evaluated on the levels of acylcarnitines. Long-chain acylcarnitines which are responsible for shuttling into mitochondria of long-chain fatty acids, increased in liver diseases compared with the controls (Fig. 4a), suggesting that more fatty acids can enter mitochondria [43]. Moreover, there was a trend that the more severe the liver injury was, the more long-chain acylcarnitines accumulated in plasma.

With the decline of free carnitine and increase of long-chain acylcarnitines in liver disease state, FFAs were readily translocated onto carnitines by carnitine palmitoyl transferase 1 (CPT 1), reflected by higher level of (C16 + C18)-CN/carnitine (Fig. 5a). CPT 1 has been reported to be regulated by long-chain FFAs, independently from peroxisome proliferator-activated receptor alpha (PPARα) [47], a sensor for FFAs. However, the lower level of carnitine and higher level of long-chain acylcarnitnes indicated a deficiency of carnitine palmitoyl transferase 2 (CPT 2), demonstrated by reversely increased (C16 + C18:1)-CN/C2-CN (Fig. 5b). CPT 2 locates at the inner side of inner mitochondria membrane, responsible for fatty acid conjugated back to CoA releasing carnitine to mitochondrial matrix. Also, the increasing ratio of C16-CN to C8-CN (Fig. 5c) and decreasing medium and short-chain acylcarnitines (Fig. 4b) in liver diseases, indicated that once the fatty acids were conjugated back to CoA in the mitochondria matrix, they may be easily oxidized into acetyl-CoA. That is to say, FAO in liver diseases is mainly limited by CPT 2. Seelaender et al. have also found that CPT 2 activity shrank in rats bearing the Walker 256 carcinosarcoma and the expression of CPT 2 may be affected by a prostaglandin, PGE(2) [48].

Free carnitine decreased progressively in non-malignant liver diseases with a little rebound in HCC compared with CIR (Fig. 4b). It has been reported that carnitine strikingly decreases in PPARα (−/−) mice, due to both direct and indirect effects of PPARα inactivity. And there are reduced tissue levels of short-chain acylcarnitine intermediates and tissue accumulation of long-chain acyl-CoAs, as well [49]. So the expression level of PPARα in liver diseases needs to be investigated further. Free carnitine supplement has been proved to effectively prevent hepatitis and subsequent HCC in Long-Evans Cinnamon rats via protection of mitochondria from oxidative injury [50].

C2-CN is mainly from acetyl-CoA, the ultimate product of FAO. The significantly decreased plasma level of C2-CN in liver diseases (Fig. 4b) implies both the lack of long-chain acyl-CoA in mitochodria matrix for production of acetyl-CoA and the fast consumption of acetyl-CoA. As is mentioned above, cancer cell reprograms its energy metabolism by promoting glycolysis and suppressing the TCA cycle, which would inevitably causes the accumulation of carboxylic acid in the mitochondria. When these carboxylic acids are transported outside the mitochondria by carnitines, the relative accumulation of C2-CN and other medium and short-chain acylcarnitines would be detected in HCC compared with CIR (Fig. 4b). And these glycolytic intermediates may subsequently facilitate the proliferation of cancer cells [44]. Moreover, their accumulation may reversely inhibit the formation of long-chain acyl-CoA, resulting a relative smaller accumulation of long-chain acylcarnitines in HCC compared with CIR (Fig. 4a), and lower ratios of (C16 + C18)-CN to carnitine, and C16-CN to C8-CN as well (Fig. 5a, c). Moreover, C2-CN/carnitine, an indicator for beta-oxidation of even-numbered fatty acids, rose in CIR and HCC (Fig. 5c), resulting from more severe liver mitochondria injury with corresponding higher AST/ALT.

Conclusions

Through combining serum metabolic profiling with the RF-RFE algorithm for feature selection, important metabolic changes among liver diseases were found out from a large amount of metabolome data. These metabolic deregulations are related to pathways of tryptophan, glucocorticoid, bile acids and FFAs among the liver diseases. The deregulation of FFA metabolism is a primary focus subsequently for it is closely related with the reprogramming of energy metabolism in cancer cell. It was found that a lowered level of free carnitine and a deficiency of CPT 2 are correlated with liver disease severity. Besides, the alterations of FFA (16:1), SCD activity, acylcarnitines and some corresponding enzymes also imply the possible mechanisms related to the development of HCC. These results provide an important insight into metabolic deregulations contributing to the development of liver diseases and may further help to explore new target for preventing deterioration of liver diseases and differential diagnosis of HCC.

Notes

Acknowledgments

The study has been supported by the State Key Science and Technology Project for Infectious Diseases (2008ZX10002-017 and 2008ZX10002-019) from State Ministry of Science and Technology of China, and the key foundation (no. 20835006) and the creative research group project (no. 21021004) from National Natural Science Foundation of China.

References

  1. 1.
    Feitelson M (1992) Hepatitis-B virus-infection and primary hepatocellular-carcinoma. Clin Microbiol Rev 5:275–301Google Scholar
  2. 2.
    Villanueva A, Newell P, Chiang DY, Friedman SL, Llovet JM (2007) Genomics and signaling pathways in hepatocellular carcinoma. Semin Liver Dis 27:55–76CrossRefGoogle Scholar
  3. 3.
    Farazi PA, DePinho RA (2006) Hepatocellular carcinoma pathogenesis: from genes to environment. Nat Rev Cancer 6:674–687CrossRefGoogle Scholar
  4. 4.
    Feitelson MA, Duan LX (1997) Hepatitis B virus× antigen in the pathogenesis of chronic infections and the development of hepatocellular carcinoma. Am J Pathol 150:1141–1157Google Scholar
  5. 5.
    Kao JH, Chen PJ, Lai MY, Chen DS (2003) Basal core promoter mutations of hepatitis B virus increase the risk of hepatocellular carcinoma in hepatitis B carriers. Gastroenterology 124:327–334CrossRefGoogle Scholar
  6. 6.
    Thorgeirsson SS, Grisham JW (2002) Molecular pathogenesis of human hepatocellular carcinoma. Nat Genet 31:339–346CrossRefGoogle Scholar
  7. 7.
    Wang XW, Hussain SP, Huo TI, Wu CG, Forgues M, Hofseth LJ, Brechot C, Harris CC (2002) Molecular pathogenesis of human hepatocellular carcinoma. Toxicology 181:43–47CrossRefGoogle Scholar
  8. 8.
    Brechot C (2004) Pathogenesis of hepatitis B virus-related hepatocellular carcinoma: old and new paradigms. Gastroenterology 127:S56–S61CrossRefGoogle Scholar
  9. 9.
    Marchio A, Meddeb M, Pineau P, Danglot G, Tiollais P, Bernheim A, Dejean A (1997) Recurrent chromosomal abnormalities in hepatocellular carcinoma detected by comparative genomic hybridization. Genes Chromosomes Cancer 18:59–65CrossRefGoogle Scholar
  10. 10.
    Yamashita T, Honda M, Takatori H, Nishino R, Minato H, Takamura H, Ohta T, Kaneko S (2009) Activation of lipogenic pathway correlates with cell proliferation and poor prognosis in hepatocellular carcinoma. J Hepatol 50:100–110CrossRefGoogle Scholar
  11. 11.
    Nicholson JK, Lindon JC, Holmes E (1999) ‘Metabonomics’: understanding the metabolic responses of living systems to pathophysiological stimuli via multivariate statistical analysis of biological NMR spectroscopic data. Xenobiotica 29:1181–1189CrossRefGoogle Scholar
  12. 12.
    van der Greef J, Stroobant P, van der Heijden R (2004) The role of analytical sciences medical systems biology. Curr Opin Chem Biol 8:559–565CrossRefGoogle Scholar
  13. 13.
    Cobbold JFL, Patel JH, Goldin RD, North BV, Crossey MME, Fitzpatrick J, Wylezinska M, Thomas HC, Cox IJ, Taylor-Robinson SD (2010) Hepatic lipid profiling in chronic hepatitis C: an in vitro and in vivo proton magnetic resonance spectroscopy study. J Hepatol 52:16–24CrossRefGoogle Scholar
  14. 14.
    Shariff MIF, Gomaa AI, Cox IJ, Patel M, Williams HRT, Crossey MME, Thillainayagam AV, Thomas HC, Waked I, Khan SA, Taylor-Robinson SD (2011) Urinary metabolic biomarkers of hepatocellular carcinoma in an Egyptian population: a validation study. J Proteome Res 10:1828–1836CrossRefGoogle Scholar
  15. 15.
    Soga T, Sugimoto M, Honma M, Mori M, Igarashi K, Kashikura K, Ikeda S, Hirayama A, Yamamoto T, Yoshida H, Otsuka M, Tsuji S, Yatomi Y, Sakuragawa T, Watanabe H, Nihei K, Saito T, Kawata S, Suzuki H, Tomita M, Suematsu M (2011) Serum metabolomics reveals γ-glutamyl dipeptides as biomarkers for discrimination among different forms of liver disease. J Hepatol. doi:10.1016/j.jhep.2011.01.031
  16. 16.
    Gao HC, Lu Q, Liu X, Cong H, Zhao LC, Wang HM, Lin DH (2009) Application of H-1 NMR-based metabonomics in the study of metabolic profiling of human hepatocellular carcinoma and liver cirrhosis. Cancer Sci 100:782–785CrossRefGoogle Scholar
  17. 17.
    Yin PY, Wan DF, Zhao CX, Chen J, Zhao XJ, Wang WZ, Lu X, Yang SL, Gu JR, Xu GW (2009) A metabonomic study of hepatitis B-induced liver cirrhosis and hepatocellular carcinoma by using RP-LC and HILIC coupled with mass spectrometry. Mol Biosyst 5:868–876CrossRefGoogle Scholar
  18. 18.
    Handrick V, Vogt T, Frolov A (2010) Profiling of hydroxycinnamic acid amides in Arabidopsis thaliana pollen by tandem mass spectrometry. Anal Bioanal Chem 398:2789–2801CrossRefGoogle Scholar
  19. 19.
    Yin P, Zhao X, Li Q, Wang J, Li J, Xu G (2006) Metabonomics study of intestinal fistulas based on ultraperformance liquid chromatography coupled with Q-TOF mass spectrometry (UPLC/Q-TOF MS). J Proteome Res 5:2135–2143CrossRefGoogle Scholar
  20. 20.
    Zhang X, Wei D, Yap Y, Li L, Guo S, Chen F (2007) Mass spectrometry-based “omics” technologies in cancer diagnostics. Mass Spectrom Rev 26:403–431CrossRefGoogle Scholar
  21. 21.
    Dunn WB, Bailey NJC, Johnson HE (2005) Measuring the metabolome: current analytical technologies. Analyst 130:606–625CrossRefGoogle Scholar
  22. 22.
    Breiman L (2001) Random forests. Mach Learn 45:5–32CrossRefGoogle Scholar
  23. 23.
    Xu P, Jelinek F (2007) Random forests and the data sparseness problem in language modeling. Comput Speech Lang 21:105–152CrossRefGoogle Scholar
  24. 24.
    Pang H, Lin A, Holford M, Enerson BE, Lu B, Lawton MP, Floyd E, Zhao H (2006) Pathway analysis using random forests classification and regression. Bioinformatics 22:2028–2036CrossRefGoogle Scholar
  25. 25.
    Truong Y, Lin X, Beecher C (2004) Learning a complex metabolomic dataset using random forests and support vector machines. In: Proceedings of the tenth ACM SIGKDD international conference on knowledge discovery and data mining, Seattle, WA, USAGoogle Scholar
  26. 26.
    Menze BH, Kelm BM, Masuch R, Himmelreich U, Bachert P, Petrich W, Hamprecht FA (2009) A comparison of random forest and its Gini importance with standard chemometric methods for the feature selection and classification of spectral data. BMC Bioinformatics 10:213–228CrossRefGoogle Scholar
  27. 27.
    Strobl C, Boulesteix A-L, Zeileis A, Hothorn T (2007) Bias in random forest variable importance measures: illustrations, sources and a solution. BMC Bioinformatics 8:25–45CrossRefGoogle Scholar
  28. 28.
    Granitto P, Furlanello C, Biasioli F, Gasperi F (2006) Recursive feature elimination with random forest for PTR-MS analysis of agroindustrial products. Chemom Intell Lab Syst 83:83–90CrossRefGoogle Scholar
  29. 29.
    Gika HG, Theodoridis GA, Wingate JE, Wilson ID (2007) Within-day reproducibility of an HPLC-MS-based method for metabonomic analysis: application to human urine. J Proteome Res 6:3291–3303CrossRefGoogle Scholar
  30. 30.
    Smilde AK, van der Werf MJ, Bijlsma S, van der Werff-van-der Vat BJC, Jellema RH (2005) Fusion of mass spectrometry-based metabolomics data. Anal Chem 77:6729–6736CrossRefGoogle Scholar
  31. 31.
    Chandra B, Kothari R, Paul P (2010) A new node splitting measure for decision tree construction. Pattern Recogn 43:2725–2731CrossRefGoogle Scholar
  32. 32.
    Chen J, Zhao X, Fritsche J, Yin P, Schmitt-Kopplin P, Wang W, Lu X, Haring HU, Schleicher ED, Lehmann R, Xu G (2008) Practical approach for the identification and isomer elucidation of biomarkers detected in a metabonomic study for the discovery of individuals at risk for diabetes by integrating the chromatographic and mass spectrometric information. Anal Chem 80:1280–1289CrossRefGoogle Scholar
  33. 33.
    Bordner AJ, Gorin AA (2008) Comprehensive inventory of protein complexes in the Protein Data Bank from consistent classification of interfaces. BMC Bioinformatics 9:234–244CrossRefGoogle Scholar
  34. 34.
    Moroni F (1999) Tryptophan metabolism and brain function: focus on kynurenine and other indole metabolites. Eur J Pharmacol 375:87–100CrossRefGoogle Scholar
  35. 35.
    Wolf H (1974) Studies on tryptophan metabolism in man—effect of hormones and vitamin-B6 on urinary-excretion of metabolites of kynurenine pathway. Scand J Clin Lab Invest 33:1664–1667Google Scholar
  36. 36.
    Taylor MW, Feng GS (1991) Relationship between interferon-gamma, indoleamine 2,3-dioxygenase, and tryptophan catabolism. FASEB J 5:2516–2522Google Scholar
  37. 37.
    Yoshida R, Urade Y, Tokuda M, Hayaishi O (1979) Induction of indoleamine 2,3-dioxygenase in mouse lung during virus-infection. Proc Natl Acad Sci USA 76:4084–4086CrossRefGoogle Scholar
  38. 38.
    Heyes MP, Saito K, Lackner A, Wiley CA, Achim CL, Markey SP (1998) Sources of the neurotoxin quinolinic acid in the brain of HIV-1-infected patients and retrovirus-infected macaques. FASEB J 12:881–896Google Scholar
  39. 39.
    Takikawa O, Kuroiwa T, Yamazaki F, Kido R (1988) Mechanism of interferon-gamma action-characterization of indoleamine 2,3-dioxygenase in cultured human-cells induced by interferon-gamma and evaluation of the enzyme-mediated tryptophan degradation in its anticellular activity. J Biol Chem 263:2041–2048Google Scholar
  40. 40.
    Eyigun CP, Dayan S, Sengul A, Ozguven V, Alga OH, Hacibetkasoglu A (1995) Sera cortisol levels in patients with chronic hepatitis B virus infection and its prognostic significance. Mikrobiyol Bul 29:388–396Google Scholar
  41. 41.
    Gaillard MC, Song E, Nogueira CM, Kilroesmith TA (1995) Elastase binding-capacity of alpha(2)-macroglobulin and its association with glucocorticoid concentration in southern African black patients with hepatocellular-carcinoma. Clin Chim Acta 240:179–185CrossRefGoogle Scholar
  42. 42.
    Longo N, Filippo CAD, Pasquali M (2006) Disorders of carnitine transport and the carnitine cycle. Am J Med Genet C Semin Med Genet 142C:77–85CrossRefGoogle Scholar
  43. 43.
    Mihalik SJ, Goodpaster BH, Kelley DE, Chace DH, Vockley J, Toledo FGS, DeLany JP (2010) Increased levels of plasma acylcarnitines in obesity and type 2 diabetes and identification of a marker of glucolipotoxicity. Obesity 18:1695–1700CrossRefGoogle Scholar
  44. 44.
    Hanahan D, Weinberg RA (2011) Hallmarks of cancer: the next generation. Cell 144:646–674CrossRefGoogle Scholar
  45. 45.
    Cao HM, Gerhold K, Mayers JR, Wiest MM, Watkins SM, Hotamisligil GS (2008) Identification of a lipokine, a lipid hormone linking adipose tissue to systemic metabolism. Cell 134:933–944CrossRefGoogle Scholar
  46. 46.
    Dobrzyn P, Dobrzyn A (2006) Stearoyl-CoA desaturase: a new therapeutic target of liver steatosis. Drug Dev Res 67:643–650CrossRefGoogle Scholar
  47. 47.
    Louet JF, Chatelain F, Decaux JF, Park EA, Kohl C, Pineau T, Girard J, Pegorier JP (2001) Long-chain fatty acids regulate liver carnitine palmitoyltransferase I gene (L-CPT I) expression through a peroxisome-proliferator-activated receptor alpha (PPAR alpha)-independent pathway. Biochem J 354:189–197CrossRefGoogle Scholar
  48. 48.
    Seelaender MCL, Curi R, Colquhoun A, Williams JF, Zammitt VA (1998) Carnitine palmitoyltransferase II activity is decreased in liver mitochondria of cachectic rats bearing the Walker 256 carcinosarcoma: effect of indomethacin treatment. Biochem Mol Biol Int 44:185–193Google Scholar
  49. 49.
    Makowski L, Noland RC, Koves TR, Xing WB, Ilkayeva OR, Muehlbauer MJ, Stevens RD, Muoio DM (2009) Metabolic profiling of PPAR alpha(−/−) mice reveals defects in carnitine and amino acid homeostasis that are partially reversed by oral carnitine supplementation. FASEB J 23:586–604CrossRefGoogle Scholar
  50. 50.
    Chang BJ, Nishikawa M, Nishiguchi S, Inoue M (2005) l-carnitine inhibits hepatocarcinogenesis via protection of mitochondria. Int J Cancer 113:719–729CrossRefGoogle Scholar

Copyright information

© Springer-Verlag 2012

Authors and Affiliations

  • Lina Zhou
    • 1
  • Quancai Wang
    • 2
  • Peiyuan Yin
    • 1
  • Wenbin Xing
    • 3
  • Zeming Wu
    • 1
  • Shili Chen
    • 1
  • Xin Lu
    • 1
  • Yong Zhang
    • 3
  • Xiaohui Lin
    • 2
  • Guowang Xu
    • 1
  1. 1.CAS Key Laboratory of Separation Science for Analytical ChemistryDalian Institute of Chemical Physics, Chinese Academy of SciencesDalianChina
  2. 2.School of Computer Science and TechnologyDalian University of TechnologyDalianChina
  3. 3.The Sixth People’s HospitalDalianChina

Personalised recommendations