Metabolomics DOI 10.1007/s11306-013-0596-8

ORIGINAL ARTICLE

Serum metabolite profiling of B-cell non-Hodgkin’s lymphoma using UPLC-QTOFMS and GC-TOFMS Junyi Zhou • Shijun Yu • Yan Wang • Xue Gu • Qian Wu • Yun Xue • Gao Shan Huiping Zhang • Weili Zhao • Chao Yan

•

Received: 15 June 2013 / Accepted: 15 October 2013 Ó Springer Science+Business Media New York 2013

Abstract A global metabolic profiling was generated with serum samples of patients with B-cell non-Hodgkin’s lymphoma (NHL) and healthy controls using two different analytical platforms for metabonomics, UPLC-QTOFMS and GCTOFMS, in conjunction with multivariate data analysis and ROC analysis. Significant difference in metabolic characteristics was observed between B-cell NHL and healthy control by OPLS-DA. A total of 37 differential metabolites for B-cell NHL were identified. Some significant changes in metabolites were detected, indicating that there were disturbances of key metabolic pathways, including bile acids, glycerophospholipids, fatty acids metabolism, steroid biosynthesis, glycolysis, as well as glycine, serine and threonine metabolism associated with B-cell NHL. A panel of metabolite markers composed of choline, arachidonic acid, LysoPC (17:0), PA (16:0/16:0) and

Junyi Zhou and Shijun Yu have contributed equally to this work.

Electronic supplementary material The online version of this article (doi:10.1007/s11306-013-0596-8) contains supplementary material, which is available to authorized users. J. Zhou  Y. Wang  X. Gu  Q. Wu  Y. Xue  H. Zhang  C. Yan (&) School of Pharmacy, Shanghai Jiao Tong University, 800 Dongchuan Road, Shanghai 200240, China e-mail: [email protected] S. Yu  W. Zhao (&) State Key Laboratory of Medical Genomics, Shanghai Institute of Hematology, Shanghai Rui-Jin Hospital, School of Medicine, Shanghai Jiao Tong University, 197 Rui Jin Er Road, Shanghai 200025, China e-mail: [email protected] G. Shan Zhejiang Institute of Microbiology, 9 Huanggushan Road, Hangzhou 310012, Zhejiang, China

coproporphyrin from UPLC-QTOFMS and another panel of markers composed of benzenebutanoic acid, b-hydroxypyruvic acid, D-2-hydroxyoctanoic acid, pyruvic acid and arachidonic acid derived from GC-TOFMS were selected. A ROC curve analysis of these markers resulted in an AUC of 0.968 and 1.00 for the UPLC-QTOFMS and GC-TOFMS analysis, respectively. These biochemical changes provide a novel molecular diagnostic approach which could be helpful to further understand the pathogenesis and identify the therapeutic target of B-cell NHL. Keywords Non-Hodgkin’s lymphoma  B-cell  Metabonomics  UPLC-QTOFMS  GC-TOFMS  Serum

1 Introduction Non-Hodgkin’s lymphoma (NHL) is a heterogeneous group of lymphoid malignancies. NHL can be divided into the type of precursor cells that the lymphoma is raised from, either B or T lymphocytes, or natural killer cells. B-cell NHL is the most common type of NHL. Incidence of NHL has been increasing over the past 20 years (Emily and Michael 2012), ranging among the ten most frequent cancers worldwide. Diagnostic imaging, such as contrastenhanced computed tomography (CT) and positron emission tomography (PET) have been applied for the early detection of NHL and the screening of high-risk patients. However, a high false-positive rate could not be ignored. For instance, the positive predictive value (PPV) of PET is substantially lower and more variable than its negative predictive value (NPV). In aggressive NHL, the PPV is probably in the range of 40–50 % when using visual

123

J. Zhou et al.

assessment as the interpretation of PET Scans (Juweid and Hoekstra 2011). Han et al. (2009) reported an even lower PPV of interim PET of 33 % in 40 NHL patients with a median follow-up of 24 months. To date, although pathological study remains the gold standard for NHL diagnosis, noninvasive methods and new biomarkers for the early diagnosis of NHL remain of great interest. As B-cell lymphomas are the most common form of NHL and can arise from any stage in the B-cell maturation process (Martinez-Climent et al. 2010), the development of new biomarkers with high sensitivity and specificity in a less invasive manner is needed to improve early detection of NHL. Metabonomics is an effective and noninvasive diagnostic method that was defined as the comprehensive qualitative and quantitative analysis of all metabolites in cells, tissues, or biofluids following a genetic modification or physiological stimulus (Nicholson et al. 1999). Metabonomic-based diagnostics explores metabolites in a biological system and its response to a stress situation such as disease. Tumor cells exhibit a number of significant metabolic perturbations such as the Warburg effect that are already exploited in diagnosis and therapy and that could provide a rich source of predictive biomarkers. This study suggests that the metabonomic method may be a valuable and feasible tool to explore the disturbance in specific metabolism and biosynthesis associated with B-cell NHL. So far, little is known about how the B-cell NHL impact changes in metabolic profiles of the patients. There are very few studies dealing with prognostication or response to therapy of NHL. In search for alternative methods with improved detection rates, and/or better compliance rates in screening for NHL, a metabonomic approach with broad unbiased search for changes in the metabolic profile is a possible solution. Interesting results have been published by Yoo et al. (2010) by use of MALDI-TOF analysis, investigating the metabonomic profile of urine samples from patients with NHL. The aim of this study was a non-targeted mass spectrometry-based investigation of the metabolic effects in B-cell NHL patients, with the aim of identifying a larger proportion of the whole metabolome. Here, we report a serum metabonomic study on a cohort of 54 B-cell NHL patients and 50 healthy controls using ultra performance liquid chromatography-quadrupole time-of-flight mass spectrometry (UPLC-QTOFMS) and a cohort of 22 B-cell NHL patients and 21 healthy controls using gas chromatography-time-of-flight mass spectrometry (GC-TOFMS). It was intended to gain knowledge of important metabolic variations associated with B-cell NHL, which can be utilized for improved NHL detection, diagnosis, and therapeutic strategies.

123

2 Materials and methods 2.1 Chemicals Methanol (MeOH) was of HPLC grade and obtained from Tedia (Fairfield, OH, USA); acetonitrile (ACN) and formic acid were of gradient grade and obtained from Merck (Darmstadt, Germany), and distilled water was produced by the Milli-Q Reagent Water System (Millipore, MA, USA). The internal standard L-2-chlorophenylalanine was obtained from Shanghai Intechem Technology Co. Ltd. Pyridine was analytical grade and purchased from China National Pharmaceutical Group Corporation (Shanghai, China). BSTFA (1 % TMCS), heptadecanoic acid, methoxyamine, leucine-enkephalin and the reference standards of arachidonic acid, choline, chenodeoxycholic acid, glycocholic acid and cholic acid were supplied by Sigma Corporation (St. Louis, MO, USA). Lysophosphatidylcholines (LysoPC (14:0), LysoPC (P-16:0), LysoPC (16:0), LysoPC (17:0), LysoPC (18:1(11Z)), LysoPC (18:0), LysoPC (18:0)) and cholesterol were purchased from Avanti Polar Lipids (AL, USA). 2.2 Sample collection Serum samples for UPLC-QTOFMS analysis were obtained from 54 B-cell NHL patients and 50 healthy volunteers. The cohort for GC-TOFMS analysis is a subgroup of that for UPLC-QTOFMS, obtained from 22 B-cell NHL patients and 21 healthy volunteers. Samples were collected with an informed consent and approval from the local ethics committee. Patients enrolled in this research were not on any medication before sample collection. The clinical diagnosis and pathological reports of all the patients were obtained from Shanghai Rui-Jin Hospital. The healthy volunteers were selected by a routine physical examination in the same hospital and any subjects with inflammatory conditions or hematological disorders were excluded. The venous blood samples were collected in the morning from each subject before breakfast. The samples were drawn into the separation gel coagulation-promoting tubes and then after 30 min, centrifuged at 3,000 rpm for 20 min at 4 °C to fractionate the serum for biochemical parameter analysis and metabolic profiling experiments. Detailed clinical information of serum samples is provided in Table 1. All samples were stored at -80 °C until use. There was no significant difference for the age and sex between B-cell NHL and healthy controls. The P value of Student’s t test for age between the two groups was 0.90 for UPLC result and 0.91 for GC result. The values of Chi square with Yate’s correction were both larger than 0.84 for sex between the two groups in UPLC and GC result.

Serum metabolic profile of B-cell NHL Table 1 Demographics and characteristic of B-cell NHL patients and healthy controls

Characteristic

Patients for UPLCQTOFMS

Healthy controls for UPLCQTOFMS

Patients for GCTOFMS

Healthy controls for GC-TOFMS

Number of all patients/controls

54

50

22

21

Age, years (median, range)

56, 26–76

53, 28–77

56, 26–76

53, 27–77

Male

34

26

12

11

Female

20

24

10

10

Gender

Histologic type Diffuse large B cell lymphoma Follicular lymphoma

38 4

12 4

Mantle cell lymphoma

3

1

Small B-cell lymphoma

4

2

MALT lymphoma

2

1

Splenic marginal zone lymphoma

2

1

Burkitt lymphoma

1

1

ECOG score 0 or 1

37

12

C2

17

10

Yes

27

13

No

27

9

34

12

20

10

Low

26

9

Low-intermediate

11

5

High-intermediate

14

6

High

3

2

Extranodal involvement

Ann Arbor Stage I–II III–IV International Prognostic Index

2.3 Sample preparation and analysis by UPLC-QTOFMS Serum samples were thawed at room temperature before analysis. 50 lL L-2-chlorophenylalanine in water (0.2 mg/ mL) as internal standard was added to each 100 lL serum sample in a 1.5 mL conical plastic test tube and vortexed for 30 s. The sample was then added into 350 lL mixture of organic solvents (MeOH–ACN 2:1; v/v, according to our previous study and others’ publication (Qiu et al. 2009)) and vortexed for 2 min to precipitate the proteins and extract the analytes. The mixture was kept at 4 °C for 15 min and then centrifuged at 15,000 rpm for 15 min at about 4 °C. 200 lL of supernatant was centrifuged again at 15,000 rpm for 15 min, transferred into a new tube and stored at 4 °C during UPLC-QTOFMS analysis. An inhouse quality control (QC) was prepared by pooling and mixing the same volume of each sample from healthy people according to our previous work (Zhang et al. 2011). The QC sample was injected frequently (every ten

samples) throughout the analytical run and monitored for changes in spectrometer response. An ACQUITYTM UPLC system (Waters Corp., Milford, MA, USA) was used. Reversed-phase separation was performed on a 2.1 9 100 mm ACQUITYTM 1.7 lm BEH C18 column (Waters Corp., Milford, MA, USA). The column temperature was kept constant at 40 °C. The mobile phase contained (A) water with 0.1 % v/v formic acid and (B) acetonitrile with 0.1 % v/v formic acid. The gradient duration was 18 min at a flow rate of 0.4 mL/min without split and the detailed gradient conditions were described as follows: 0–3 min, 1–20 % B; 3–5 min, 20–60 % B; 5–12 min, 60–99 % B; 12–14.5 min, 99 % B; 14.5–15.5 min, 99–5 % B; 15.5–16.2 min, 5–1 % B; 16.2–18 min, 1 % B. A 5 lL aliquot of each sample was injected onto the column. Mass spectrometry was performed on a Waters Q-TOF Micromass (Waters Corp., Milford, MA, USA) operating both in positive and negative ion mode at the resolution of 8000. Nitrogen was used as desolvation and cone gas. The

123

J. Zhou et al.

desolvation gas was set to 700 L/h at a temperature of 350 °C, and the cone gas was set to 50 L/h. Source temperature was set to 100 °C. The capillary voltage was 3.0 kV in positive ion mode and 2.4 kV in negative ion mode with cone voltage of 35 V in positive ion mode and 55 V in negative ion mode. Argon was employed as collision gas and the collision energy was set at 6.0 eV. The data acquisition rate was set to 0.3 s scan time with a 0.02 s interscan time. Data between m/z 50 and 1000 were recorded in centroid mode. Lock Mass calibration was applied using a solution of leucine enkephaline (0.5 mg/L, m/z 556.2707 for positive and 554.2561 for negative ion mode) at 0.1 mL/min. The deviations of m/z for the selected ions were always less than 0.005 kDa. A ‘‘control-patient-’’ run order was chosen to avoid systematic bias resulting from instrumental drift. The same run order was used for both ionization modes. QCs were injected repeatedly until stability of peak retention and signal intensity was obtained prior to the analysis of the experimental samples. 2.4 UPLC-QTOFMS data analysis The UPLC-QTOFMS ESI?/- raw data were analyzed by the MarkerLynx Applications Manager version 4.1 (Waters Corp., Milford, MA, USA). The parameters used were retention time (RT) range 0.5–13 min, mass range 50–1,000 kDa, mass tolerance 0.02 kDa, isotopic peaks were excluded for analysis, noise elimination level was set at 6.00, minimum intensity was set to 15 % of base peak intensity, maximum masses per RT was set at 6 and, finally, RT tolerance was set at 0.1 min. A list of the ion intensities of each peak detected was generated, using RT and the m/z data pairs as the identifier for each ion. To obtain consistent differential variables, the resulting matrix was further reduced by removing any peaks with missing value (ion intensity = 0) in more than 60 % samples (Cheng et al. 2012). The internal standard was also used for data quality control (reproducibility). Any known pseudo positive peaks, such as peaks caused by noise, column bleed and solvents, were removed from the data set. The detectable spectral features were then normalized to the internal standard within that sample to correct for the MS response shift from the first injection to the last injection due to the long duration of UPLC–MS analysis. QC samples were used for within batch variation correction (van der Kloet et al. 2009). As metabolite distributions did not always achieve normality, Box-Cox power transformations were implemented for each metabolite. The normalized and transformed data were then statistically analyzed. To reflect the differences between B-cell NHL patients and healthy individuals, multivariate statistical analysis was performed using SIMCA-P software version 12.0

123

(Umetrics AB, Umea˚, Sweden). The unsupervised principle component analysis (PCA) and supervised orthogonal partial least squares—discriminant analysis (OPLS-DA) models were constructed with the data. Cross validation (seven times) was used to calculate the number of significant components. Pareto (Par) scaling was used in the supervised analysis models. To prevent model overfitting, the supervised models (PLS-DA) were validated with a permutation test that was repeated 200 times. Plots for variable importance for projection (VIP) were created in order to detect features differing from patients and controls. Potential candidates for discrimination were selected on the basis of the threshold VIP [1 which is frequently used in metabolomics data analysis (Qiu et al. 2010; Ali et al. 2012). In addition to the multivariate statistical method, the Student’s t test and Wilcoxon-Mann–Whitney test were also applied to measure the significance of each metabolite. Metabolites with both multivariate and univariate statistical significance (VIP [1 and P value \0.05) were considered metabolites responsible for the differentiation of B-cell NHL from healthy controls. The metabolites were identified via online ChemSpider Database integrated in MarkerLynx and other metabonomic database, such as METLIN (http://metlin.scripps.edu/) and HMDB database (http://www.hmdb.ca/) with mass tolerance of less than 0.01 kDa. The list of potential candidates can be confirmed further by performing MS/MS experiments (the collision energy was 20–40 eV) and applying isotope ratios and the nitrogen rule (Theodoridis et al. 2012). The discriminating features were then compared to an in-house standard database prepared by injecting standard compounds under identical analytical conditions. Matches in retention times, parent mass ions, mass fragment patterns and isotopic patterns were considered for compound identification. These compounds were highlighted as putative markers and were selectively used in ROC analysis performed by SPSS software (IBM, SPSS 19.0). The corresponding fold change shows how these selected differential metabolites varied between the B-cell NHL and healthy control groups. To further understand the pathogenesis of the disease and identify potential biomarkers, the KEGG (http://www. genome.jp/kegg/ligand.html) database was queried with the names of the metabolites. 2.5 Sample preparation and analysis by GC-TOFMS A 100 lL aliquot of serum sample was spiked with the internal standard solution (10 lL of L-2-chlorophenylalanine in water, 0.3 mg/mL) and vortexed for 20 s. The mixed solution was extracted with 300 lL of methanol/ chloroform (3:1) and vortexed for 30 s (Qiu et al., 2009). After storing for 10 min at -20 °C, the samples were

Serum metabolic profile of B-cell NHL

centrifuged at 12,000 rpm for 15 min at 4 °C. An aliquot of the 300 lL supernatant was transferred to a glass sampling vial to vacuum-dry at room temperature. The residue was derivatized using a two-step procedure. First, 80 lL of methoxyamine (15 mg/mL in pyridine) was added to the vial and kept at 37 °C for 120 min, followed by 80 lL of BSTFA (1 % TMCS) at 70 °C for 60 min. Each 1 lL aliquot of the derivatized solution was injected in splitless mode into an Agilent 7890 gas chromatograph coupled with a Pegasus 4D HT time-of-flight mass spectrometer (LECO Corporation, St Joseph, MI, USA). The B-cell NHL and control samples were run in the order of ‘‘control-patient-’’, alternately, to minimize systematic analytical deviations. Similar to the UPLCQTOFMS analysis, the QC sample was injected every 10 samples during the whole sample analysis. Separation was achieved on a DB-5MS capillary column (30 m 9 250 lm I.D., 0.25 lm film thickness; (5 %-phenyl)-methyl-polysiloxane bonded and cross-linked; Agilent J&W Scientific, Folsom, CA, USA), with helium as the carrier gas at a constant flow rate of 1.0 mL/min. The temperature of injection, transfer interface and ion source was set to 280, 270, and 220 °C, respectively. The GC temperature programming was set to 0.2 min isothermal heating at 80 °C, followed by 10 °C/min oven temperature ramps to 180 °C, 5 °C/min to 240 °C, and 20 °C/min to 280 °C, and a final 11 min maintenance at 280 °C. Electron impact ionization (70 eV) at full scan mode (m/z 20–600) was used, with an acquisition rate of 10 spectrum/second in the TOFMS setting. 2.6 GC-TOFMS data analysis The GC-TOFMS data was analyzed by ChromaTOF software (v 4.50.8.0, LECO, St Joseph, MI, USA). After alignment with Statistic Compare component, the CSV file was obtained with three dimension data sets including sample information, peak RT and peak intensities. Any known pseudo positive peaks, such as peaks caused by noise, column bleed and BSTFA derivatization procedure, were removed from the data set. The data set was normalized using the internal standard in each sample, and was Box-Cox power transformed. The resulting data was mean centered and Pareto scaled during chemometric data analysis in the SIMCA-P 12.0 Software package. Similar to the UPLC-QTOFMS data analysis, PCA and OPLS-DA were carried out to discriminate between B-cell NHL patients and healthy controls. In addition to the multivariate statistical method, the Student’s t test and WilcoxonMann–Whitney test were also applied to measure the significance of each metabolite. Metabolites annotation with the NIST 11 Mass Spectral Library linked to ChromaTOF software were manually checked with a similarity of more

than 70 % in addition to the reference standard compounds. ROC analysis was performed using the values determined by the top five metabolites of the VIP result.

3 Results and discussion 3.1 Validation of UPLC-QTOFMS and GC-TOFMS conditions and analysis of metabolic pattern The metabonomic method used in this study established a noninvasive approach that revealed a global view of the metabolism that may be applicable to the surveillance of high-risk populations. Typical UPLC-QTOFMS base peak intensity (BPI) chromatograms of human serum from the B-cell NHL group and the control group are shown in Supplementary Fig. S1, Online Resource 1. Typical GCTOFMS total ion current (TIC) chromatograms of serum samples from a patient and a healthy control are shown in Supplementary Fig. S2, Online Resource 1. There were 2,447 detectable spectral features for ESI? and 1,658 for ESI- in UPLC-QTOFMS data set. And the detectable spectral features in GC-TOFMS were 318 in total, which were used in the following analysis. In PCA analysis, the repeated QC injections were clustered in a small space in scores plots (Supplementary Fig. S3, Online Resource 1). This indicated that the relative differences between the repeated injections are much smaller than the biological differences between the test samples. Extracted ion chromatograms (EICs) of selected ions of internal standard using UPLC-QTOFMS analysis were also compared between the various QC injections. The variations (RSD) of the RT and peak area of internal standard were \0.3 and \7 %, respectively. The RSD of the peak area of internal standard for each real sample was \25 % prior to normalization. Supplementary Fig. S3 also shows the PCA score plots representing the distribution between the B-cell NHL and the control group in two dimensions. The obvious separation suggested that serum biochemical perturbation significantly occurred in patient group. OPLS-DA was used to enhance the classification performance. Supplementary Fig. S4 shows a clear separation between the patients and healthy controls using UPLC-QTOFMS positive ion mode analysis, negative ion mode analysis and GC-TOFMS analysis by OPLS-DA. However, a separate OPLS-DA model without the healthy group failed to discriminate different pathological stages (Ann Arbor Stage and International Prognostic Index Score) of B-cell NHL patients neither in UPLCQTOFMS analysis nor in GC-TOFMS analysis. A permutation test performed with 200 random permutations in the PLS-DA models shows that R2 and Q2 values from the permuted analysis are significantly lower

123

J. Zhou et al. Table 2 Representative differential metabolites contributed for the separation between the B-cell NHL patients and the healthy controls derived from UPLC-QTOFMS, their relative intensities in metabolic profiles and metabolic pathways No.

RT (min)_m/z

VIPa

Adduct

P valueb

P valuec

Fold changed (B-cell NHL/ control)

Metabolite

Metabolic pathway

Cholinee

Glycine, serine and threonine metabolism

20 ,30 ,40 Trihydroxyacetophenonef

Others

Positive ion mode 1

0.6_104.11

[M ? H]?

3.37

1.65 9 10-21

2.59 9 10-16

-1.87

2

5.2_169.05

[M ? H]?

2.55

7.43 9 10-28

1.58 9 10-18

-15.68

3

8.7_197.08

[M ? H]?

1.21

2.78 9 10-29

2.66 9 10-18

-3.58

Maltyl isobutyratef

Others

4

8.1_201.09

[M ? H]

?

2.69

1.23 9 10

-26

-18

-2.82

Dehydrotremetonef

Others

5

12.7_369.35

[M ? H– H2O]?

1.09

1.78 9 10-7

2.31 9 10-7

Cholesterole

Steroid biosynthesis

6

10.1_379.28

[M ? H]?

2.13

3.75 9 10-28

7.87 9 10-18

-4.37

MG(18:1(9Z)/0:0/0:0)f

Others

1.31

1.71 9 10

-11

-10

-1.83

6-Keto-prostaglandin F1af

Arachidonic acid metabolism

Glycochenodeoxycholic acide

Bile acid biosynthesis

-1.95

Glycocholic acide

Bile acid biosynthesis

-1.52

LysoPC (14:0)e

Glycerophospholipid metabolism

?

9.35 9 10

1.33

7

7.8_393.23

[M ? Na]

8

4.7_414.3

[M ? H– 2H2O]?

1.23

2.59 9 10-2

2.40 9 10-4

9

7.9_466.33

[M ? H]?

1.95

5.84 9 10-16

3.47 9 10-14

10

5.4_468.31

[M ? H]

?

1.94

2.23 9 10

11

6.6_480.34

[M ? H]?

2.84

1.71 9 10-20

5.55 9 10-15

-1.72

LysoPC (P-16:0)e

Glycerophospholipid metabolism

12

6.3_496.34

[M ? H]?

10.72

1.34 9 10-8

3.45 9 10-8

-1.19

LysoPC (16:0)e

Glycerophospholipid metabolism

13

6.9_510.36

[M ? H]?

3.56

2.34 9 10-23

4.21 9 10-16

-1.82

LysoPC (17:0)e

Glycerophospholipid metabolism

14

6.6_522.36

[M ? H]?

5.00

3.19 9 10-6

2.87 9 10-5

-1.25

LysoPC (18:1(11Z))e

Glycerophospholipid metabolism

15

7.6_524.37

[M ? H]?

15.25

7.44 9 10-19

2.10 9 10-14

-1.46

LysoPC (18:0)e

Glycerophospholipid metabolism

-5

6.89 9 10

1.47 9 10

-5

1.76

Negative ion mode 8.5_303.23

[M–H]-

7.04

7.90 9 10-22

2.21 9 10-14

-1.92

Arachidonic Acide

Arachidonic acid metabolism

17

8.3_327.23

[M–H]

-

4.90

7.08 9 10

-13

-13

-2.18

Docosahexaenoic acide

Biosynthesis of unsaturated fatty acids

18

4.6_407.28

[M–H]-

2.21

2.02 9 10-3

2.31 9 10-2

2.70

Cholic acide

Bile acid biosynthesis

19

5.9_500.28

[M–H]

-

3.30

6.06 9 10

-9

-9

1.46

Glycerophospho-NArachidonoyl Ethanolaminef

Others

20

4.3_510.25

[M–H2O–H]-

1.88

1.52 9 10-6

3.88 9 10-7

1.92

N-[(3a,5b,7b)-7-hydroxy-24oxo-3-(sulfooxy)cholan-24yl]-Glycinef

Others

21

3.4_609.22

[M ? FA–H]-

1.72

2.13 9 10-8

2.67 9 10-8

3.12

Aerobactinf

Lysine degradation

PA(16:0/16:0)f

Glycerolipid metabolism

Coproporphyrinf

Porphyrin and chlorophyll metabolism

16

a

22

8.5_629.45

[M–H2O–H]

23

3.1_653.26

[M–H]-

-

-21

3.33

1.25 9 10

2.77

6.61 9 10-9

5.39 9 10

5.35 9 10

1.26 9 10

-14

8.52 9 10-9

-2.99 3.33

Variable importance in the projection (VIP) was obtained from OPLS-DA with a threshold of 1.0

b

P values were calculated from student’s t test (P \ 0.05)

c

P values were calculated from Wilcoxon-Mann–Whitney test

d

Fold change (FC) was calculated from the arithmetic mean values of each group. Fold change with a positive value indicates a relatively higher concentration present in B-cell NHL patients while a negative value means a relatively lower concentration as compared to the healthy controls e

Metabolites formally identified by standard samples

f

Metabolites putatively annotated by databases searching

than corresponding original values. The validation test indicated the robustness, the goodness of fit of the data and consequently the predictive ability of the

123

metabonomic method used in this study. The model quality results are summarized in Supplementary Table S1, Online Resource 2.

Serum metabolic profile of B-cell NHL

3.2 Identification of differential metabolites and construction of diagnostic models Twenty-three endogenous metabolites contributing to the differences between the disease and control groups from UPLC-QTOFMS positive ion mode and negative ion mode were identified using MS spectral databases and 13 of them were confirmed using reference standards (Table 2). The related metabolic pathways were searched in KEGG database. Among the metabolites identified from UPLCQTOFMS, 20 ,30 ,40 -trihydroxyacetophenone was the serum metabolite found to be the most depleted in B-cell NHL patients, compared to controls, showing the greatest fold change (FC = -15.68). Coproporphyrin was the metabolite mostly increased (FC = 3.33) in B-cell NHL patients. The significantly altered serum metabolites also included decreased docosahexaenoic acid, glycocholic acid, choline and LysoPC (17:0), and elevated cholic acid, glycochenodeoxycholic acid, and cholesterol in the serum of B-cell NHL patients compared with healthy controls. Sixteen differential metabolites identified from GCTOFMS are demonstrated in Table 3 using a method similar to UPLC-QTOFMS analysis. Among these, two metabolites were also detected in the GC-TOFMS analysis with the same direction. They are decreased levels of arachidonic acid and docosahexaenoic acid in the B-cell NHL patients. The most significantly altered serum metabolites included decreased benzenebutanoic acid (FC = -33.98), D-2-hydroxyoctanoic acid, b-hydroxypyruvic acid and L-serine, and elevated pyruvic acid (FC = 10.37), 3-hydroxydecanoic acid, and 2,3-butanediol in the serum of B-cell NHL patients. Eight representative metabolites obtained from UPLCQTOFMS analysis with characteristic expression levels between the two groups are shown in Fig. 1a. They were selected from the metabolites annotated with the aid of available reference standards in our lab. Although the OPLS-DA model failed to distinguish the stages of B-cell NHL by multivariate statistical method, there are a number of serum metabolites which showed a consistent trend of alteration (up- or down-regulation) from healthy control to B-cell NHL with extranodal involvement (Fig. 1b). The concentrations of these metabolites altered at different pathological states. The patients (n = 27) with one or more extranodal disease sites such as liver, spleen, stomach, intestine, brain, bone marrow, and muscle, are at higher risk and may have worse prognosis than the patients without extranodal involvement (n = 27). Chen et al. (1979) found that extranodal involvement was less frequent in the nodular (19 %) than in the diffuse (52 %) group, showing that it was associated with relapse-free and actuarial survival. We hypothesized that the trends of these metabolites in presence and absence of extranodal disease

might be due to the different tumor cell subtypes of the patients of the two groups. Thus, these metabolites can help us to further understand the underlying biological alterations associated with NHL progression. Similar to UPLCQTOFMS analysis, five metabolites obtained from GCTOFMS analysis with characteristic expression levels between controls and B-cell NHL patients are shown in Fig. 1c. They were the top five metabolites from Table 3 ranked by VIP value. Figure 1d shows the four representative metabolites obtained from GC-TOFMS analysis with characteristic expression levels among controls and different extranodal involvements. To test the robustness of the model and further validate this noninvasive UPLC-QTOFMS and GC-TOFMS profiles and identify metabolites with the potential diagnostic potential, the receiver operating characteristic (ROC) curve analysis was performed using the values of marker metabolites. Choline, arachidonic acid, LysoPC (17:0), PA (16:0/16:0) and coproporphyrin were selected from the results of UPLC-QTOFMS analysis, according to their VIP value and P value between the two groups, and their chemical class (the marker metabolites represent different chemical class and metabolic pathway). Similarly, benzenebutanoic acid, b-hydroxypyruvic acid, D-2-hydroxyoctanoic acid, pyruvic acid and arachidonic acid derived from GC-TOFMS analysis (the same as the differential metabolites in Fig. 1c) were selected by P value to construct another diagnostic model using ROC analysis. The area under the curve (AUC) was used to assess the candidate biomarkers of B-cell NHL, which is equal to the probability that this classifier will rank a randomly chosen positive instance higher than a randomly chosen negative one (Hsieh et al. 2011). ROC analysis by the cross-validated predicted Y-values (CVPY) of the PLS-DA model based on data of the UPLC-QTOFMS profile, revealed AUC of 0.968 with a sensitivity of 100 % and specificity of 94.5 % for discriminating B-cell NHL patients from healthy people, while AUC reached 1.00 for the result of GCTOFMS, with a sensitivity of 100 % and specificity of 100 %. To confirm the reliability of the potential markers from UPLC-QTOFMS, we took a new batch of samples of B-cell NHL patients and healthy controls by using the same method. The clinical information of these samples is provided in the Supplementary Table S2, Online Resource 2. The ROC curve based on the same five representative metabolites repeatedly yielded satisfactory result. The AUC reached 0.971 with a sensitivity of 100 % and specificity of 94.7 % for the extra batch of samples. The OPLS-DA model score plots from both positive and negative ion mode showed clear separations between B-cell NHL patients and healthy controls, with satisfactory modeling and predictive abilities (Supplementary Table S3, Online Resource 2). Consequently, these metabolites could

123

J. Zhou et al. Table 3 Representative differential metabolites contributed for the separation between the B-cell NHL patients and the healthy controls derived from GC-TOFMS, their relative intensities in metabolic profile and metabolic pathways No.

1

RT (min)

VIPa

P valueb

P valuec

Fold changed (B-cell NHL/control)

8.0

2.29

8.32 9 10-3

2.38 9 10-3

-7

-6

10.37

1.34

2

8.5

3.88

5.32 9 10

2.39 9 10

3

9.1

2.09

7.54 9 10-3

1.77 9 10-3

1.56

4

11.1

2.97

2.77 9 10-4

5.69 9 10-5

-1.40

-3

-4

-1.26

5

14.8

2.69

1.34 9 10

6.76 9 10

6

15.0

5.22

7.44 9 10-14

6.88 9 10-9

-33.98

7

15.1

3.64

5.19 9 10-7

1.57 9 10-6

-2.04

8

17.8

2.36

3.38 9 10-3

1.47 9 10-3

1.41

-13

-11

Metabolite

Metabolic pathway

Isopropanole Pyruvic acid

Others e

Glycolysis/Gluconeogenesis

3-Hydroxydecanoic acide

Others

Glycerolf

Galactose metabolism f

Pyroglutamic acid

Glutathione metabolism

Benzenebutanoic acide

Butanoate metabolism

L-Serine

f

Glycine, serine and threonine metabolism

2,3-Butanediolf

Butanoate metabolism

-3.48

b-Hydroxypyruvic acide

Glycine, serine and threonine metabolism

9

18.3

4.83

2.67 9 10

1.33 9 10

10

22.2

2.10

8.28 9 10-3

4.60 9 10-3

-1.86

(S)-2-Acetolactatee

Valine, leucine and isoleucine biosynthesis

11

22. 9

3.05

1.46 9 10-4

9.95 9 10-6

-1.87

10-Hendecenoic acidf

Others

12

25.1

2.78

4.74 9 10-4

8.24 9 10-5

-1.64

Linoleic acidf

Linoleic acid metabolism

-7

-7

f

13 14

26.9 29.3

3.84 2.72

5.56 9 10 1.13 9 10-3

4.67 9 10 2.11 9 10-4

-1.81 -1.91

Arachidonic acid Docosahexaenoic acidf

Arachidonic acid metabolism Biosynthesis of unsaturated fatty acids

15

29.5

2.84

3.59 9 10-4

1.05 9 10-4

-1.71

MG(16:0/0:0/0:0)e

Others

-7

-3.56

D-2-Hydroxyoctanoic acide

Others

16 a

32.4

4.41

-10

4.13 9 10

1.80 9 10

Variable importance in the projection (VIP) was obtained from OPLS-DA with a threshold of 1

b

P values were calculated from student’s t test (P \ 0.05)

c

P values were calculated from Wilcoxon-Mann–Whitney test

d

Fold change was calculated from the arithmetic mean values of each group. Fold change with a positive value indicates a relatively higher concentration present in B-cell NHL patients while a negative value means a relatively lower concentration as compared to the healthy controls

e

Metabolites were identified using available library databases

f

Verified by reference compounds

potentially serve as promising biomarkers for the early detection and prognosis of B-cell NHL. 3.3 Study of the changed metabolic trend The changed metabolic trend is the terminal behavior of changed metabolic pattern induced by B-cell NHL. In the present study, a total of 37 differential metabolites were identified using the two analytical platforms (arachidonic acid and docosahexaenoic acid were detected in both instruments). Variation of the level of hypoxanthine, the urinary marker for NHL reported by Yoo et al. (2010), was not significant in this study. Lysophosphatidylcholines (LysoPCs, LPC) are products or metabolites of phosphatidylcholines (PCs). Among the metabolites summarized in Table 2, six are phospholipids. The significantly lower levels of lysophosphatidylcholine

123

were observed in B-cell NHL samples compared with healthy controls. Moreover, LPC regulates a variety of biological processes, including cell proliferation, tumor cell invasiveness, and inflammation (Yang et al. 2006). Low levels of LPCs imply an anti-inflammatory status in NHL patients, and represent a severe immune suppression status. Similar LPC trends have also been found in other malignant diseases, such as leukemia, renal cell carcinoma, gastrointestinal cancer, and liver diseases (Lin et al. 2011; Wang et al. 2012). Bile acids (BAs) represent characteristic constituents of human and animal bile, and play critical roles in lipid absorption and metabolism. They are signaling molecules coordinating hepatic triglyceride, glucose and energy homeostasis which reflect internal secretion, especially during diseases and medications. These characters made BAs important target analytes in physiological and

Serum metabolic profile of B-cell NHL

Fig. 1 a Bar charts of mean intensity of eight representative differential metabolites in serum samples of B-cell NHL patients and healthy controls (mean ± SE) using UPLC-QTOFMS analysis. b Bar charts of mean intensity of six representative metabolites in serum samples of B-cell NHL patients with different extranodal involvements and healthy controls (mean ± SE) using UPLCQTOFMS analysis. c Bar charts of mean intensity of five

representative differential metabolites in serum samples of B-cell NHL patients and healthy controls (mean ± SE) using GC-TOFMS analysis. d Bar charts of mean intensity of four representative metabolites in serum samples of B-cell NHL patients with different extranodal involvements and healthy controls (mean ± SE) using GC-TOFMS analysis

pathological tests, as well as marker compounds in metabonomic studies (Qiao et al. 2012). The present study justifies the finding of increased serum chenodeoxycholic acid, glycochenodeoxycholic acid and cholic acid, and decreased serum glycocholic acid, suggests that liver dysfunction happened in B-cell NHL. Bacterial dehydrolases may be responsible for the appearance of deoxycholate. Additionally, bile acids markedly alter the expression of various genes involved in cholesterol and phospholipid homeostasis, which results in cell death and inflammation and leads to severe liver injury (Matsubara et al. 2011). The metabolic transformation of bile acids in B-cell NHL patients is characterized in Supplementary Fig. S5, Online Resource 1. Fatty acids are the primary materials to produce phospholipids, which are important in forming biomembranes. The present research also showed that the serum levels of arachidonic acid (AA) and docosahexaenoic acid (DHA) were significantly decreased in B-cell NHL patients.

Arachidonic acid plays a central role in inflammation related to injury and many diseased states. PGs (e.g. PGI2, TXA2) derived from AA show multiple beneficial effects against cardiovascular risks, including reducing blood pressure and inhibiting platelet aggregation (Zhang et al. 2009). Phospholipase A2 mediates hydrolysis of fatty acids, especially arachidonate, from sn-2 position of membrane phospholipids, and it plays an important role in this rate limiting step for the biosynthesis of AA (Moreno 1993). Thus, the down-regulation of AA and LPCs in this study seems reasonable from this point of view. Cyclooxygenase-2 (COX-2) plays a vital role in the AA metabolism which converts arachidonate into prostaglandin E2 (PGE2). A recent study (Mestre et al. 2012) showed that COX-2 was expressed on Reed-Sternberg cells in one-third of Hodgkin’s Lymphoma (HL) patients and was a major independent, unfavorable prognostic factor in early stage HL. Thus, the down-regulated AA in our study may be

123

J. Zhou et al.

partially explained by the over expression of COX-2 in NHL. Docosahexaenoic acid, a kind of omega-3 fatty acid derived from linolenic acid, is needed in tissue in order to maximize the beneficial formation of eicosanoids, lipoxins and resolvins to prevent atherosclerosis. DHA had been shown to exert antioxidant effects (Hashimoto et al. 2002; Guillot et al. 2008) and significant free radical scavenging capacity (Innis 2007) in the brain. In this study, a significantly lower level of DHA was observed in patients with B-cell NHL, indicating the presence of an excessive oxidative stress that might have resulted from diminished neuron-protection. In the present study we have also found an increase of coproporphyrin, which belongs to the pathway of porphyrin and chlorophyll metabolism. The studies by Madhuri et al. (2003) and Masilamani et al. (2004) show elevated levels of porphyrins in cancer patients compared to healthy patients. Pinellia et al. (2005) reported that patients with Hodgkin’s disease (HD) showed altered urinary porphyrin metabolism, and suggested that the increased levels of urinary coproporphyrins seem to be due to the disease itself, disease progression should be associated with higher levels of porphyrin excretion. Elevation of serum coproporphyrin level may be due to low heme levels stimulating enzymatic processes, such as ALA synthase, ALA dehydrase or uroporphyrinogen synthase, thus leading to a high output of ALA, PBG and coproporphyrins. Pyruvic acid was found by GC-TOFMS at higher levels in patients with B-cell NHL, presumably due to the increased glycolysis to meet the increased intracellular energy demand, which is termed as the ‘‘Warburg effect’’. Glucose, pyruvate, and lactate are all interrelated through the glycolytic chain and therefore directly affect lipid metabolism through the metabolic intermediate acetylCoA, the primary carbon building block in both unsaturated lipid and cholesterol biosynthesis. Indeed, many clinical metabonomics studies have identified significant elevations in pyruvate, lactate and alanine in various cancers such as colorectal, liver and kidney cancers (Ng et al. 2011). L-serine is a key metabolite in the pathway of sphingolipid metabolism. Pyruvate would generate the intermediate needed for synthesis of serine. It is crucial to denote that serine is needed for the metabolism of fats and fatty acids, muscle growth and to maintain a healthy immune system and was previously found to decrease in plasma under inflammatory conditions (Suliman et al. 2005). The down regulation of L-serine in our study confirmed that the elevation of pyruvic acid significantly perturbed the glycine, serine and threonine metabolism pathway and fatty acid metabolism. Biomarker patterns containing a group of biomarkers of different pathways could be more effective in

123

discriminating cancer patients and more informative to elucidate the pathophysiology of a cancer, rather than the attempt to pick out a single one. According to the biomarker patterns discovered in this study, B-cell NHL patients could correctly be diagnosed and differentiated from healthy controls. Furthermore, several metabolic pathways have been found closely related to the pathophysiological development of B-cell NHL.

4 Concluding remarks This work was for the first time to investigate holistic metabolic profiles from serum samples of patients with B-cell NHL. Importantly, the metabonomic method used in this study established a noninvasive approach that revealed a global view of the metabolism that may be applicable to the surveillance of high-risk populations. This non-targeted two chromatography—mass spectrometry platforms— based metabolomic approach confirmed the powerful differentiation ability of the B-cell NHL patients and normal controls, with the identification of as many as 37 novel differential metabolites in serum, which could be utilized for improved NHL detection and prognosis. The results of UPLC-QTOFMS were validated with another batch of samples. We observed the same most important variables using the extra batch of NHL patients and controls, which indicates the robustness of the method. Our results may also provide an insight into the understanding of the biochemical network and pathway in the B-cell NHL and other types of NHL, further investigation towards this direction is in progress. Acknowledgments This work was supported by the National Natural Science Foundation of China (21175092 and 21105064) and Specially-funded Programme on the Development of National Key Scientific Instrument and Equipment (2011YQ150072, 2011YQ 15007204, 2011YQ15007207, 2011YQ15007210). Also, this work was supported, in part, by the National Natural Science Foundation of China (20705021 and 81172254), and Shanghai Commission of Science and Technology (11JC1407300). We gratefully acknowledge Dr. Lei Feng and Dr. Yumin Liu from Instrumental Analysis Center of Shanghai Jiao Tong University for expert assistance in UPLCQTOFMS and GC-TOFMS analysis, and Dr. Xi Zhang from Waters Technologies (Shanghai) Ltd. for metabolites identification. Conflict of interest The all authors declared that they have no conflict of interest in the submission of this manuscript.

References Ali, K., Iqbal, M., Korthout, H., et al. (2012). NMR spectroscopy and chemometrics as a tool for anti-TNFa activity screening in crude extracts of grapes and other berries. Metabolomics. doi:10.1007/ s11306-012-0406-8.

Serum metabolic profile of B-cell NHL Chen, M. G., Prosnitz, L. R., Gonzalez-Serva, A., & Fischer, D. B. (1979). Results of radiotherapy in control of stage I and II nonHodgkin’s lymphoma. Cancer, 43(4), 1245–1254. Cheng, Y., Xie, G., Chen, T., et al. (2012). Distinct urinary metabolic profile of human colorectal cancer. Journal of Proteome Research, 11(2), 1354–1363. Emily, J. G., & Michael, R. B. (2012). Overview of non-Hodgkin’s lymphoma. Disease-a-month, 58(4), 208–218. Guillot, N., Debard, C., Calzada, C., Vidal, H., Lagarde, M., & Vericel, E. (2008). Effects of docosahexaenoic acid on some megakaryocytic cell gene expression of some enzymes controlling prostanoid synthesis. Biochemical and Biophysical Research Communications, 372(4), 924–928. Han, H. S., Escalo´n, M. P., Hsiao, B., Serafini, A., & Lossos, I. S. (2009). High incidence of false-positive PET scans in patients with aggressive non-Hodgkin’s lymphoma treated with rituximab-containing regimens. Annals of Oncology, 20(2), 309–318. Hashimoto, M., Hossain, S., Shimada, T., et al. (2002). Docosahexaenoic acid provides protection from impairment of learning ability in Alzheimer’s disease model rats. Journal of Neurochemistry, 81(5), 1084–1091. Hsieh, S. Y., He, J. R., Yu, M. C., et al. (2011). Secreted ERBB3 isoforms are serum markers for early hepatoma in patients with chronic hepatitis and cirrhosis. Journal of Proteome Research, 10(10), 4715–4724. Innis, S. M. (2007). Dietary (n-3) fatty acids and brain development. The Journal of Nutrition, 137(4), 855–859. Juweid, M. E., & Hoekstra, O. S. (2011). Positron emission tomography. Methods in molecular biology (pp. 1–19). New York: Springer Verlag. Lin, L., Huang, Z., Gao, Y., Yan, X., Xing, J., & Hang, W. (2011). LC-MS based serum metabonomic analysis for renal cell carcinoma diagnosis, staging, and biomarker discovery. Journal of Proteome Research, 10(3), 1396–1405. Madhuri, S., Vengadesan, N., Aruna, P., Koteeswaran, D., Venkatesan, P., & Ganesan, S. (2003). Native fluorescence spectroscopy of blood plasma in the characterization of oral malignancy. Photochemistry and Photobiology, 78(2), 197–204. Martinez-Climent, J. A., Fontan, L., Gascoyne, R. D., Siebert, R., & Prosper, F. (2010). Lymphoma stem cells: Enough evidence to support their existence? Haematologica, 95(1), 293–302. Masilamani, V., Al-Zhrani, K., Al-Salhi, M., Al-Diab, A., & AlAgeily, M. (2004). Cancer diagnosis by autofluorescence of blood components. Journal of Luminescence, 109(3–4), 143–154. Matsubara, T., Tanaka, N., Patterson, A. D., Cho, J. Y., Krausz, K. W., & Gonzalez, F. J. (2011). Lithocholic acid disrupts phospholipid and sphingolipid homeostasis leading to cholestasis in mice. Hepatology, 53(4), 1282–1293. Mestre, F., Gutierrez, A., Ramos, R., et al. (2012). Expression of COX-2 on Reed-Sternberg cells is an independent unfavorable prognostic factor in Hodgkin lymphoma treated with ABVD. Blood, 119(25), 6072–6079. Moreno, J. J. (1993). Effect of aristolochic acid on arachidonic acid cascade and in vivo models of inflammation. Immunopharmacology, 26(1), 1–9.

Ng, D. J. Y., Pasikanti, K. K., & Chan, E. C. Y. (2011). Trend analysis of metabonomics and systematic review of metabonomicsderived cancer marker metabolites. Metabolomics, 7(2), 155–178. Nicholson, J. K., Lindon, J. C., & 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(11), 1181–1189. Pinellia, A., Mussinib, C., Burattic, M., Parmiggiani-Veneziab, M., & Trivulzioa, S. (2005). Increased urinary coproporphyrin excretion observed in patients with differently staged Hodgkin’s disease treated with chemotherapy. Pharmacological Research, 51(3), 283–288. Qiao, X., Ye, M., Liu, C., et al. (2012). A tandem mass spectrometric study of bile acids: Interpretation of fragmentation pathways and differentiation of steroid isomers. Steroids, 77(3), 204–211. Qiu, Y., Cai, G., Su, M., et al. (2009). Serum metabolite profiling of human colorectal cancer using GC-TOFMS and UPLCQTOFMS. Journal of Proteome Research, 8(10), 4844–4850. Qiu, Y., Cai, G., Su, M., et al. (2010). Urinary metabonomic study on colorectal cancer. Journal of Proteome Research, 9(3), 1627–1634. Suliman, M. E., Qureshi, A. R., Stenvinkel, P., Pecoits-Filho, R., Barany, P., et al. (2005). Inflammation contributes to low plasma amino acid concentrations in patients with chronic kidney disease. The American Journal of Clinical Nutrition, 82(2), 342–349. Theodoridis, G. A., Gika, H. G., Want, E. J., & Wilson, I. D. (2012). Liquid chromatography–mass spectrometry based global metabolite profiling: A review. Analytica Chimica Acta, 711, 7–16. van der Kloet, F. M., Bobeldijk, I., Verheij, E. R., & Jellema, R. H. (2009). Analytical error reduction using single point calibration for accurate and precise metabolomic phenotyping. Journal of Proteome Research, 8, 5132–5141. Wang, B., Chen, D., Chen, Y., et al. (2012). Metabonomic profiles discriminate hepatocellular carcinoma from liver cirrhosis by ultraperformance liquid chromatography–mass spectrometry. Journal of Proteome Research, 11(2), 1217–1227. Yang, J., Zhao, X., Liu, X., et al. (2006). High performance liquid chromatography-mass spectrometry for metabonomics: Potential biomarkers for acute deterioration of liver function in chronic hepatitis B. Journal of Proteome Research, 5(3), 554–561. Yoo, B. C., Kong, S. Y., Jang, S. G., et al. (2010). Identification of hypoxanthine as a urine marker for non-Hodgkin lymphoma by low-mass-ion profiling. BMC Cancer, 10, 55. Zhang, H., Wang, Y., Gu, X., Zhou, J., & Yan, C. (2011). Metabolomic profiling of human plasma in pancreatic cancer using pressurized capillary electrochromatography. Electrophoresis, 32(3–4), 340–347. Zhang, Q., Wang, G., J., A., et al. (2009). Application of GC/MSbased metabonomic profiling in studying the lipid-regulating effects of Ginkgo biloba extract on diet-induced hyperlipidemia in rats. Acta Pharmacologica Sinica, 30(12), 1674–1687.

123