Journal of Pharmaceutical and Biomedical Analysis 100 (2014) 369–380

Contents lists available at ScienceDirect

Journal of Pharmaceutical and Biomedical Analysis journal homepage: www.elsevier.com/locate/jpba

Metabolomics provide new insights on lung cancer staging and discrimination from chronic obstructive pulmonary disease Stanislaw Deja a , Irena Porebska b , Aneta Kowal b , Adam Zabek c , Wojciech Barg d , Konrad Pawelczyk e , Ivana Stanimirova f , Michal Daszykowski f , Anna Korzeniewska b , Renata Jankowska b , Piotr Mlynarz c,∗ a

Faculty of Chemistry, Opole University, Pl. Kopernika 11a, 45-040 Opole, Poland Department and Clinic of Pulmonology and Lung Cancers, Wroclaw Medical University, Grabiszynska 105, 53-439 Wroclaw, Poland c Department of Bioorganic Chemistry Wrocław University of Technology, Wybrzeze Wyspianskiego 27, 50-370 Wroclaw, Poland d Department of Physiology, Wroclaw Medical University, T. Chalubinskiego 10, 50-368 Wroclaw, Poland e Department and Clinic of Thoracic Surgery, Wroclaw Medical University, Grabiszynska 105, 53-430 Wroclaw, Poland f Institute of Chemistry, The University of Silesia, Szkolna 9, 40-006 Katowice, Poland b

a r t i c l e

i n f o

Article history: Received 22 April 2014 Accepted 13 August 2014 Available online 21 August 2014 Keywords: COPD—chronic obstructive pulmonary disease Lung cancer Metabolomics 1 H NMR spectroscopy Metabolic fingerprinting

a b s t r a c t Chronic obstructive pulmonary disease (COPD) and lung cancer are widespread lung diseases. Cigarette smoking is a high risk factor for both the diseases. COPD may increase the risk of developing lung cancer. Thus, it is crucial to be able to distinguish between these two pathological states, especially considering the early stages of lung cancer. Novel diagnostic and monitoring tools are required to properly determine lung cancer progression because this information directly impacts the type of the treatment prescribed. In this study, serum samples collected from 22 COPD and 77 lung cancer (TNM stages I, II, III, and IV) patients were analyzed. Then, a collection of NMR metabolic fingerprints was modeled using discriminant orthogonal partial least squares regression (OPLS-DA) and further interpreted by univariate statistics. The constructed discriminant models helped to successfully distinguish between the metabolic fingerprints of COPD and lung cancer patients (AUC training = 0.972, AUC test = 0.993), COPD and early lung cancer patients (AUC training = 1.000, AUC test = 1.000), and COPD and advanced lung cancer patients (AUC training = 0.983, AUC test = 1.000). Decreased acetate, citrate, and methanol levels together with the increased N-acetylated glycoproteins, leucine, lysine, mannose, choline, and lipid (CH3 (CH2 )n ) levels were observed in all lung cancer patients compared with the COPD group. The evaluation of lung cancer progression was also successful using OPLS-DA (AUC training = 0.811, AUC test = 0.904). Based on the results, the following metabolite biomarkers may prove useful in distinguishing lung cancer states: isoleucine, acetoacetate, and creatine as well as the two NMR signals of N-acetylated glycoproteins and glycerol. © 2014 Elsevier B.V. All rights reserved.

1. Introduction Lung cancer diagnostics, treatment, and aftercare monitoring remain a great challenge in oncology. This highly preventable disease is the most frequent cancer worldwide (e.g., 1.38 million people diagnosed with lung cancer died in 2008) [1]. Despite the progress made in molecular oncology, lung cancer prognosis remains poor (15% five-year survival rate) [2], underscoring

∗ Correspondence to: Bioorganic Group, Department of Chemistry, Wroclaw Uni˙ Wyspianskiego 27, 55-093 Wroclaw, Poland. versity of Technology, Wybrzeze Fax: +48 713204597. E-mail address: [email protected] (P. Mlynarz). http://dx.doi.org/10.1016/j.jpba.2014.08.020 0731-7085/© 2014 Elsevier B.V. All rights reserved.

the need for new treatment regimens based on highly specific biological therapies and further development of diagnostic tools. The unfavorable prognosis of lung cancer is attributed to the following observations: early stage lung cancer is asymptomatic and clinical symptoms typically appear in the late stages of this disease. Usually, patients with lung cancer are in advanced age and concomitant diseases limit available method of treatment. Positive screening results based on low-dose computed tomography (LDCT) offers hope for improving this situation, but this procedure is still not a routine clinical practice [3]. The important problem related to LDCT results is overdiagnosis [4]. The discrimination between more or less aggressive as well as benign or malignant tumour is still not well established in LDCT [5].

370

S. Deja et al. / Journal of Pharmaceutical and Biomedical Analysis 100 (2014) 369–380

The need for a simple, cheap, and readily available screening test that could serve as a stand-alone screening function or improve the specificity of low-dose computed tomography remains. The most common type of lung cancer is non-small cell lung cancer (NSCLC), which accounts for approximately 85% of all lung cancer cases, followed by small cell lung cancer (SCLC), which accounts for approximately 15% of cases [6–8]. Other types of lung cancer are relatively rare. The tumor node metastasis (TNM) staging system is widely accepted for NSCLC classification. Patients are classified according to four stages (I to IV). However, marked heterogeneity among patients, especially in the third stage, is observed in clinical practice. Patients within group III can be further divided into two sub-groups IIIA and IIIB; however, considerable controversy remains regarding this subdivision, which is primarily related to the treatment of N2 (IIIA) and T4 (IIIB) patients [9]. Combined treatment, including surgery and chemotherapy or chemoradiotherapy, produces unsatisfactory long-term survival for stage III. In stage IV patients, the results of chemotherapy and target treatment are still disappointing. Adverse events resulting from combined treatment are common, thereby explaining why some patients are unable to complete the planned therapy. This issue is of particular concern in stage IV patients. Some of the patients rapidly decline, whereas others on the same treatment regimen exhibit longer survival rates. Therefore, proper classification of lung cancer patients is important to prescribe the appropriate type of treatment and new predictive markers are expected. In addition to lung cancer, chronic obstructive pulmonary disease (COPD) is a common lung disease. Cigarette smoking is a common risk factor for NSCLC and COPD. COPD is thought to increase the risk of lung cancer [10] and is even considered an independent risk factor [11]. Patients diagnosed with lung cancer often experience COPD symptoms [12]. Based on increasing data in the literature, inflammation is thought to be involved in the pathogenesis of both diseases [13–15]. Chronic inflammation is a potential factor associated with tumor development. The following examples typify the evolution of chronic inflammation to cancer: infection (e.g., Helicobacter pylori in gastric carcinoma), chronic immune-mediated inflammatory disorders (e.g., inflammatory bowel disease in colorectal cancer) as well as the development of lung cancer from COPD [16]. Therefore, COPD may increase the risk of developing lung cancer from chronic inflammation regardless of cigarette smoking. However, recent studies have implicated smoking as the primary risk factor for lung cancer despite the fact that COPD is strongly associated with lung cancer [17]. Cigarette smoke induces not only inflammation, but also contains genotoxic factors. All these elements together result in accumulation of genetic errors, which are essential for lung cancer development. Recently recommended prevention strategies highlight the importance of smoking cessation in the prevention of COPD and lung cancer. Secondarily, these measures seek to prevent the development of lung cancer in COPD patients [16]. Therefore, a chemically measurable indicator that differentiates between these two states is needed. Both pathological states should be reflected in qualitative and quantitative molecular changes in the substrate and products of disrupted biochemical pathways at the cellular level. Thus, these disease hallmarks should be reflected in the chemical composition of body fluids. Data-rich analytical techniques, such as nuclear magnetic resonance (NMR) spectroscopy and mass spectrometry (MS), have the potential to generate a highly informative metabolic fingerprint during single measurements. Instead of relying on one biomarker for classification, it is possible to combine two or more biomarkers to enhance classification power. The combination allows for more accurate diagnoses, reflecting the biochemical pathways perturbed by pathogenesis. A multi-biomarker approach

enables to distinguish between similar diseases [18]. Therefore, metabolomic-based studies have been extensively used for disease recognition, prognosis, predicting response to treatment and recovery monitoring. Recently, numerous COPD and lung cancer studies have revealed the great potential of metabolomic-based approaches in disease diagnostics and stratification [19–25]. Moreover this molecular technique, based on laboratory analysis of small blood sample, is minimally invasive for patient. Considering the above aspects of NSCLC, the primary goal of our study was to perform a metabolomic analysis and generate metabolic fingerprints of NSCLC patients from various well-defined disease stages and to compare them with COPD patients. The metabolic data were used to construct discriminant models to aid in the differentiation of COPD patients from NSCLC patients at various stages. In addition, we sought to identify potential metabolic biomarkers related to these differences. The development of diagnostic models may be potentially useful in a clinical setting, providing valuable information concerning patient diagnosis, stratification and monitoring.

2. Materials and methods 2.1. Clinical population In total, 77 non-small cell lung cancer (NSCLC) and 22 COPD patients were included in this study. For this study, the term “lung cancer” refers to non-small cell histological types. These patients were hospitalized in the Department of Pulmonology and Lung Cancer or Thoracic Surgery at Wroclaw University of Medicine from August 2011 to March 2013. A COPD diagnosis was made according to the Global Initiative for Chronic Obstructive Lung Disease (GOLD) [26], which considers clinical symptoms and the spirometry test. Lung cancer diagnoses were made on the basis of pathological exams of tissue specimens obtained either from diagnostic bronchoscopy or curative surgery. NSCLC patients were classified according to the VII edition of the International Association for the Study of Lung Cancer (IASLC) TNM system [27]. The stage of patients undergoing therapeutic surgery was pathological, and the remaining cases were clinical. NSCLC patient characteristics are presented in Table 1. All of the COPD patients were in a stabilization period without symptom exacerbation. Lung cancer patient sera were collected before disease treatment (surgery, radio- or chemotherapy), whereas the pharmacological treatment of the coexisting diseases was continued. During sample collection, COPD patients were treated according to disease stage and concomitant diseases. The study protocol was approved by the ethics committee of Wroclaw University of Medicine, and all of the patients provided written informed consent (KB-12/2010 and KB-263/2013).

2.2. NMR spectroscopy sample preparation Blood samples (9 mL each) were collected in the morning and 12 h after the last meal and centrifuged at 4000 rpm for 10 min at 4 ◦ C. The serum samples were then immediately frozen and stored at −80 ◦ C until further analysis. Prior to the metabolomic experiments, the serum samples were thawed at room temperature and vortexed. Next, the mixtures containing 200 ␮L of serum and 400 ␮L of saline solution (0.9% NaCl in 15% D2 O) were mixed again [28]. After centrifugation (15000 rpm for 10 min), a 550-␮L aliquot of each sample supernatant was subsequently transferred to a 5-mm NMR tube. Samples were maintained at 4 ◦ C before measurement.

S. Deja et al. / Journal of Pharmaceutical and Biomedical Analysis 100 (2014) 369–380

371

Table 1 The characteristics of COPD and NSCLC patients participating in the study. Training data set

Total number Gender Female Male Mean age [range] Smoking habits Never Current Former Unknown Histological type Squamous cell Adenocarcinoma Adenosquamous Large cell Unknown* Stage I II III IV

Test data set

COPD

NSCLC (with COPD)

COPD

NSCLC (with COPD)

12

48(19)

10

29(10)

7 5 61[39–78]

11(4) 37(15) 63[50–78] (62[50–75])

2 8 63[52–81]

5(0) 24(10) 65[51–80] (65[51–80])

1 3 4 4

3(0) 29(13) 13(4) 3(2)

1 2 5 2

1 20(10) 7 1

N/A N/A N/A N/A N/A

20(10) 20(8) 1 1 6(1)

N/A N/A N/A N/A N/A

11(4) 7(2) 0 3(2) 8(2)

N/A N/A N/A N/A

12 (8) 12 (4) 12 (4) 12 (3)

N/A N/A N/A N/A

5 (3) 6 (2) 9 (2) 9 (3)

* Unknown non-small cell lung cancer subtype refers to small tissue samples obtained from various patients with advanced cancer exclusively undergoing bronchoscopy. To provide a better comparison between COPD and NSCLC the data set is composed of 12 COPD samples and 24 NSCLC samples (6 from each cancer stage form I to IV). N/A data not available.

2.3.

1H

NMR measurements

NMR spectra were recorded at 300 K using an Avance II spectrometer (Bruker, GmbH, Germany) operating at proton frequency of 600.58 MHz. A one-dimensional Carr–Purcell–Meiboom–Gill (CPMG) NMR spin echo pulse sequence with water suppression was employed to filter out broad spectral resonances from macromolecules and thus better visualize low molecular weight metabolites. For each sample, 128 consecutive scans were collected with a 400-␮s spin-echo delay, 80 loops, a 3.5-s relaxation delay, 64 K TD, and 20.01 ppm SW. The spectra were processed with 0.3 Hz line broadening, manually phased and baseline corrected using Topspin 1.3 software (Bruker, GmbH, Germany), and referenced to an ␣-glucose signal (ı = 5.225 ppm). Serum samples are typically characterized by small variations in signal chemical shifts; however, an alignment is usually required, particularly when applying the fingerprint data analysis approach [29]. For this reason, a correction of peak positions was performed using the icoshift algorithm [30] implemented in Matlab (Matlab v. 8.1, Mathworks, Inc.). Finally, the dataset was binned into 8633 integrals of equal width (0.001 ppm).

differences. OPLS-DA divides the systematic variation in the NMR data matrix into predictive components, which model covariance between variables and responses, and orthogonal components, which collect variation unrelated to responses. This approach simplifies data interpretation and allows for the selection of relevant variables [32]. 2.5. Statistical analysis A set of quantified metabolites (with regard to signal integrals) was evaluated for statistical importance with nonparametric Kruskal–Wallis one-way ANOVA (KW-ANOVA) and Mann–Whitney–Wilcoxon (MWW) tests using the STATISTICA software (v 10, StatSoft, Tulsa, USA). Metabolites achieving the following criteria were identified as potential biomarkers for NSCLC staging: a rank correlation coefficient (Spearman) value between the NSCLC stages (1, 2, 3, and 4), a metabolite relative integral >0.3, and a KW-ANOVA P-value < 0.05. Given that COPD may promote lung cancer development in certain cases, a second set of COPD calculations were included with a level number of 0 (0, 1, 2, 3, and 4). 2.6. Evaluation and validation

2.4. Multivariate data analysis Two samples were excluded from the data analysis due to additional unidentified resonances observed at 0.86, 1.17, and 1.52 ppm (see supplementary information Fig. S1) that likely resulted from medical treatment. These samples are not included in Table 1. A total of 99 serum 1 H NMR CPMG spectra corresponding to 77 NSCLC and 22 COPD patients were considered for multivariate analysis. Each spectrum consisted of 8633 data points after removing the residual water region (4.300–5.167 ppm). The spectra were normalized using the Probabilistic Quotient Normalization (PQN) method available in the Matlab environment [31] and Pareto scaling. The preprocessed data were then transferred to the SIMCA-P+ software (v 13.0, Umetrics, Umeå, Sweden), where orthogonal projections to latent structures-discriminant analysis (OPLS-DA), a supervised pattern recognition approach was used to construct predictive models and identify metabolite fingerprint

Receiver operating characteristic (ROC) curves and area under curve (AUC) values were used to evaluate the OPLS-DA prediction performance. A perfcurve function from the Matlab statistical toolbox (Matlab v. 8.1, Mathworks, Inc.) was adopted for calculations. Specificity and sensitivity were determined according to sample class prediction using the 7-fold cross-validated predicted values of the fitted response for observations in the model set and the predicted values of the modeled response for observations in the test set (from SIMCA-P+ software). 2.7. Metabolite set enrichment analysis To identify the most altered metabolic pathways, a set of significantly altered metabolites was used as the input for the Metabolite Set Enrichment Analysis (MSEA) [33]. The MSE analysis was performed with a free on-line tool (http://www. msea.ca/MSEA/faces/Home.jsp). Over representation analysis

372

S. Deja et al. / Journal of Pharmaceutical and Biomedical Analysis 100 (2014) 369–380

(ORA) was used for comprehensive screening of affected pathways. P-values and false discovery rates (FDR) are reported. 3. Results The mean 1 H NMR CPMG spectrum of 45 assigned metabolites in NSCLC patient sera is presented as an example in Fig. 1. The metabolite resonances were identified according to assignments published in the literature [34] and on-line databases (http://hmdb.ca). One advantage of the fingerprint approach is that it can create a diagnostic model even when comprehensive knowledge about the metabolic composition of analyzed samples is unavailable. However, the data must be of suitable quality. Therefore, after the application of the icoshift algorithm, the NMR spectra were carefully inspected to avoid artifacts. The additional application of a relatively small binning step (0.001 ppm) prevented a substantial reduction in spectral resolution and maintained complicated data structure. A discriminant model should provide satisfactory prediction power to be considered as a potential tool for clinical practice or the selection of biomarkers. Therefore, it is crucial to validate the methods, e.g., test the prediction performance of the model. Various model validation procedures that employ a training set of samples as well as an independent set of samples, referred to as the test set, are available. Multivariate models used in metabolomic studies, such as the commonly used PLS-DA and OPLS-DA, are typically described by the coefficient of determination (R2 ) and cross-validated R2 (Q2 ). The validation procedure typically involves the permutation test or a cross-validated analysis of variance (CV-ANOVA) that are calculated using the training set samples exclusively. However, the risk of model overfitting, i.e. excellent discrimination performance for the training set samples but poor recognition for the test set samples, should always be taken into consideration [35]. The best approach to evaluate a model’s predictive performance, and thus implemented in this study, is to use an independent dataset. In our study, a balanced training set of samples (an equal number of samples from each group) was determined. The training set consists of various representative samples from each group that encompass all possible sources of variation. The representative samples were found by applying the Kennard and Stone algorithm [36] to individual groups. The training set groups are described in Table 1. They contain 12 representative samples for COPD and each of NSCLC stage. Therefore, in E-NSCLC vs. COPD and A-NSCLC vs. COPD comparisons the number of samples was 24 vs. 12. However in NSCLC vs. COPD comparison, only half of test NSCLC samples was used (six from each cancer stage) to avoid unbalanced comparison (24 vs. 12 instead of 48 vs. 12). 3.1. Discrimination between NSCLC and COPD patients The COPD patient group was compared with the complete NSCLC patient group and displayed good separation in the OPLSDA score plot (Fig. 2a). This separation was also evident between COPD and two independent NSCLC group subsets, an early nonsmall cell lung cancer (E-NSCLC) subgroup composed of stages I and II and an advanced non-small cell lung cancer (A-NSCLC) subgroup composed of stages III and IV (Fig. 2 b and c). All OPLS-DA models (established using a training set) had Q2 values greater than 0.4 (Table 2), but a test set was adopted to evaluate prediction errors. ROC curves are the most common metric used to describe the performance of medical diagnostic tests for discriminant problems, e.g., healthy vs. diseased individuals [37–39] and are recommended for metabolomic biomarker discovery studies [40]. Based on the predicted response values, ROC curves were

plotted, and their corresponding AUC values were calculated for the training and test sets (see Fig. 2). For all three comparisons, AUC values were equal or close to 1, suggesting that the NMRbased fingerprints have the potential to be used to distinguish COPD and NSCLC patients even during the early stages of cancer development. For a more detailed analysis, four additional models were constructed to compare the metabolic fingerprints of patients with a particular NSCLC stage with the COPD patient group fingerprints. Interestingly, the models’ performances were independent of the lung cancer stage. For all four groups, the subsequent models aided in achieving almost complete separation and good prediction. It is important to note that the AUC value was equal to 1 even for samples representing stage I (Table 2). This finding confirms that discrimination is possible even at early disease stages. Among the identified metabolites, acetate, citrate, and methanol concentrations were considerably reduced in lung cancer patients, whereas the concentrations of NAC1, leucine, lysine, mannose, choline, lipids (L3 + L4), and two unidentified compounds appearing at 1.39 and 1.41 ppm were elevated in all three comparisons (Table 3, Fig. S2). E-NSCLC vs. COPD and A-NSCLC vs. COPD subcomparisons revealed that various metabolites contributed to the separation between groups. Additionally, a relative increase in isoleucine, valine, 3-methyl-2-oxovalerate, 3-hydroxybutyrate, acetone, acetoacetate, isobutyrate, lactate, creatinine, ␣-glucose, lipids (L6), and an unidentified compound with a resonance of 1.05 ppm as well as reduced levels of glutamine and TMA were only observed in the E-NSCLC vs. COPD comparison (Table 3, Fig. S3). Another set of metabolites was significantly different only for the A-NSCLC vs. COPD comparison in which high levels of glycerol, NAC2, and glycine, and low levels of glyceryl of lipids (L8) were found (Table 3, Fig. S4). Patients with comorbid diseases (suffering simultaneously from NSCLC and COPD) were classified by a model as NSCLC (Figs. S5 and S6). Metabolic fingerprints showed their predominant features of NSCLC patient group. Distinction between comorbid and NSCLC subjects could not be performed, while with COPD was possible (Fig. S7). Therefore, patients with comorbidity were included into NSCLC set. 3.2. Biomarkers for NSCLC staging Metabolomics potential can be used to evaluate cancer staging [41–43]. To differentiate the metabolomic fingerprints of early stage (I and II) patients from advanced stage (III and IV) lung cancer patients, the OPLS-DA model was built (Fig. 3). Regardless of the model parameters (Q2 and P-value of CV-ANOVA), the model’s performance provides strong evidence suggesting that metabolic fingerprints contain information related to NSCLC progression (AUC from 0.8 to 0.9) (Table 2). In our study, we observed that A-NSCLC patients exhibited reduced levels of isoleucine, acetoacetate, lactate, glyceryl of lipids (L8), creatinine, acetone, valine, isobutyrate, and unidentified compounds with the resonances of 1.05 and 1.41 ppm compared with E-NSCLC patients. Simultaneously, creatine, NAC1, NAC2, and glycerol levels were elevated in A-NSCLC patients compared with E-NSCLC patients (Table 4). Of the identified metabolites, the staging biomarker candidates should reflect changes in relation to cancer progression in all four stages. In addition to statistically significant alterations among disease stages, a biomarker should exhibit an unambiguous trend reflective of these changes (e.g., subsequent increase or decrease during all stages). Therefore, we consider two criteria: P-value 0.3. For a preliminary overview, metabolite resonances

S. Deja et al. / Journal of Pharmaceutical and Biomedical Analysis 100 (2014) 369–380

373

Fig. 1. Mean 1 H NMR CPMG spectrum obtained from NSCLC patient serum. The following metabolites are identified: 1, Lipid (L1): LDL CH3 (CH2 )n ; 2, Lipid (L2): VLDL CH3 (CH2 )n ; 3, Leucine; 4, Isoleucine; 5, Valine; 6, 3-Methyl-2-oxovalerate; 7, Isobutyrate; 8, Lipid (L3): LDL CH3 (CH2 )n ; 9, Lipid (L4): VLDL CH3 (CH2 )n ; 10, Lactate; 11, Threonine; 12, Alanine; 13, Lipid (L5): VLDL CH2 CH2 C O; 14, Lysine; 15, Acetate; 16, Lipid (L6): CH2 CH CH ; 17, N-acetylated glycoproteins 1 (NAC1); 18, N-acetylated glycoproteins 2 (NAC2); 19, Lipid (L7): CH2 C O; 20, Acetone; 21, Acetoacetate; 22, 3-Hydroxybutyrate; 23, Pyruvate; 24, Glutamine; 25, Citrate; 26, Dimethylamine (DMA); 27, Trimethylamine (TMA); 28, Creatine; 29, Creatinine; 30, Dimethyl sulfone; 31, Choline; 32, Phosphocholine (PC); 33, Glycerophosphocholine (GPC); 34, Glucose; 35, Methanol; 36, Glycine; 37, Glycerol; 38, Lipid (L8): CH2 O C( O) ; 39, Mannose; 40, Lipid (L9): unsaturated lipids CH CH ; 41, Urea; 42, Tyrosine; 43, Histidine; 44, Phenylalanine; 45, Formate.

that were correlated with different groups of samples from patients with a specific NSCLC stage are highlighted in the STOCSY-like spectrum (colored according to the correlation coefficient rank) in Fig. 4. Given that some cases of COPD may be regarded as initiating cancer development, this disease was indicated by 0 and compared with consecutive NSCLC stages (1, 2, 3, and 4). This comparison should reveal potential metabolic trends initiated in COPD that progress in NSCLC (Fig. 4a). Cancer stage is negatively correlated with glutamine and acetate and positively correlated with leucine only when COPD is included in the analysis (Fig. 4a). The graphical representation of alterations in metabolite levels indicated a clear relation between lung cancer progression and NAC1, NAC2, and glycerol resonances. These three NMR signals are positively correlated with cancer stage regardless of the inclusion of COPD in the comparison (Fig. 4b).

The results of the statistical and correlation analyses obtained for relative integrals of the metabolites are presented in Table 4. Finally, only six metabolites could be considered as candidate biomarkers of lung cancer staging. It was found that during the progression of the pathological state, the levels of two metabolites, namely isoleucine and acetoacetate, were decreasing, whereas the levels of creatine, NAC1, NAC2, and glycerol were increasing (see Supplementary information Fig. S8). 4. Discussion 4.1. Metabolic changes associated with lung diseases Stage I lung cancer patients and healthy control cohorts have been recently studied using GC–MS and LC–MS techniques.

Table 2 Summary of discriminant models and validation using the test dataset. Comparison

Apred

Aorth

R2 X

R2 Y

Q2 Y

P value (CV-ANOVA)

AUC training

AUC test

COPD vs. NSCLC COPD vs. E-NSCLC COPD vs. A-NSCLC COPD vs. NSCLC stage I COPD vs. NSCLC stage II COPD vs. NSCLC stage III COPD vs. NSCLC stage IV E-NSCLC vs. A-NSCLC

1 1 1 1 1 1 1 1

2 2 3 2 2 3 2 4

0.682 0.694 0.663 0.692 0.713 0.710 0.635 0.732

0.762 0.809 0.909 0.871 0.894 0.936 0.946 0.908

0.568 0.651 0.595 0.635 0.599 0.643 0.774 0.298

2.3E−04 1.3E−05 7.6E−04 4.4E−03 8.7E−03 2.0E−02 1.0E−04 1.5E−01

0.972 1.000 0.983 1.000 1.000 0.979 1.000 0.811

0.993 1.000 1.000 1.000 1.000 1.000 1.000 0.904

Apred —number of predictive components, Aorth —number of orthogonal components.

374

S. Deja et al. / Journal of Pharmaceutical and Biomedical Analysis 100 (2014) 369–380

Fig. 2. Discrimination between: (a) all NSCLC and COPD, (b) E-NSCLC and COPD, (c) A-NSCLC and COPD; left: ROC curves; right: OPLS-DA prediction scatter plots; solid symbols: training set; empty symbols: predicted test set; circles: COPD; inverted triangles: NSCLC; triangles: E-NSCLC; diamonds: A-NSCLC.

These studies indicate that lung cancer patients exhibit perturbed metabolism of amino acids, lipids, fatty acids, glycol, and steroids [44]. The authors of a study using NMR spectroscopy of human blood plasma to detect early lung cancer stages (mainly stages I and II) reported decreased concentrations of glucose, citrate, formate, acetate, alanine, glutamine, histidine, tyrosine, valine, and methanol and elevated levels of lactate and pyruvate [19]. In addition to metabolic profiling, studies that exclusively focus on amino acid profiles using Aminoindex technology are available in the literature [45]. This technology was used to build predictive models for lung cancer patients and healthy controls [46,47] and to identify the amino acid characteristics of five types of cancer [48]. Interestingly, these studies, which were conducted in Japan, identified a similar set of amino acids that were significantly altered, including proline,

isoleucine, leucine, phenylalanine, histidine, and ornitine. The discrepancies among the pool of amino acids and other metabolites identified in lung cancer studies worldwide can be explained by regional variability in subjects’ diet and ethnicity [49]. This observation suggests that diagnostic models should also be evaluated using individuals from a specific geographical origin. In addition, results may also vary if different biofluids (e.g., plasma or serum) are used in lung cancer metabolomic studies [50]. On the other hand, COPD-specific metabolites are the result of altered protein turnover in the blood. For example, decreased lipoproteins and N, N-dimethylglycine together with increased glutamine, phenylalanine, 3-methylhistidine, and ketone bodies were observed in all of the patients, whereas decreased branched chain amino acids (BCAA) were only observed in stage GOLD IV patients

S. Deja et al. / Journal of Pharmaceutical and Biomedical Analysis 100 (2014) 369–380

375

Table 3 List of statistically significant metabolites useful for distinguishing COPD and NSCLC patients. Metabolite

NAC1 Leucine Lysine Acetate Citrate Methanol Mannose Choline Lipid (L3 + L4) Unknown signal 1.39 ppm Unknown signal 1.41 ppm Isoleucine Valine Glutamine 3-Methyl-2-oxovalerate 3-Hydroxybutyrate Acetone Acetoacetate Isobutyrate Lactate Creatinine TMA Alpha-glucose Lipid (L6) Unknown signal 1.05 ppm NAC2 Glycine Lipid (L8) Glycerol

NSCLC vs. COPD

E NSCLC vs. COPD

A NSCLC vs. COPD

%Difference

p MWW

%Difference

p MWW

%Difference

p MWW

14.8 16.2 13.7 −11.5 −16.0 −15.6 18.1 7.7 11.9 15.4 11.8 7.3 12.0 −5.3 9.8 3.7 18.7 11.7 16.5 6.4 15.5 −9.7 5.0 6.5 13.7 6.5 11.2 −7.4 13.6

0.0001 0.0001 0.0005 0.0009 0.0026 0.0025 0.0003 0.0036 0.0020 0.0000 0.0002 0.0380 0.0051 0.0611 0.1145 0.0343 0.1089 0.0126 0.0006 0.1419 0.1611 0.0085 0.0329 0.0138 0.0114 0.0123 0.0159 0.1561 0.0015

8.8 16.8 12.1 −8.4 −16.4 −15.9 17.1 7.5 13.8 19.7 17.7 12.1 19.7 −8.5 10.8 2.9 31.6 14.1 25.9 23.0 21.9 −13.5 9.6 9.1 18.8 −1.1 8.3 −1.5 5.1

0.0213 0.0004 0.0013 0.0195 0.0081 0.0021 0.0028 0.0163 0.0006 0.0000 0.0001 0.0005 0.0014 0.0195 0.0464 0.0311 0.0063 0.0019 0.0000 0.0119 0.0085 0.0010 0.0482 0.0113 0.0012 0.4758 0.1763 0.8892 0.3895

18.1 14.2 14.6 −14.2 −13.7 −14.8 21.5 9.0 10.1 14.6 7.8 0.2 8.4 −2.8 0.2 6.9 6.0 6.3 12.8 1.9 13.1 −8.5 0.5 3.9 6.2 10.5 15.8 −15.4 18.9

0.0000 0.0003 0.0020 0.0004 0.0057 0.0153 0.0004 0.0050 0.0279 0.0001 0.0048 0.6358 0.0519 0.2611 0.3545 0.0912 0.7398 0.1322 0.0555 0.7720 0.9380 0.1150 0.0611 0.0519 0.1474 0.0003 0.0040 0.0121 0.0000

% difference was calculated using median values and reflects the difference in the lung cancer patient group compared with the COPD group.

[51]. These results are consistent with the studies described by ´ Rodrıguez et al., who reported reduced levels of valine, alanine, and isoleucine in COPD patients compared with healthy volunteers [52]. Based on COPD studies, two main processes appear to characterize the disease: increased protein turnover and simultaneous intake of BCAA for gluconeogenesis processes [51–53]. It should be mentioned that not all findings may be directly transferred and compared between studies given that various analytical techniques and normalization methods may be utilized. In our study, the probabilistic quotient normalization (PQN) was used because it enables further univariate data analyses [54]. However, this normalization procedure can produce different results compared with other normalization methods [55], especially normalization to the constant sum. The ability to discriminate between NSCLC and COPD patients in our study is particularly interesting because these diseases share the same pathogenic factors, such as exposure to tobacco smoke

and chronic inflammation. This study also has greater implications because COPD can be a comorbid disease or increase the risk of lung cancer development. Therefore, it is important to monitor COPD patients to detect ongoing and emerging pathological alterations. The diagnosis of early stage lung cancer in COPD patients group is essential because of impaired lung function and thus high risk of implementation of invasive intervention—lobectomy or pneumonectomy in this group may occur. In many cases of early lung cancer, especially peripherally located, the less extensive surgical procedures or new developing stereotactic ablative radiation therapy could be applied [56,57]. Moreover, our study evaluated patients with COPD, NSCLC, and COPD with NSCLC. Despite of the comorbidity of these diseases, we were able to separate these diseases in our model. The metabolic fingerprints of NSCLC patients simultaneously suffering from COPD were significantly different from patients with only COPD, and patients with NSCLC and COPD were classified together with NSCLC,

Fig. 3. Discrimination between: E-NSCLC and A-NSCLC; left: ROC curve; right: OPLS-DA prediction scatter plot; solid symbols: training set; empty symbols: predicted test set; triangles: E-NSCLC; diamonds: A-NSCLC.

376

S. Deja et al. / Journal of Pharmaceutical and Biomedical Analysis 100 (2014) 369–380

Table 4 Metabolites significantly altered during NSCLC progression. Metabolite

Isoleucine# Acetoacetate# Lipid (L8) Creatinine Acetone Unknown doublet 1.05 ppm Valine Isobutyrate Acetate Lactate Unknown signal 1.41 ppm Unknown signal 1.39 ppm Leucine Lysine Glycine Creatine# NAC2# NAC1# Glycerol# a b c #

A NSCLC vs. E NSCLC

COPD staging I, II, III, IV

Staging I, II, III, IV

%Difference

P value a

rb

P value c

rb

P value c

−10.6% −6.9% −14.1% −7.2% −19.5% −10.6% −9.5% −10.5% −6.3% −17.1% −8.4% −4.3% −2.3% 2.2% 6.9% 13.4% 11.8% 8.6% 13.1%

0.0004 0.0282 0.0033 0.0077 0.0087 0.0321 0.0466 0.0261 0.2177 0.0194 0.0228 0.1163 0.4279 0.9145 0.0866 0.0155 0.0072 0.0050 0.0002

−0.06 −0.02 −0.28 −0.06 −0.03 0.05 0.10 0.14 −0.34 0.02 0.24 0.35 0.30 0.34 0.31 0.21 0.40 0.50 0.51

0.0011 0.0001 0.0141 0.0341 0.0392 0.0171 0.0202 0.0000 0.0100 0.0095 0.0000 0.0000 0.0009 0.0002 0.0479 0.0608 0.0027 0.0000 0.0000

−0.36 −0.34 −0.29 −0.27 −0.24 −0.22 −0.18 −0.17 −0.16 −0.15 −0.07 −0.01 0.00 0.16 0.22 0.32 0.35 0.35 0.44

0.0034 0.0010 0.0160 0.0487 0.0642 0.1423 0.2309 0.0015 0.5254 0.0137 0.0018 0.0061 0.3067 0.0134 0.2666 0.0462 0.0258 0.0087 0.0013

P value calculated using the Mann–Whitney–Wilcoxon test. Spearman correlation coefficient. P value calculated using Kruskal–Wallis one-way analysis of variance. Biomarkers potentially useful for NSCLC staging (r > 0.3, P value < 0.05).

Fig. 4. Correlation spectra. Colors denote Spearman correlation coefficient values calculated between each bin and: (a) COPD (0), stage I NSCLC (1), stage II NSCLC (2), stage III NSCLC (3), stage IV NSCLC (4); (b) stage I NSCLC (1), stage II NSCLC (2), stage III NSCLC (3), and stage IV NSCLC (4).

S. Deja et al. / Journal of Pharmaceutical and Biomedical Analysis 100 (2014) 369–380

377

Fig. 5. Results of the metabolite set enrichment analysis (MSEA) for E-NSCLC patients. The horizontal bar graph summarizes the most affected metabolic pathways in early NSCLC serum compared with COPD patients. The most significant differences include protein biosynthesis (P value = 3.51E−5, FDR = 0.00281), valine leucine and isoleucine degradation (P value = 8.9E−4, FDR = 0.03), and ketone body metabolism (P value = 0.00113, FDR = 0.03). FDR: false discovery rate.

suggesting that cancer metabolism primarily influences human serum metabolome composition. Therefore, carcinogenesis has a stronger effect on humans than COPD, which is considered a bodywasting disease. Even early lung cancer has a stronger influence on biochemical processes than chronic inflammation. This observation impacts assessments of patient health because it suggests that metabolomic-based diagnostics can distinguish between disease entities even in diseases with similar risk factors and symptoms. Interestingly, in our study reduced levels of acetate, citrate, and methanol were observed in NSCLC patients compared with COPD patients. Rocha et al. reported similar results for lung cancer patients compared with healthy controls [19]. Using 13 C stable isotope-resolved metabolomic experiments in human lung cancer tissues, increased citrate levels were observed in tumors compared with healthy tissues, suggesting that the Krebs cycle is more active in tumor tissues [58]. This finding suggests that decreased blood citrate concentrations may reflect the energetic demands of tumor tissues. In our study, only two amino acids were affected in all stages of lung cancer compared with COPD. Specifically, increased leucine and lysine levels were observed. These amino acids are exclusively

ketogenic amino acids. Opposite leucine trends (decreased leucine levels in COPD patients and increased levels in lung cancer patients compared with healthy individuals) were also reported previously in two lung diseases [46–48,51,53]. This behavior makes leucine a promising potential biomarker to distinguish lung cancer from COPD. The metabolite set enrichment analysis indicated that the main differences between COPD and the early stages of lung cancer are related to protein biosynthesis, BCAA (valine, leucine, and isoleucine) degradation and ketone body metabolism (Fig. 5). This finding is consistent with protein and BCAA degradation known to occur in COPD patients as well as the increased ketone body production in COPD, and lung cancer patients [23,51–53]. However, it seems that ketone body production is more pronounced in early lung cancer patients compared with COPD patients. 4.2. Metabolic patterns of lung cancer progression The good performance of discriminant models indicates that the E-NSCLC and A-NSCLC patients have different metabolic fingerprints. Lung cancer stratification is very important, and proper

378

S. Deja et al. / Journal of Pharmaceutical and Biomedical Analysis 100 (2014) 369–380

disease recognition influences the type of prescribed treatment. To the best of our knowledge, this study is the first to demonstrate differences between early and advanced lung cancer stages using NMR-based metabolomic fingerprints of human serum. The results obtained by Hori et al. using GC-MS uncovered differences between healthy individuals and early and advanced lung cancer patients; however, a limited number of samples were analyzed [23]. Interestingly, in our study, well-known hallmarks of cancer metabolism, including the Warburg effect (increased lactate levels) and ketosis (increased levels of all three ketone bodies), were only significantly altered during early stages of the disease. In contrast, the strongest differences for A-NSCLC patients were related to glycerol, suggesting lipid degradation or/and lung cell membrane rearrangement. The 1 H NMR signals referred to as NAC1 and NAC2 are characteristic for acute-phase glycoproteins. However, it should be emphasized that although the NAC1 signal was a common feature distinguishing all NSCLC patients from COPD patients, the NAC2 signal was specific for A-NSCLC patient serum. Elevated NAC signal intensity was reported for various cancer and non-cancer diseases [59–62]. Our recently published work suggests that NAC is one of the components that can distinguish active inflammatory bowel diseases from those in remission [63]. Therefore, this signal is not specific to lung cancer. Recently, an NMR-based serum metabolomics approach was used to differentiate metastatic and localized breast cancer patients [64]. This stratification procedure shares common features with our study distinguishing E-NSCLC and A-NSCLC patients. Indeed, we identified three metabolites significantly altered in an identical fashion: NAC1, NAC2, and glycerol. Contrary to metastatic breast cancer, acetoacetate is decreased in advanced lung cancer compared with early stages of the disease. In contrast, significantly decreased isoleucine was identified between early and advanced stages of lung cancer patients. This finding indicates that although numerous serum metabolic differences are common in various types of cancer (e.g., NAC, lipids, lactate), some specific metabolites may be altered in a way that may allow for diagnosis of a specific type of cancer. There is an important clinical application of metabolomic profile in lung cancer patients. Patients with less advanced lung cancer who have meatabolomic profile characteristic for advanced cancer should have expanded diagnostics and may require a more aggressive treatment. 4.3. Potential background of metabolic changes in lung cancer Various metabolite concentrations were increased only in the E-NSCLC patient group, whereas A-NSCLC and COPD patients displayed comparable levels. We hypothesize that this finding might be partially related to similar levels of lung tissue destruction, systemic changes and whole body impairment for both diseases. As the disease advances, the lungs are injured and might resemble COPD. Therefore, the serum metabolome should possess signatures indicative of lung dysfunction and systemic changes at advanced lung cancer stages, and these changes should not be present during early lung cancer stages. In turn, early lung cancer patients mainly experience metabolic alterations that are specifically related to tumor development. This finding would partially explain why the levels of certain metabolites are significantly altered in early stage lung cancer patients compared with COPD patients; in contrast, the levels of monitored metabolites tend to be similar between late stage lung cancer and COPD patients. Conversely, alterations in metabolic concentrations between early and advanced stage lung cancer may be explained by the postulated “reverse Warburg effect,” which states that the ketone bodies and lactate produced in fibroblasts adjacent to the tumor are re-utilized by cancer cells and stimulate breast cancer metastasis [65]. Notably, we observed a decrease in lactate and two ketone

bodies (acetone and acetoacetate) during the progression from early to advanced lung cancer. If the “reverse Warburg effect” holds true for other types of cancer, e.g., lung cancer, it could partially explain the decrease ketone bodies and lactate, which are used as a fuel by the advanced hypoxic tumors. However, this hypothesis needs to be verified experimentally. In summary, metabolic profile alterations in patients suffering from lung diseases reflect various findings related with tumor metabolism, weight loss, protein degradation, lung dysfunction, and chronic inflammation. Using our model, it is possible to accurately differentiate between similar diseases or different stages of the same disease. 5. Conclusion To the best of our knowledge, this is first report of metabolic differences between COPD and different stages of lung cancer using NMR spectroscopy analysis of human serum. Moreover, we identified characteristic metabolic alterations between early and advanced stage NSCLC patients. Six metabolites fulfilling the criteria for lung cancer staging (P value < 0.05, r > 0.3) were identified and considered to be useful for disease differentiation. Three of the metabolites remain valid when COPD is considered as starting point for lung cancer: two signals of acetyl-glycoproteins (NAC1 and NAC2, most likely acute-phase glycoproteins) and glycerol. The additional biomarkers identified include isoleucine, acetoacetate, and creatine; however, these metabolites only correlate with lung cancer stages, not COPD. The present study underscores the potential use of NMR-based serum metabolomics for early lung cancer screening in COPD patients with increased risk for lung cancer. Additionally, this technique can support current lung cancer staging methodologies and aid in proper tumor stratification. The biomarker candidates discovered in this study should be critically validated in future studies involving additional analytical techniques. All patients included in our study are under observation so the prognostic and treatment predictive value of studied biomarkers should be further evaluated. Conflict of interest None declared. Acknowledgments This study was supported by National Science Centre Poland (grant no. N N 402515939). Stanislaw Deja was a recipient of PhD scholarship under a project funded by the European Social Fund. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.jpba.2014.08.020. References [1] http://www.cancerresearchuk.org/cancer-info/cancerstats/world/. [2] http://www.cancerresearchuk.org/cancer-help/type/lungcancer/treatment/statistics-and-outlook-for-lung-cancer#outcome. [3] National Lung Screening Trial Research Team, D.R. Aberle, A.M. Adams, C.D. Berg, W.C. Black, J.D. Clapp, R.M. Fagerstrom, I.F. Gareen, C. Gatsonis, P.M. Marcus, J.D. Sicks, Reduced lung-cancer mortality with low-dose computed tomographic screening, N. Engl. J. Med. 365 (5) (2011) 395–409. [4] G. Veronesi, P. Maisonneuve, M. Bellomi, C. Rampinelli, I. Durli, R. Bertolotti, L. Spaggiari, Estimating overdiagnosis in low-dose computed tomography screening for lung cancer: a cohort study, Ann. Intern. Med. 157 (11) (2012) 776–784. [5] P. Nanavaty, M.S. Alvarez, W.M. Alberts, Lung cancer screening: advantages, controversies, and applications, Cancer Control 21 (1) (2014) 9–14.

S. Deja et al. / Journal of Pharmaceutical and Biomedical Analysis 100 (2014) 369–380 [6] R.M. Julian, Y. Ping, D.C. Stephen, E.S. Steven, A.A. Alex, Non-small cell lung cancer: epidemiology, risk factors,treatment, and survivorship, Mayo Clin. Proc. 83 (5) (2008) 584–594. [7] S. Navada, P. Lai, A.G. Schwartz, G.P. Kalemkerian, Temporal trends in small cell lung cancer: analysis of the national surveillance, epidemiology, and endresults (SEER) database, J. Clin. Oncol. 24 (18 suppl) (2006) S7082 (Meeting Abstracts). [8] E.S. Santos, A. Castrellon, M. Blaya, L.E. Raez, Controversies in the management of stage IIIA non-small-cell lung cancer, Expert Rev. Anticancer Ther. 8 (12) (2008) 1913–1929. [9] A. Punturieri, E. Szabo, T.L. Croxton, S.D. Shapiro, S.M. Dubinett, Lung cancer and chronic obstructive pulmonary disease: needs and opportunities for integrated research, J. Natl. Cancer Inst. 101 (8) (2009) 554–559. [10] E. Potton, F. McCaughan, S.M. Janes, Chronic obstructive pulmonary disease and lung cancer, Respir. Med. 5 (2) (2009) 34–37. [11] H. Wang, L. Yang, L. Zou, D. Huang, Y. Guo, M. Pan, Y. Tan, H. Zhong, W. Ji, P. Ran, N. Zhong, J. Lu, Association between chronic obstructive pulmonary disease and lung cancer: a case-control study in southern Chinese and a meta-analysis, PLoS ONE 7 (9) (2012) e46144. [12] H. Yao, I. Rahman, Current concepts on the role of inflammation in COPD and lung cancer, Curr. Opin. Pharmacol. 9 (4) (2009) 375–383. [13] T.L. Petty, Are COPD and lung cancer two manifestations of the same disease? Chest 128 (4) (2005) 1895–1897. [14] C.M. Ardies, Inflammation as cause for scar cancers of the lung, Integr. Cancer Ther. 2 (3) (2003) 238–246. [15] J.S. Brody, A. Spira, State of the art. Chronic obstructive pulmonary disease, inflammation, and lung cancer, Proc. Am. Thorac. Soc 3 (6) (2006) 535–537. [16] K. Vermaelen, G. Brusselle, Exposing a deadly alliance: novel insights into the biological links between COPD and lung cancer, Pulm. Pharmacol. Ther. 26 (5) (2013) 544–554. [17] H.A. Powell, B. Iyen-Omofoman, D.R. Baldwin, R.B. Hubbard, L.J. Tata, Chronic obstructive pulmonary disease and risk of lung cancer the importance of smoking and timing of diagnosis, J. Thorac. Oncol. 8 (4) (2013) e34–e35. [18] L.K. Newby, A.B. Storrow, W.B. Gibler, J.L. Garvey, J. Tucker, Bedside multimarker testing for risk stratification in chest pain units: the chest pain evaluation by creatine kinase-MB, myoglobin, and troponin I (CHECKMATE) study, Circulation 103 (14) (2001) 1832–1837. [19] C.M. Rocha, J. Carrola, A.S. Barros, A.M. Gil, B.J. Goodfellow, I.M. Carreira, J. Bernardo, A. Gomes, V. Sousa, L. Carvalho, I.F. Duarte, Metabolic signatures of lung cancer in biofluids: NMR-based metabonomics of blood plasma, J. Proteome Res. 10 (9) (2011) 4314–4324. [20] J. Carrola, C.M. Rocha, A.S. Barros, A.M. Gil, B.J. Goodfellow, I.M. Carreira, J. Bernardo, A. Gomes, V. Sousa, L. Carvalho, I.F. Duarte, Metabolic signatures of lung cancer in biofluids: NMR-based metabonomics of urine, J. Proteome Res. 10 (1) (2011) 221–230. [21] I.F. Duarte, C.M. Rocha, A.S. Barros, A.M. Gil, B.J. Goodfellow, I.M. Carreira, J. Bernardo, A. Gomes, V. Sousa, L. Carvalho, Can nuclear magnetic resonance (NMR) spectroscopy reveal different metabolic signatures for lung tumours? Virchows Arch. 457 (6) (2010) 715–725. [22] K.W. Jordan, C.B. Adkins, L. Su, E.F. Halpern, E.J. Mark, D.C. Christiani, L.L. Cheng, Comparison of squamous cell carcinoma and adenocarcinoma of the lung by metabolomic analysis of tissue–serum pairs, Lung Cancer 68 (1) (2010) 44–50. [23] S. Hori, S. Nishiumi, K. Kobayashi, M. Shinohara, Y. Hatakeyama, Y. Kotani, N. Hatano, Y. Maniwa, W. Nishio, T. Bamba, E. Fukusaki, T. Azuma, T. Takenawa, Y. Nishimura, M. Yoshida, A metabolomic approach to lung cancer, Lung Cancer 74 (2) (2011) 284–292. [24] C.M. Rocha, A.S. Barros, A.M. Gil, B.J. Goodfellow, E. Humpfer, M. Spraul, I.M. Carreira, J.B. Melo, J. Bernardo, A. Gomes, V. Sousa, L. Carvalho, I.F. Duarte, Metabolic profiling of human lung cancer tissue by 1H high resolution magic angle spinning (HRMAS) NMR spectroscopy, J. Proteome Res. 9 (1) (2010) 319–332. [25] L. Wang, Y. Tang, S. Liu, S. Mao, Y. Ling, D. Liu, X. He, X. Wang, Metabonomic profiling of serum and urine by 1H NMR-based spectroscopy discriminates patients with chronic obstructive pulmonary disease and healthy individuals, PLoS ONE 8 (6) (2013) e65675. [26] K.F. Rabe, S. Hurd, A. Anzueto, P.J. Barnes, S.A. Buist, P. Calverley, Y. Fukuchi, C. Jenkins, R. Rodriguez-Roisin, C. van Weel, J. Zielinski, Global initiative for chronic obstructive lung disease. Global strategy for the diagnosis, management, and prevention of chronic obstructive pulmonary disease: GOLD executive summary, Am. J. Respir. Crit. Care. Med. 176 (6) (2007) 532–555. [27] P. Goldstraw, Lung and pleural tumours, in: P. Goldstraw (Ed.), International Association for the Study of Lung Cancer. Staging Manual in Thoracic Oncology, Editorial Rx Press, Florida, 2009, p. 164. [28] O. Beckonert, H.C. Keun, T.M. Ebbels, J. Bundy, E. Holmes, J.C. Lindon, J.K. Nicholson, Metabolic profiling, metabolomic and metabonomic procedures for NMR spectroscopy of urine, plasma, serum and tissue extracts, Nat. Protoc. 2 (11) (2007) 2692–2703. [29] A. Smolinska, L. Blanchet, L.M. Buydens, S.S. Wijmenga, NMR and pattern recognition methods in metabolomics: from data acquisition to biomarker discovery: a review, Anal. Chim. Acta 750 (2012) 82–97. [30] F. Savorani, G. Tomasi, S.B. Engelsen, Icoshift: a versatile tool for the rapid alignment of 1D NMR spectra, J. Magn. Reson. 202 (2) (2010) 190–202. [31] F. Dieterle, A. Ross, G. Schlotterbeck, H. Senn, Probabilistic quotient normalization as robust method to account for dilution of complex biological mixtures. Application in 1 H NMR metabonomics, Anal. Chem. 78 (13) (2006) 4281–4290.

379

[32] J. Trygg, S. Wold, Orthogonal projections to latent structures (OPLS), J. Chemom. 16 (2002) 119–128. [33] J. Xia, D.S. Wishart, MSEA: a web-based tool to identify biologically meaningful patterns in quantitative metabolomic data, Nucleic Acids Res. 38 (2010) W71–W77. [34] J.K. Nicholson, P.J. Foxall, M. Spraul, R.D. Farrant, J.C. Lindon, 750 MHz 1H and 1H-13 C NMR spectroscopy of human blood plasma, Anal. Chem. 67 (5) (1995) 793–811. [35] J.A. Westerhuis, H.C.J. Hoefsloot, S. Smit, D.J. Vis, A.K. Smilde, E.J.J. van Velzen, J.P.M. van Duijnhoven, F.A. van Dorsten, Assessment of PLS-DA cross validation, Metabolomics 4 (1) (2008) 81–88. [36] R.W. Kennard, L.A. Stone, Computed aided design of experiments, Technometrics 11 (1969) 137–148. [37] N.A. Obuchowski, M.L. Lieber, F.H.Jr. Wians, ROC curves in clinical chemistry: uses, misuses, and possible solutions, Clin. Chem. 50 (7) (2004) 1118–1125. [38] M.H. Zweig, G. Campbell, Receiver-operating characteristic (ROC) plots: a fundamental evaluation tool in clinical medicine, Clin. Chem. 39 (4) (1993) 561–577. [39] M.S. Pepe, R. Etzioni, Z. Feng, J.D. Potter, M.L. Thompson, M. Thornquist, M. Winget, Y. Yasui, Phases of biomarker development for early detection of cancer, J. Natl. Cancer Inst. 93 (14) (2001) 1054–1061. [40] J. Xia, D.I. Broadhurst, M. Wilson, D.S. Wishart, Translational biomarker discovery in clinical metabolomics: an introductory tutorial, Metabolomics (2013) 280–299. [41] M. Cao, L. Zhao, L. Chen, W. Xue, D. Lin, NMR-based metabolomic analysis of human bladder cancer, Anal. Sci. 28 (5) (2012) 451–456. [42] L. Wang, J. Chen, L. Chen, P. Deng, Q. Bu, P. Xiang, M. Li, W. Lu, Y. Xu, H. Lin, T. Wu, H. Wang, J. Hu, X. Shao, X. Cen, Y.L. Zhao, 1H-NMR based metabonomic profiling of human esophageal cancer tissue, Mol. Cancer 4 (2013) 12–25. [43] S. Deja, T. Dawiskiba, W. Balcerzak, M. Orczyk-Pawiłowicz, M. Głód, D. Pawełka, P. Młynarz, Follicular adenomas exhibit a unique metabolic profile. 1H NMR studies of thyroid lesions, PLoS ONE 8 (12) (2013) e84637. [44] T. Wen, L. Gao, Z. Wen, C. Wu, C.S. Tan, W.Z. Toh, C.N. Ong, Exploratory investigation of plasma metabolomics in human lung adenocarcinoma, Mol. Biosyst. 9 (9) (2013) 2370–2378. [45] Y. Noguchi, Q.W. Zhang, T. Sugimoto, Y. Furuhata, R. Sakai, M. Mori, M. Takahashi, T. Kimura, Network analysis of plasma and tissue amino acids and the generation of an amino index for potential diagnostic use, Am. J. Clin. Nutr. 83 (2) (2006), 513S-9S. [46] J. Maeda, M. Higashiyama, A. Imaizumi, T. Nakayama, H. Yamamoto, T. Daimon, M. Yamakado, F. Imamura, K. Kodama, Possibility of multivariate function composed of plasma amino acid profiles as a novel screening index for non-small cell lung cancer: a case control study, BMC Cancer 10 (2010) 690. [47] M. Shingyoji, T. Iizasa, M. Higashiyama, F. Imamura, N. Saruki, A. Imaizumi, H. Yamamoto, T. Daimon, O. Tochikubo, T. Mitsushima, M. Yamakado, H. Kimura, The significance and robustness of a plasma free amino acid (PFAA) profilebased multiplex function for detecting lung cancer, BMC Cancer 13 (2013) 77. [48] Y. Miyagi, M. Higashiyama, A. Gochi, M. Akaike, T. Ishikawa, T. Miura, N. Saruki, E. Bando, H. Kimura, F. Imamura, M. Moriyama, I. Ikeda, A. Chiba, F. Oshita, A. Imaizumi, H. Yamamoto, H. Miyano, K. Horimoto, O. Tochikubo, T. Mitsushima, M. Yamakado, N. Okamoto, Plasma free amino acid profiling of five types of cancer patients and its application for early detection, PLoS ONE 6 (9) (2011) e24143. [49] F. Guarner, J.R. Malagelada, Gut flora in health and disease, Lancet 361 (9356) (2003) 512–519. [50] D.C. Wedge, J.W. Allwood, W. Dunn, A.A. Vaughan, K. Simpson, M. Brown, L. Priest, F.H. Blackhall, A.D. Whetton, C. Dive, R. Goodacre, Is serum or plasma more appropriate for intersubject comparisons in metabolomic studies? An assessment in patients with small-cell lung cancer, Anal. Chem. 83 (17) (2011) 6689–6697. [51] B.K. Ubhi, J.H. Riley, P.A. Shaw, D.A. Lomas, R. Tal-Singer, W. MacNee, J.L. Griffin, S.C. Connor, Metabolic profiling detects biomarkers of protein degradation in COPD patients, Eur. Respir. J. 40 (2) (2012) 345–355. ´ ´ S. Dıaz-Moralli, ´ ´ [52] D.A. Rodrıguez, G. Alcarraz-Vizan, M. Reed, F.P. Gomez, F. Falciani, U. Günther, J. Roca, M. Cascante, Plasma metabolic profile in COPD patients: effects of exercise and endurance training, Metabolomics 8 (2012) 508–516. [53] L. Wang, Y. Tang, S. Liu, S. Mao, Y. Ling, D. Liu, X. He, X. Wang, Metabonomic profiling of serum and urine by 1H NMR-based spectroscopy discriminates patients with chronic obstructive pulmonary disease and healthy individuals, PLoS ONE 8 (6) (2013) e65675. [54] T. De Meyer, D. Sinnaeve, B. Van Gasse, E.R. Rietzschel, M.L. De Buyzere, M.R. Langlois, S. Bekaert, J.C. Martins, W. Van Criekinge, Evaluation of standard and advanced preprocessing methods for the univariate analysis of blood serum 1 H-NMR spectra, Anal. Bioanal. Chem. 398 (4) (2010) 1781–1790. [55] S.M. Kohl, M.S. Klein, J. Hochrein, P.J. Oefner, R. Spang, W. Gronwald, Stateof-the art data normalization methods improve NMR-based metabolomic analysis, Metabolomics 8 (Suppl 1) (2012) S146–S160. [56] C.B. Smith, S.J. Swanson, G. Mhango, J.P. Wisnivesky, Survival after segmentectomy and wedge resection in stage I non-small-cell lung cancer, J. Thorac. Oncol. 8 (1) (2013) 73–78. [57] J.H. Heinzerling, B. Kavanagh, R.D. Timmerman, Stereotactic ablative radiation therapy for primary lung tumors, Cancer J. 17 (1) (2011) 28–32.

380

S. Deja et al. / Journal of Pharmaceutical and Biomedical Analysis 100 (2014) 369–380

[58] T.W. Fan, A.N. Lane, R.M. Higashi, M.A. Farag, H. Gao, M. Bousamra, D.M. Miller, Altered regulation of metabolic pathways in human lung cancer discerned by 13 C stable isotope-resolved metabolomics (SIRM), Mol. Cancer 8 (2009) 41. [59] I. Bertini, S. Cacciatore, B.V. Jensen, J.V. Schou, J.S. Johansen, M. Kruhøffer, C. Luchinat, D.L. Nielsen, P. Turano, Metabolomic NMR fingerprinting to identify and predict survival of patients with metastatic colorectal cancer, Cancer Res. 72 (1) (2012) 356–364. [60] G.M. Torri, J. Torri, J.M. Gulian, J. Vion-Dury, P. Viout, P.J. Cozzone, Magnetic resonance spectroscopy of serum and acute-phase proteins revisited: a multiparametric statistical analysis of metabolite variations in inflammatory, infectious and miscellaneous diseases, Clin. Chim. Acta 279 (1999) 77–96. [61] J.D. Bell, J.C. Brown, J.K. Nicholson, P.J. Sadler, Assignment of resonances for “acute-phase” glycoproteins in high resolution proton NMR spectra of human blood plasma, FEBS Lett. 215 (2) (1987) 311–315.

[62] A.N. Zira, S.E. Theocharis, D. Mitropoulos, V. Migdalis, E. Mikros, 1H NMR metabonomic analysis in renal cell carcinoma: a possible diagnostic tool, J. Proteome Res. 9 (8) (2010) 4038–4044. ˛ ´ D. Pawełka, M. Banasik, [63] T. Dawiskiba, S. Deja, A. Mulak, A. Zabek, E. Jawien, ´ K. Korta, K. A. Mastalerz-Migas, W. Balcerzak, K. Kaliszewski, J. Skóra, P. Barc, ´ P. Szyber, A. Litarski, P. Młynarz, Serum and urine metabolomic Pormanczuk, fingerprinting in diagnostics of inflammatory bowel diseases, World J. Gastroenterol. 20 (1) (2014) 163–174. [64] E. Jobard, C. Pontoizeau, B.J. Blaise, T. Bachelot, B. Elena-Herrmann, O. Trédan, A serum nuclear resonance-based metabolomic signature of advanced metastatic human breast cancer, Cancer Lett. 343 (1) (2014) 33–41. [65] G. Bonuccelli, A. Tsirigos, D. Whitaker-Menezes, S. Pavlides, Ketones and lactate fuel tumor growth and metastasis, Cell Cycle 9 (17) (2010) 3506–3514.