Screening, Verification, and Optimization of Biomarkers for Early

Apr 28, 2015 - (20, 21) Blank and QC samples were injected every 5 h to test the stability of ...... Gika , H. G.; Theodoridis , G. A.; Wingate , J. E...
0 downloads 0 Views 4MB Size
Download PDF
Article pubs.acs.org/jpr

Screening, Verification, and Optimization of Biomarkers for Early Prediction of Cardiotoxicity Based on Metabolomics Yubo Li,†,‡ Liang Ju,†,‡ Zhiguo Hou,†,‡ Haoyue Deng,†,‡ Zhenzhu Zhang,†,‡ Lei Wang,†,‡ Zhen Yang,†,‡ Jia Yin,†,‡ and Yanjun Zhang*,† †

Tianjin State Key Laboratory of Modern Chinese Medicine and ‡School of Traditional Chinese Materia Medica, Tianjin University of Traditional Chinese Medicine, 312 Anshan West Road, Tianjin 300193, China S Supporting Information *

ABSTRACT: Drug-induced cardiotoxicity seriously affects human health and drug development. However, many conventional detection indicators of cardiotoxicity exhibit significant changes only after the occurrence of severe heart injuries. Therefore, we investigated more sensitive and reliable indicators for the evaluation and prediction of cardiotoxicity. We created rat cardiotoxicity models in which the toxicity was caused by doxorubicin (20 mg/kg), isoproterenol (5 mg/kg), and 5fluorouracil (125 mg/kg). We collected data from rat plasma samples based on metabolomics using ultra-performance liquid chromatography quadrupole time-offlight mass spectrometry. Following multivariate statistical and integration analyses, we selected 39 biomarker ions of cardiotoxicity that predict cardiotoxicity earlier than biochemical analysis and histopathological assessment. Because drugs with different toxicities may cause similar metabolic changes compared with other noncardiotoxic models (hepatotoxic and nephrotoxic models), we obtained 10 highly specific biomarkers of cardiotoxicity. We subsequently used a support vector machine (SVM) to develop a predictive model to verify and optimize the exclusive biomarkers. L-Carnitine, 19-hydroxydeoxycorticosterone, LPC (14:0), and LPC (20:2) exhibited the strongest specificities. The prediction rate of the SVM model is as high as 90.0%. This research provides a better understanding of drug-induced cardiotoxicity in drug safety evaluations and secondary development and demonstrates novel ideas for verification and optimization of biomarkers via metabolomics. KEYWORDS: metabolomics, biomarkers for early prediction of cardiotoxicity, support vector machine, UPLC−Q-TOF-MS

1. INTRODUCTION The toxic effects of drugs in safety evaluations and secondary development are of great concern to medical researchers.1 The incidence of acute cardiotoxicity caused by adverse drug reactions is fairly high, and recovery is often difficult. Many drugs can cause serious cardiotoxicity, which has many negative symptoms, including high blood pressure and heart failure, and can lead to death.2 Many drugs, such as the potent anthracycline anticancer drug doxorubicin (DOX), the βadrenergic receptor agonist isoproterenol (ISO), the pyrimidine chemotherapy drug 5-fluorouracil (5-FU), and traditional Chinese medicines (e.g., aconite root and nux vomica), have limited applications in both secondary development and clinics because of the cardiotoxicity effects.3,4 Therefore, the early identification of drug-induced toxic effects is important for drug safety assessment and disease diagnosis, monitoring, and prevention. There is a period of time between cardiotoxic drug exposure and absorption during which drug-induced toxicities can be identified early by rapid and reliable prediction methods. Currently, biochemical analyses [lactate dehydrogenase (LDH), creatine kinase (CK), creatine kinase isoenzyme (CK-MB), and malondialdehyde (MDA)] are used as cardiotoxic indicators to identify heart injury in conventional drug safety evaluations.5,6 © XXXX American Chemical Society

However, these biochemical indicators have severe limitations because they exhibit significantly increased concentrations only after substantial heart injury has occurred.7 Therefore, a rapid and effective method for the detection and early prevention of cardiotoxicity must be developed, which could effectively reduce drug development costs and monitor drug toxicity in clinics. As a research method, metabolomics can reflect changes in endogenous substances in different physiological or pathological states. Metabolomics has unique advantages in drug safety evaluation, toxicity prediction, and disease diagnosis because of its high sensitivity, broad coverage, and good reproducibility.8−11 Ultra-performance liquid chromatography coupled with quadrupole time-of-flight mass spectrometry (UPLC−Q-TOF-MS), which provides high resolution, high sensitivity, and rapid separation for untargeted metabolomics, is a better method to detect and identify complex endogenous substances compared with other analytical techniques.12−14 In addition, new data processing and analysis techniques for metabolomics provide a broader range of applications. The support vector machine (SVM) is a branch of neural networks Received: October 28, 2014

A

DOI: 10.1021/pr501116c J. Proteome Res. XXXX, XXX, XXX−XXX

Article

Journal of Proteome Research that is based on empirical risk minimization principles; it is widely used in the fields of medicine, disease diagnosis and image processing and is based on classification and regression learning processes.15−18 Therefore, metabolomics combined with SVM could be beneficial for future research. In this study, we developed rat models of cardiotoxicity with several drugs (including DOX, ISO, and 5-FU). We subsequently conducted nontargeted metabolomic analysis using UPLC−Q-TOF-MS with plasma samples for different time periods. By using multivariate statistical analysis [principal component analysis (PCA) and partial least-squares-discriminant analysis (PLS-DA)], we identified unified biomarkers in which significant changes were induced by different drugs in the early stage of cardiotoxicity; these changes were referred to as biomarkers for the early prediction of cardiotoxicity. To avoid changes in the same metabolites caused by different toxicity drugs, we combined the plasma samples of hepatotoxicity and nephrotoxicity with the plasma samples of cardiotoxicity for verification and optimization via the SVM. Finally, exclusive biomarkers for the early prediction of cardiotoxicity were obtained. The aim of this study is to develop a metabolomics method with high sensitivity and strong specificity for the prediction of cardiotoxicity that will provide a reliable basis for the diagnosis of drug-induced cardiotoxicity.

Table 1. Dose, Mode of Administration, and Sampling Time at the Process of Preliminary Screening Biomarkers for Early Prediction of Cardiotoxicity drug

grouping

number

dose (mg/kg)

NS DOX

NS DOX-6h DOX-12h DOX-24h ISO-12h ISO-1d ISO-3d

10 10 10 10 10 10 10

2 mL 20 20 20 5 5 5

5-FU-1d 5-FU-2d

10 10

125 125

5-FU-3d

10

125

ISO

5-FU

mode of administration i.p., single-dose i.p., single-dose i.p., single-dose i.p., single-dose i.p., single-dose i.p., single-dose i.p., successive administration i.g., single-dose i.g., successive administration i.g., successive administration

sampling time 1 day 6h 12 h 24 h 12 h 1 day 3 days 1 day 2 days 3 days

2.3. Sample Collection and Preparation

Prior to sample collection, all animals were fasted for 12 h but allowed access to water to avoid the effects of food on the final results. After blood collection from each group, all animals were sacrificed, and the hearts were immediately removed and stored in 10% formalin solution. The plasma and serum were separated via centrifugation at 3500 rpm for 15 min at 4 °C. The plasma samples were collected and stored at −80 °C until the metabolomic analysis; the serum was used to test the concentrations of LDH, CK, CK-MB, and MDA using an automatic chemical analyzer. The plasma was thawed at room temperature. We used 300 μL of acetonitrile in 100 μL of plasma for the protein precipitation. The resultant mixture was ultrasonicated in cold water for 10 min, vortexed for 1 min, and then centrifuged at 13 000 rpm for 15 min at 4 °C. The supernatants were used for the metabolomic analysis.19 The pathological features of the hearts were examined by hematoxylin and eosin (H&E) staining. The H&E staining process was performed as follows. The heart was trimmed and embedded in paraffin wax. Then, 5 μm thick slices were cut from the hearts and affixed to glass slides. The slices were deparaffinized with xylene, hydrated, stained with hematoxylin for 10 min, differentiated by hydrochloric alcohol, stained with eosin, dehydrated by gradient alcohol, and then cleaned in xylene. Histopathological changes were identified under a light microscope at 100× and 200× magnification. To minimize animal suffering, all experimental procedures were conducted in accordance with the Chinese national legislation and local guidelines. This study was approved by the Animal Ethics Committee of Tianjin University of Traditional Chinese Medicine under permit number TCM-2012−078F01.

2. EXPERIMENTAL SECTION 2.1. Reagents and Materials

High-pressure liquid chromatography (HPLC)-grade acetonitrile and formic acid were purchased from Oceanpak (Gothenburg, Sweden) and ROE (USA), respectively. Distilled water was obtained from Wahaha (Hangzhou, China). Normal saline (NS), DOX, ISO, and 5-FU were purchased from Queensland Technology Co., Ltd. (Tianjin, China) and dissolved in saline solution prior to use. 2.2. Animal Treatment

The experimental animals were purchased from Sibei Fu (Beijing) Experimental Animals Technology Co., Ltd. with license number “SCXK (Jing) 2011−0004”. The animal study was performed at the Institute of Radiation Medicine, Chinese Academy of Medical Sciences (Tianjin, China). Male Wistar rats (weight, 200 ± 20 g) were raised in the following conditions: temperature, 25 ± 1 °C; light/dark cycle, 12 h/12 h. Prior to the experiment, the animals were allowed free access to food and clean water and were acclimated for 1 week. The animal treatment included the following stages. 2.2.1. Screening Stage. One-hundred rats were randomly divided into 10 groups to screen the biomarkers for the early prediction of cardiotoxicity, which included the NS, DOX-6h, DOX-12h, DOX-24h, ISO-12h, ISO-1d, ISO-3d, 5-FU-1d, 5FU-2d, and 5-FU-3d groups. Prior to this experiment, we optimized the animal models. The groups, doses, administration modes, and sampling times are shown in Table 1. 2.2.2. Verification and Optimization Stage. Seventy rats were randomly divided into seven groups, which included the NS group, two cardiotoxicity groups (ISO and 5-FU), two hepatotoxicity groups [Radix Bupleuri and carbon tetrachloride (CCl4)], and two nephrotoxicity groups (gentamicin and etimicin). The groups, doses, administration modes, and sampling times are shown in Table 2.

2.4. Chromatographic and Mass Spectrometric Conditions

The data acquisition was performed on an UPLC−Q-TOF-MS system (Waters, USA). An aliquot of 10 μL of supernatant was injected into an ACQUITY UPLC HSS C18 column (2.1 × 100 mm2, 1.7 μm, Waters). The column temperature was set to 40 °C, and the flow rate was 0.3 mL/min. The UPLC separation system included a binary solvent system with mobile phase A (0.1% formic acid in water) and mobile phase B (0.1% formic acid in acetonitrile). The gradient program was as follows: 99% A followed by 0−0.5 min, A: 99−99%; 0.5−2 min, A: 99−50%; 2−9 min, A: 50−1%; 9−10 min, A: 1−1%; 10−10.5 min, A: 1− 99%; and 10.5−12 min, A: 99−99%. B

DOI: 10.1021/pr501116c J. Proteome Res. XXXX, XXX, XXX−XXX

Article

Journal of Proteome Research

Table 2. Dose, Mode of Administration, and Sampling Time at the Process of Verification and Optimization of Biomarkers for Early Prediction of Cardiotoxicity grouping

drug

number

dose (mg/kg)

mode of administration

sampling time

NS ISO 5-FU gentamicin etimicin Radix Bupleuri CCl4

NS ISO 5-FU gentamicin etimicin Radix Bupleuri CCl4

10 10 10 10 10 10 10

2 mL 5 125 100 100 10 5 mL

i.p., single-dose i.p., single-dose i.g., single-dose i.p., successive administration i.p., single-dose i.g., successive administration subcutaneous injection, single-dose

1 day 12 h 1 day 7 days 3 days 5 days 2 days

Figure 1. At different time of administration, the clinical cardiac biochemistry results (LDH, CK, CK-MB, and MDA) in rats, respectively, compared DOX group, ISO group, and 5-FU group with NS group (n = 10). ∗ means that the group is significantly increased compared with NS group (p < 0.05).

software (Umetrics AB, Umea, Sweden). PCA and PLS-DA are widely used in data processing in metabolomics. In this experiment, PCA was used to determine the similarity between the groups to eliminate outlier samples, and PLS-DA was used to identify significantly changed metabolites in the plasma samples between the different groups. A score plot was used to establish model visualization. Variables with significant differences in their group contribution (VIP > 1) were considered biomarkers, and their molecular structures were further identified. Metabolites were identified by standard experiments, including MS/MS analysis and data matching with HMDB (http://www.hmdb.ca/) and KEGG (http://www.genome.jp/ kegg/). SPSS 17.0 software was used to perform t-tests to determine if the metabolites exhibited statistically significant changes. Venn diagrams (http://bioinfogp.cnb.csic.es/tools/ venny/index.html) were used to identify the correlations of the biomarkers with different drug effects. The identified biomarkers were considered to be unified biomarkers when significant changes were induced by different drugs in the early stage of cardiotoxicity. 2.5.2. Verification and Optimization of Biomarkers for the Early Prediction of Cardiotoxicity. We utilized the metabolomic data of hepatotoxicity and nephrotoxicity to examine the specificity using t-tests compared with the NS

The UPLC system was coupled to a Q-TOF-MS equipped with an electrospray ionization apparatus in positive mode. The MS parameters were as follows: drying gas temperature, 325 °C; drying gas flow, 10 mL/min; desolvation gas flow, 600 L/h; capillary voltage, 3.5 kV; nebulizer pressure, 350 psi; and evaporative and auxiliary gases, high-purity nitrogen. Reference ions [M + H]+ = 556.2771 were employed to ensure accuracy during the spectral acquisition. The data acquisition range was 50−1000 Da. All samples were randomly injected. The samples were singled out from each group and mixed together to form quality control (QC) samples. The QC samples contained all information in the plasma samples and were used to optimize and supervise the process of injection.20,21 Blank and QC samples were injected every 5 h to test the stability of both the samples and the system during acquisition. 2.5. Data Process

2.5.1. Screening of Biomarkers for the Early Prediction of Cardiotoxicity. The raw data of the control and model groups were collected with MarkerLynx Version 4.1 (Waters Corp., Manchester, USA) based on the UPLC−QTOF-MS. The raw data were subsequently processed for peak discovery and peak alignment and filtered to determine potential discriminant variables. The exported data were then processed using multivariate data analysis with SIMCA-P+ 11.5 C

DOI: 10.1021/pr501116c J. Proteome Res. XXXX, XXX, XXX−XXX

Article

Journal of Proteome Research

cardiotoxicity at 1 d post-DOX, 3 d post-ISO, and 3 d post5-FU administrations. The 6 and 12 h DOX administrations, the 12 h and 1 d ISO administrations, and the 1 and 2 d 5-FU administrations did not expose cardiotoxicity. To predict the early cardiotoxicity effects, we chose the 1 d post-DOX, 3 d post-ISO, and 3 d post-5-FU administrations as the models with severe heart damage based on the results of the biochemical analysis and histopathological assessment; thus, we inferred that DOX administered for 6 h, ISO administered for 12 h, and 5-FU administered for 1 d exhibited early stages of cardiotoxicity.

group. We then processed the biomarkers for verification and optimization combined with hepatotoxicity and nephrotoxicity to filter out exclusive biomarkers for the early prediction of cardiotoxicity by SVM. The peak areas of the biomarkers served as input variables to verify and optimize the exclusive biomarkers for accuracy and specificity in SVM. The SVM model was developed in the Matlab (Matlab R2010a, USA) kernel to map from low-dimensional to high-dimensional spaces, and a penalty factor was used to determine the characteristics of the subspace-regulated learning. The confidence and experience risk ratio ranges were determined using the cross-validation method.22,23 In the SVM classification process, the computer was trained by the training set. Next, the classification model was established. Finally, the test set was used to determine the accuracy rate of the model.

3.2. Screening for Biomarkers for the Early Prediction of Cardiotoxicity

Figure S2 of the Supporting Information shows the positive-ion BPI chromatograms of the QC plasma samples obtained by UPLC−Q-TOF-MS. We used six QC samples to determine the precision and repeatability of the detection method; the relevant results are shown in Table S1. The relative standard deviations of the peak areas and retention times of 20 randomly selected chromatographic peaks were less than 15%, which indicates that the sample detection method meets metabolomics requirements. The data that met the metabolomics test conditions were applied in the multivariate statistical analysis. PCA and PLS-DA can identify differences between different groups.25 In this experiment, PCA was used to remove outlier samples, and PLSDA was subsequently used to identify differential metabolites between different groups. The PCA plots are shown in Figure S3. The results indicate that some samples are outliers, and these samples were removed. The PLS-DA scores are shown in Figure 3, panels A−C; the results indicate that the DOX, ISO, and 5-FU group data differ from the NS group at different times. A temporal correlation was also identified. On the basis of the PLS-DA model, we selected the biomarkers with VIP > 1 as potential markers. Here, p < 0.05 was considered significant. To identify biomarkers that are relatively stable and consistently change, we used a Venn diagram to separately integrate the ion information on the DOX, ISO, and 5-FU groups at different times and identified 106, 77, and 69 common ions, respectively (Figure 4). We further conducted an integration analysis of the common ions of the different drugs and ultimately screened 39 ions that were initially related to cardiotoxicity (Table 3), of which nine biomarkers were identified with mass spectrometry information (Table S2 and Figure S4).26 Of the substances that were screened in this study, several substances appear to be unknown. We will attempt to identify these unknown substances in future studies.

3. RESULTS AND DISCUSSION 3.1. Biochemical Analysis and Histopathological Assessment

The detection and analysis of the enzymatic spectrum of serum are important to understand the extent of organ damage caused by toxic effects.24 To assess the toxic effects of drugs on the heart, the levels of LDH, CK, CK-MB, and MDA in the serum from the model group were compared with NS group by t-test. The biochemical results of our experiments are summarized in Figure 1. Compared with the NS group, the levels of the 1 d post-DOX, 3 d post-ISO, and 3 d post-5-FU treatment groups significantly increased (p < 0.05); the serum levels of the other groups did not significantly change. The significant changes (1 d post-DOX, 3 d post-ISO, and 3 d post-5-FU treatments) indicated cardiotoxicity in the organism. The histopathological results demonstrated that compared with the NS group, the 1 d post-DOX, 3 d post-ISO, and 3 d post-5-FU administration groups exhibited heart tissue injury, which can manifest as myocardial disarrangement, disorders, degeneration and necrosis, and cell invasion. The cardiac tissue of the other groups did not exhibit damage (Figure 2). The histopathological results of the CCl4 , Radix Bupleuri, gentamicin, and etimicin groups at the verification and optimization stages are shown in Figure S1 of the Supporting Information. The results of the biochemical tests and histopathological examinations indicated that the bodies were exposed to

3.3. Verification and Optimization of Biomarkers for the Early Prediction of Cardiotoxicity

Different toxic drugs may induce similar metabolic processes in vivo and also cause changes in the same biomarkers. Because the liver and kidney are important metabolic and excretory organs, biomarkers for the early prediction of cardiotoxicity were verified by the results of metabolomics research on hepatotoxicity and nephrotoxicity to identify biomarkers with high specificity. We obtained unified biomarkers for the early prediction of cardiotoxicity caused by three drugs. The results of 10 biomarker ions for the early prediction of cardiotoxicity have different content variance in hepatotoxicity and nephrotoxicity groups, which compared with cardiotoxicity groups (Table 4).

Figure 2. Effects of 6 h/12 h/1 d post-DOX administration (B1−B3), 12 h/1 d/3 d post-ISO administration (C1−C3), and 1 d/2 d/3 d post-5-FU administration (D1−D3) for the heart tissue were assessed by histopathology compared with NS administration (A). D

DOI: 10.1021/pr501116c J. Proteome Res. XXXX, XXX, XXX−XXX

Article

Journal of Proteome Research

Figure 3. PLS-DA score plot (A, B, and C) of DOX groups, ISO groups, and 5-FU groups compared with NS groups, respectively. R2X, R2Y, and Q2 were used to evaluate whether the PLS-DA model is satisfactory. The closer to 1 the R2Y/Q2 ratio of PLS-DA model is, the more stable and reliable the model is. (A) R2X = 0.435, R2Y = 1, Q2 = 0.999; (B) R2X = 0.387, R2Y = 1, Q2 = 0.999; (C) R2X = 0.393, R2Y = 1, Q2 = 0.999.

exhibited good predictive ability and provides a strong motivation for additional testing of relevant samples. 3.4. Biological Significance of Biomarkers for the Early Prediction of Cardiotoxicity

We established three models of different cardiotoxicity mechanisms and screened the common metabolites, that is, the early cardiotoxicity biomarkers. The screened early cardiotoxicity biomarkers can reflect generally the biochemical mechanisms of cardiotoxicity. According to KEGG, the common markers that are related to cardiotoxicity are caused by different pathways. To further illustrate this finding, the mechanisms (mainly phospholipid metabolism and energy metabolism) are summarized in Figure 6. LPCs are a type of endogenous phospholipid and are classified as such. The drug can have damaging effects on the physical structure of the cell membrane.28 LPCs also play an important role in pathological changes in myocardial tissue.19 During drug-induced cardiotoxicity, the protein kinase C (PKC) pathway is activated, and phosphatase A2 activity is enhanced; thus, LPC generation decreases.29 Lysophospholipid receptors (LPL-Rs) are members of the G protein-coupled receptor family and perform messenger functions in life activities. Some drugs can activate these receptors and send signals to activate lysophospholipids, which can affect the normal function of cardiomyocytes. There is a close link between energy metabolism and cardiac function. When pathological changes occurred in heart function, energy metabolism also exhibited abnormalities. Fatty acids provide energy for cardiometabolism. The fatty acid β-oxidation pathway plays an essential role in myocardial mitochondrial function.30 Cardiotoxicity can lead to increases in myocardial oxygen consumption, which may further aggravate the β-oxidation of fatty acids. L-Carnitine is vital for the uptake of fatty acids as energy decreases and may weaken the mitochondrial dysfunction of cardiocytes.10,19 Amino acid metabolism, which is an important basic pathway of life, plays an important role in energy metabolism and the biological conversion process. For example, L-tryptophan is a precursor material of energy metabolism and may be converted to a substance in the metabolic Krebs cycle.19 These changes occur in the organism to meet the energy requirements of heart damage.

Figure 4. Venn diagram of DOX groups, ISO groups, and 5-FU groups that separately integrated the ion information at different time points and, respectively, found 106, 77, and 69 common ions. For integration analysis of different drugs’ ion information, we screened 39 ions that were initially related to cardiotoxicity in the early stage.

The classified model was established by the 10 highly specific biomarkers that compared with the noncardiotoxicity models. The peak area of two-thirds of the samples was randomly divided into the training set, which was used to build the classified model, and the peak area of the other one-third was divided into the test set, which was used to determine the accuracy rate of the model in the SVM classification process.27 By removing the potential biomarker metabolomics data, prediction models with good accuracy were established. The prediction accuracy rate of the model established by the potential biomarkers for the early prediction of cardiotoxicity was 90.0%. The Bestc, Bestg, and CV accuracy parameters from the cross-validation method are shown in Figure 5. Some differences in the prediction accuracy can be identified when several potential biomarkers are removed, as shown in Figure S5. The model prediction accuracy rate decreased when Lcarnitine, 19-hydroxydeoxycorticosterone, LPC (14:0), and LPC (20:2) were removed, which indicates that these substances have positive impacts on the model and strong specificity for the early prediction of cardiotoxicity. The model

3.5. Significance of Biomarkers for the Early Prediction of Cardiotoxicity

Cardiotoxicity can lead to functional changes in the organism, such as palpitations and arrhythmia, and may further develop E

DOI: 10.1021/pr501116c J. Proteome Res. XXXX, XXX, XXX−XXX

Article

Journal of Proteome Research

Table 3. Thirty-nine Potential Biomarkers’ Ions for Early Prediction of Cardiotoxicity in the Positive Ion Mode by Multivariate Statistical Analysis and Integration Analysis of DOX Group, ISO Group, and 5-FU Group no.

tRc (min)

obsd m/z

calcd m/z

error (ppm)

1 2

0.71 0.73

97.0289 162.1135

162.1130

−3.08

3 4 5

0.75 0.75 2.12

140.0695 156.0441 205.0971

205.0977

2.93

L-tryptophan

6 7 8 9 10

2.91 2.91 2.93 3.05 3.16

172.0758 190.0866 130.0653 360.2724 347.2210

347.2222

3.46

11 12

3.23 3.60

237.1480 431.2765

431.2773

13 14 15 16 17 18

3.80 3.83 3.84 3.87 4.04 4.64

432.3109 391.2846 298.2725 287.1487 299.1999 490.2904

19 20 21 22

5.09 5.21 5.26 5.27

23

metabolite a

formula

MS/MS

C7H15NO3

162.1 [M + H]+ 103.0 [M+H−C3H9N]+

C11H12N2O2

205.0 [M + H]+ 188.0 [M+H-NH3]+

19-hydroxydeoxycorticosteroneb

C21H30O4

385.1 [M+K]+ 369.2 [M + Na]+ 347.2 [M + H]+ 329.2 [M+H−H2O]+ 109.1 [M+H−C15H22O3]+ 97.1 [M+H−C14H22O3]+

1.85

cholic acida

C24H40O5

431.2 [M + Na]+ 391.2 [M+H−H2O]+ 373.2 [M+H-2H2O]+

490.2910

1.22

LPC (14:0)b

C22H46NO7P

490.2 468.3 184.0 104.1

346.3277 412.3321 502.2947 590.3216

590.3223

1.19

LPC (22:6)b

C30H50NO7P

5.44

400.3418

400.3427

2.25

L-palmitoylcarnitine

590.3 [M + Na]+ 568.3 [M + H]+ 184.0 [M+H−C25H36O3]+ 104.1 [M+H−C25H37O6P]+ 400.3 [M + H]+ 341.2 [M+H−C3H9N]+ 144.1 [M+H−C16H32O2]+ 85.0 [M+H−C19H41NO2]+

24 25 26

5.57 5.63 5.65

468.3456 426.3576 496.3402

496.3403

0.20

LPC (16:0)b

C24H50NO7P

518.3 496.3 184.0 104.1

[M + Na]+ [M + H]+ [M+H- C19H36O3]+ [M+H−C19H33O6P]+

27 28 29 30 31

6.06 6.17 6.18 6.19 6.46

183.1171 544.3379 438.2984 480.3399 548.3712

548.3716

0.73

LPC (20:2)b

C28H54NO7P

570.3 548.3 184.0 104.1

[M + Na]+ [M + H]+ [M+H−C23H40O3]+ [M+H−C23H41O6P]+

32 33 34 35 36

6.49 6.93 7.39 7.55 7.65

508.3761 980.6017 551.3867 466.3298 510.3924

L-carnitine

F

a

a

C23H45NO4

[M + Na]+ [M + H]+ [M+H−C17H32O3]+ [M+H−C17H33O6P]+

DOI: 10.1021/pr501116c J. Proteome Res. XXXX, XXX, XXX−XXX

Article

Journal of Proteome Research Table 3. continued no.

tRc (min)

obsd m/z

37 38 39

8.27 8.28 8.95

287.2370 207.1376 516.3434

calcd m/z

error (ppm)

metabolite

formula

MS/MS

a Metabolites were identified by comparison with standards. bMetabolites were putatively identified by parsing MS and comparing with databases and literature. ctR refers to retention time.

Table 4. Specificity Study Results of Biomarkers’ Ions for Early Prediction of Cardiotoxicitya cardiotoxicity tR (min)

m/z

0.73 2.91 2.91 2.93 3.16 3.80 4.04 4.64 5.57 6.46

162.1135 190.0866 172.0758 130.0653 347.221 432.3109 299.1999 490.2904 468.3456 548.3712

metabolite

nephrotoxicity

hepatotoxicity

ISO

5-FU

gentamicin

etimicin

Radix Bupleuri

CCl4

↑ ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓

↑ ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓

↓      ↑   ↓

↓     ↑ ↑   

↓  ↑ ↑    ↑ ↓ 

  ↑ ↑    ↑  

L-carnitine

19-hydroxydeoxycorticosterone

LPC (14:0) LPC (20:2)

“↑” means the content in the model group was significantly increased compared with the NS group (p < 0.05, n = 10). “↓” means the content in the model group was significantly decreased compared with the NS group (p < 0.05, n = 10). “” means the content in the model group was not significantly changed compared with the NS group (p > 0.05, n = 10). a

with conventional biochemical detection indices. Biomarkers for the early prediction of cardiotoxicity that are obtained through screening, verification, and optimization can alert medical practitioners to the cardiotoxic effects of drugs even in the early stages of toxicity. Previous studies have focused on screening biomarkers of cardiotoxicity of single drugs but have not actively promoted cardiotoxicity studies or the application of biomarkers of cardiotoxicity.31−33 In this study, we established three cardiotoxicity models and screened the metabolic biomarkers of each group. The biomarkers we identified are more sensitive than the existing detection indicators and previously reported biomarkers of cardiotoxicity; thus, our study further promotes the application of metabolomics in the determination of cardiotoxicity. The use of SVM in metabolomics enables broader applications of metabolic biomarkers. By using the advantages of the second classification model, which is based on the elimination of single biomarkers, we calculated the prediction rate of the SVM model and determined its contribution to the toxicity prediction model to optimize biomarkers. Biomarkers for the early prediction of cardiotoxicity that are optimized by SVM can be used to screen the cardiotoxicity of drugs or biological samples. The combination of metabolomics with the SVM model can provide a unique perspective to screen and apply metabolic biomarkers. Compared with conventional biochemical analyses and histopathological assessments, biomarkers for the early prediction of cardiotoxicity based on metabolomics are advantageous in drug safety evaluations. The results demonstrate that these biomarkers are rapid, have strong specificity and high accuracy, and are less invasive; thus, they have broad application prospects. In the drug development process, this approach can provide further cost savings and shorten the development cycle. In addition, it may also provide a basis for

Figure 5. Three-dimensional view of SVM model of 10 biomarkers (the parameters are described below: Bestc = 1.3195, Bestg = 6.9644, CV accuracy = 91.3793%).

into substantial changes such as myocardial disarrangement, cell degeneration and necrosis. Therefore, the development of early cardiotoxicity markers is important for the identification of toxicity before cardiac tissues become pathologically damaged. Myocardial enzyme detection methods are often applied in heart attacks as a secondary biochemical analysis index; however, this index often identifies significant changes only after pathological damage has occurred in the heart tissue. This finding indicates that the determination of toxicity exposure is relatively delayed in existing methods. In addition, the sensitivity of related enzymes is not ideal, and their specificity is poor. By comparison, biomarkers for the early prediction of cardiotoxicity may indicate significant changes prior to heart tissue damage, that is, toxicity is detected faster using biomarkers for the early prediction of cardiotoxicity compared G

DOI: 10.1021/pr501116c J. Proteome Res. XXXX, XXX, XXX−XXX

Article

Journal of Proteome Research

Figure 6. Schematic diagram about biological significance of biomarkers for early prediction of cardiotoxicity. (□ means that the biomarkers were obtained in this experiment).

■

safety evaluations and the secondary development and preparation of traditional Chinese medicines.

Corresponding Author

*E-mail: [email protected]. Phone: +86-22-59596221 Fax: +86-22-59596221.

4. CONCLUSION This experiment identified exclusive biomarkers for the early prediction of cardiotoxicity using UPLC−Q-TOF-MS-based metabolomics. We used multivariate statistical and integration analyses to screen 39 unified biomarkers of drug-induced cardiotoxicity. SVM was subsequently applied as a pattern recognition method and combined with the metabolomics results of hepatotoxicity and nephrotoxicity to verify and optimize the screened biomarkers of cardiotoxicity. We identified 10 exclusive biomarkers for the early prediction of cardiotoxicity. The prediction rates of the SVM model for these biomarkers were as high as 90.0%. Of the biomarkers, Lcarnitine, 19-hydroxydeoxycorticosterone, LPC (14:0), and LPC (20:2) exhibited the strongest specificities. This method exhibited high sensitivity and specificity and would serve as a basis research for further diagnosis and prediction of druginduced cardiotoxicity in its early stages. The proposed method can also provide a reliable basis for the evaluation of drug safety and secondary development.

■

AUTHOR INFORMATION

Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS This project was supported by the National Basic Research Program of China (973 Program) (2011CB505300, 2011CB505302) and the National Natural Science foundation of China (No. 81273998).

■

REFERENCES

(1) Gao, Y.; Ma, Z.; Zhang, B. Significance of re-evaluation and development of Chinese herbal drugs. China J. Chin. Mater. Med. 2012, 37 (1), 1−4. (2) Sandhu, H.; Maddock, H. Molecular basis of cancer-therapyinduced cardiotoxicity: Introducing microRNA biomarkers for early assessment of subclinical myocardial injury. Clin Sci. 2014, 126, 377− 400. (3) Mori, Y.; Kondo, C.; Tonomura, Y.; Torii, M.; Uehara, T. Identification of potential genomic biomarkers for early detection of chemically induced cardiotoxicity in rats. Toxicology 2010, 271 (1−2), 36−44. (4) Lestuzzi, C.; Vaccher, E.; Talamini, R.; Lleshi, A.; Meneguzzo, N.; Viel, E.; Scalone, S.; Tartuferi, L.; Buonadonna, A.; Ejiofor, L.; Schmoll, H. J. Effort myocardial ischemia during chemotherapy with 5fluorouracil: An underestimated risk. Ann. Oncol. 2014, 25 (5), 1059− 1064. (5) Berna, M. J.; Z, Y.; Watson, D. E.; Hale, J. E.; Ackermann, B. L. Strategic use of immunoprecipitation and LC−MS/MS for trace-level protein quantification: Myosin light chain 1, a biomarker of cardiac necrosis. Anal. Chem. 2007, 79, 4199−4205. (6) Babuin, L.; Jaffe, A. S. Troponin: The biomarker of choice for the detection of cardiac injury. Can. Med. Assoc. J. 2005, 173 (10), 1191− 1202. (7) Tonomura, Y.; Mori, Y.; Torii, M.; Uehara, T. Evaluation of the usefulness of biomarkers for cardiac and skeletal myotoxicity in rats. Toxicology 2009, 266 (1−3), 48−54.

ASSOCIATED CONTENT

S Supporting Information *

Effects of the heart, liver, and kidney tissue were assessed by histopathology compared with corresponding NS group. BPI chromatogram of QC plasma samples in positive ion mode. PCA score plot of the three cardiotoxicity drug groups. Mass spectrum of standards: L-carnitine (tR = 0.73 min), Ltryptophan (tR =2.12 min), cholic acid (tR = 3.60 min), and L-palmitoylcarnitine (tR = 5.44 min). Prediction results of the ten specific biomarkers for early prediction of cardiotoxicity by SVM. The results of experimental methodology. Fragmentation mechanisms and generated fragmentation of the nine potential biomarkers for early prediction of cardiotoxicity in positive ion mode. The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/pr501116c. H

DOI: 10.1021/pr501116c J. Proteome Res. XXXX, XXX, XXX−XXX

Article

Journal of Proteome Research (8) Cong, W.; Liang, Q.; Li, L.; Shi, J.; Liu, Q.; Feng, Y.; Wang, Y.; Luo, G. Metabonomic study on the cumulative cardiotoxicity of a pirarubicin liposome powder. Talanta 2012, 89, 91−98. (9) Zhang, H. Y.; Chen, X.; Hu, P.; Liang, Q. L.; Liang, X. P.; Wang, Y. M.; Luo, G. A. Metabolomic profiling of rat serum associated with isoproterenol-induced myocardial infarction using ultra-performance liquid chromatography/time-of-flight mass spectrometry and multivariate analysis. Talanta 2009, 79 (2), 254−259. (10) Liu, Y. T.; Jia, H. M.; Chang, X.; Ding, G.; Zhang, H. W.; Zou, Z. M. The metabolic disturbances of isoproterenol induced myocardial infarction in rats based on a tissue targeted metabonomics. Mol. Biosyst. 2013, 9 (11), 2823−2834. (11) Ramirez, T.; Daneshian, M.; Kamp, H.; Bois, F. Y.; Clench, M. R.; Coen, M.; Donley, B.; Fischer, S. M.; Ekman, D. R.; Fabian, E.; Guillou, C.; Heuer, J.; Hogberg, H. T.; Jungnickel, H.; Keun, H. C.; Krennrich, G.; Krupp, E.; Luch, A.; Noor, F.; Peter, E.; Riefke, B.; Seymour, M.; Skinner, N.; Smirnova, L.; Verheij, E.; Wagner, S.; Hartung, T.; van Ravenzwaay, B.; Leist, M. Metabolomics in toxicology and preclinical research. Altex Chromatogram 2013, 30 (2), 209−225. (12) Andreadou, I.; Papaefthimiou, M.; Zira, A.; Constantinou, M.; Sigala, F.; Skaltsounis, A. L.; Tsantili-Kakoulidou, A.; Iliodromitis, E. K.; Kremastinos, D. T.; Mikros, E. Metabonomic identification of novel biomarkers in doxorubicin cardiotoxicity and protective effect of the natural antioxidant oleuropein. NMR Biomed. 2009, 22 (6), 585− 592. (13) Dong, H.; Zhang, A.; Sun, H.; Wang, H.; Lu, X.; Wang, M.; Ni, B.; Wang, X. Ingenuity pathways analysis of urine metabolomics phenotypes toxicity of Chuanwu in Wistar rats by UPLC−Q-TOFHDMS coupled with pattern recognition methods. Mol. Biosyst. 2012, 8 (4), 1206−1221. (14) Wei, R.; Li, G.; Seymour, A. B. High-throughput and multiplexed LC−MS/MRM method for targeted metabolomics. Anal. Chem. 2010, 82 (13), 5527−5533. (15) Valentini, G. Gene expression data analysis of human lymphoma using support vector machines and output coding ensembles. Artif. Intell. Med. 2002, 26 (3), 281−304. (16) Liu, Y. H.; Chen, Y. T. Face recognition using total marginbased adaptive fuzzy support vector machines. IEEE Trans. Neural Networks 2007, 18, 178−192. (17) Manivannan, K.; Aggarwal, P.; Devabhaktuni, V.; Kumar, A.; Nims, D.; Bhattacharya, P. Particulate matter characterization by gray level co-occurrence matrix based support vector machines. J. Hazard Mater. 2012, 223−224, 94−103. (18) Johnston, B. A.; Mwangi, B.; Matthews, K.; Coghill, D.; Konrad, K.; Steele, J. D. Brainstem abnormalities in attention deficit hyperactivity disorder support high accuracy individual diagnostic classification. Hum. Brain Mapp. 2014, 35, 5179−5189. (19) Liu, Y. T.; Jia, H. M.; Chang, X.; Cheng, W. H.; Zhao, X.; Ding, G.; Zhang, H. W.; Cai, D. Y.; Zou, Z. M. Metabolic pathways involved in Xin-Ke-Shu protecting against myocardial infarction in rats using ultra high-performance liquid chromatography coupled with quadrupole time-of-flight mass spectrometry. J. Pharm. Biomed. Anal. 2014, 90, 35−44. (20) Gika, H. G.; Theodoridis, G. A.; Wingate, J. E.; Wilson, I. D. Within-day reproducibility of an HPLC−MS-based method for metabonomic analysis: Application to human urine. J. Proteome Res. 2007, 6 (8), 3291−3303. (21) Xie, G. X.; Lu, L. G.; Qiu, Y. P.; Ni, Q. X.; Zhang, W.; Gao, Y. T.; Risch, H. A.; Yu, H.; Jia, W. Plasma metabolite biomarkers for the detection of pancreatic cancer. J. Proteome Res. 2015, 14, 1195−1202. (22) Luts, J.; Ojeda, F.; Van de Plas, R.; De Moor, B.; Van Huffel, S.; Suykens, J. A. A tutorial on support vector machine-based methods for classification problems in chemometrics. Anal. Chim. Acta 2010, 665 (2), 129−145. (23) Zavar, M. R. S.; Akbarzadeh, T. M. R. Evolutionary model selection in a wavelet-based support vector machine for automated seizure detection. Expert Syst. Appl. 2011, 38 (9), 10751−10758.

(24) Khalid, M. A.; Ashraf, M. Direct detection of endogenous hydroxyl radical production in cultured adult cardiomyocytes during anoxia and reoxygenation. Is the hydroxyl radical really the most damaging radical species? Circ. Res. 1993, 72 (4), 725−736. (25) Xie, G.; Zheng, X.; Qi, X.; Cao, Y.; Chi, Y.; Su, M.; Ni, Y.; Qiu, Y.; Liu, Y.; Li, H.; Zhao, A.; Jia, W. Metabonomic evaluation of melamine-induced acute renal toxicity in rats. J. Proteome Res. 2010, 9 (1), 125−133. (26) Li, Y. B.; Zhang, Z. Z.; Hou, Z. G.; Wang, L.; Wu, X.; Ju, L.; Zhang, X. X.; Zhang, Y. J. A rapid and integrated pyramid screening method to classify and identify complex endogenous substances with UPLC−Q-TOF-MS-based metabolomics. RSC Adv. 2015, 5, 202− 209. (27) Tan, Y. X.; Yin, P. Y.; Tang, L.; Xing, W. B.; Huang, Q.; Cao, D.; Zhao, X. J.; Wang, W. Z.; Lu, X.; Wang, H. Y.; Xu, G. W. Metabolomics study of stepwise hepatocarcinogenesis from the model rats to patients: Potential biomarkers effective for small hepatocellular carcinoma diagnosis. Mol. Cell. Proteomics 2012, 11 (2), M111.010694. (28) Xu, Y.; Xiao, Y. J.; Zhu, K.; Baudhuin, L. M.; Lu, J.; Hong, G.; Kim, K. S.; Cristina, K. L.; Song, L.; Williams, F. S.; Elson, P.; Markman, M.; Belinson, J. Unfolding the pathophysiological role of bioactive lysophospholipids. Curr. Drug Targets: Immune, Endocr. Metab. Disord. 2003, 3, 23−32. (29) Fuly, A. L.; Machado, A. L.; Castro, P. Lysophosphatidylcholine produced by the phospholipase A2 isolated from Lachesis muta snake venom modulates natural killer activity as a protein kinase C effector. Toxicon 2007, 50, 400−410. (30) Lena, E. H.; Ake, N. Arahidonic acid and ischemic heart disease. J. Nutr. 2005, 135 (9), 2271−2273. (31) Li, Y. B.; Zhang, X. X.; Zhou, H. F.; Fan, S. M.; Wang, Y. M.; Wang, L.; Zhang, Z. Z.; Deng, H. Y.; Zhang, Y. J. Toxicity analysis of doxorbic using plasma metabolomics technology based on rapid resolution liquid chromatography coupled with quadruple-time-offlight mass spectrometry. Anal. Methods 2014, 6, 5909−5917. (32) Andreadou, I.; Mikros, E.; Ioannidis, K.; Sigala, F.; Naka, K.; Kostidis, S.; Farmakis, D.; Tenta, R.; Kavantzas, N.; Bibli, S. I.; Gikas, E.; Skaltsounis, L.; Kremastinos, D. T.; Iliodromitis, E. K. Oleuropein prevents doxorubicin-induced cardiomyopathy interfering with signaling molecules and cardiomyocyte metabolism. J. Mol. Cell. Cardiol. 2014, 69, 4−16. (33) Guang, G.. T.; Zi, Y. L.; Wen, T. L.; Zhen, Y. Z.; Xin, D.; Wei, Z.; Wu, H. L.; Yi, F. C. Potential biomarkers in mouse myocardium of doxorubicin-induced cardiomyopathy: A metabonomic method and its application. PLoS One 2011, 6 (11), e27683.

I

DOI: 10.1021/pr501116c J. Proteome Res. XXXX, XXX, XXX−XXX