Metabolic signatures of cancer unveiled by NMR spectroscopy of human biofluids

Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

Contents lists available at SciVerse ScienceDirect

Progress in Nuclear Magnetic Resonance Spectroscopy journal homepage: www.elsevier.com/locate/pnmrs

Metabolic signatures of cancer unveiled by NMR spectroscopy of human biofluids Iola F. Duarte ⇑, Ana M. Gil CICECO, Department of Chemistry, University of Aveiro, Campus de Santiago, 3810-193 Aveiro, Portugal

a r t i c l e

i n f o

a b s t r a c t

Article history: Received 4 October 2011 Accepted 23 November 2011 Available online 29 November 2011

Ó 2011 Elsevier B.V. All rights reserved.

Keywords: Nuclear Magnetic Resonance (NMR) spectroscopy Metabonomics Metabolomics Multivariate analysis Biofluids Cancer Metabolic signature

Contents 1. 2.

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . NMR-based metabolic signatures of different cancer types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1. Biliary tract cancer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2. Bladder cancer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3. Brain cancer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4. Breast cancer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5. Colorectal cancer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.6. Esophageal cancer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.7. Kidney cancer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.8. Leukaemia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.9. Liver cancer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.10. Lung cancer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

52 54 54 54 64 64 66 66 66 67 67 67

Abbreviations: AL, acute leukaemia; Apo-IV and ApoA-I, apolipoproteins; AUC, area under curve; BE, Barret’s esophagus; BMRB, Biological Magnetic Resonance Data Bank; BPH, benign prostatic hyperplasia; CA 19-9, CEA, CA 27.29, cancer antigens; CC, cholangiocarcinoma; CCA, canonical correlation analysis; CLL, chronic lymphocytic leukaemia; COSY, Correlation Spectroscopy; CPMG, Carr–Purcell–Meiboom–Gill; CRC, colorectal cancer; CSF, cerebrospinal fluid; CT, Computed Tomography; DA, discriminant analysis; DART-MS, direct analysis in real time Mass Spectrometry; DIGE, differential gel electrophoresis; DSS, 4,4-dimethyl-4-silapentane-1-sulfonic acid; EAC, esophageal adenocarcinoma; EOC, epithelial ovarian cancer; ER, estrogen receptor; FDA, Food and Drug Administration; GCGC–MS, bidimensional Gas Chromatography coupled to Mass Spectrometry; HCA, Hierarchical Clustering Analysis; HCC, hepatocellular carcinoma; HDL, high density lipoproteins; HGD, high-grade dysplasia; HMBC, Heteronuclear Multiple Bond Correlation; HMDB, Human Metabolome Database; HRMAS, High Resolution Magic Angle Spinning; HSQC, Heteronuclear Single Quantum Correlation; LC, liver cirrhosis; LC–MS, Liquid Chromatography coupled to Mass Spectrometry; LDL, low density lipoproteins; LG, logistic regression; M-/U-IGHV, mutated/unmutated immunoglobulin heavy chain variable region genes; MCCV, Monte Carlo cross validation; MS, Mass Spectrometry; MVA, multivariate analysis; NED, no evidence of disease; NLD, non-liver disease; NMLD, non-malignant liver diseases; NMR, Nuclear Magnetic Resonance; OLK, oral leucoplakia; OPLS, orthogonal projections to latent structures; OPLS-DA, orthogonal projections to latent structures discriminant analysis; OSC, orthogonal signal correction; OSCC, oral squamous cell carcinoma; OSC-PLS-DA, orthogonal signal correction partial least squares discriminant analysis; PCA, Principal Component Analysis; PLS, partial least squares; PLS-DA, partial least squares discriminant analysis; PLS-DF, partial least squares discriminant function; PR, progesterone receptor; PSA, prostate specific antigen; RCC, renal cell carcinoma; ROC, receiver operating characteristic; SIMCA, Soft Independent Modeling of Class Analogy; STOCSY, Statistical Total Correlation Spectroscopy; SVM, Support Vector Machines; TCA, tricarboxylic acid; TOCSY, Total Correlation Spectroscopy; TSP, 3-(trimethylsilyl)-propionic acid; UBC, urinary bladder cancer; VIP, Variable Importance in the Projection; VLDL, very low density lipoproteins; WOS, Web of Science. ⇑ Corresponding author. Tel.: +351 234401424; fax: +351 234401470. E-mail address: [email protected] (I.F. Duarte). 0079-6565/$ - see front matter Ó 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.pnmrs.2011.11.002

52

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

2.11. Oral cancer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.12. Ovarian cancer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.13. Pancreatic cancer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.14. Prostate cancer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.15. Thyroid cancer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Integration of published results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Progress into the clinic: needs and future directions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3. 4.

1. Introduction

Number of papers (Web of Science)

Metabonomics entails the comprehensive analysis of low molecular weight molecules (metabolites) involved in intermediary metabolism, through advanced profiling techniques, such as Nuclear Magnetic Resonance (NMR) spectroscopy and Mass Spectrometry (MS)-based methods, combined with multivariate statistical analysis [1,2]. The general aim of this approach is to determine fluctuations on the levels of endogenous metabolites, through the analysis of a given biological matrix (e.g. tissue, biofluid, cultured cells), and to mine for consistent relationships between those metabolic variations and specific pathophysiological conditions or external perturbations, such as disease, diet or therapeutic intervention. By reflecting the upstream activity of genes and proteins, as well as being modulated by factors unrelated to the genome (e.g. interaction with commensal microorganisms and environmental agents, nutritional and other lifestyle-related aspects), metabolite levels closely express cellular function, constituting a sensitive probe for homeostasis and its regulation. As pathological conditions usually disrupt normal metabolism and homeostasis, leading to altered metabolite levels and/or profiles, metabonomics holds great potential in disease diagnosis and monitoring. Indeed, metabonomics has been increasingly recognized as a valuable complementary approach to other well-established ‘omic’ sciences (genomics and proteomics), being part of a wide biomarkers development effort, promoted by the Food and Drug Administration (FDA), to aid in the assessment of disease and toxicity (www.fda.gov/nctr/science/centers/metabolomics). Several pathologies have been extensively investigated through metabonomics including cancer, diabetes, cardiovascular and neurological diseases [3–6]. Cancer in particular has been the focus of an increasingly large number of studies in recent years (Fig. 1). Cancer cells show distinct metabolic behavior to face limited nutri-

Cancer studies using

NMR

other analytical techniques

180 160 140 120

*

100 80 60 40 20 0

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

Year of publication Fig. 1. Number of cancer-related metabonomics publications per year. The literature review was performed in the Web of Science (WOS) database, using the keywords cancer, metabolomic(s) or metabonomic(s), and Nuclear Magnetic Resonance or NMR. The data was collected up to July 2011.

68 69 70 70 70 71 72 73 73

ent and oxygen availability and to meet the high energetic and biosynthetic demands for aberrant cell proliferation. Examples of known key metabolic shifts in cancer include enhanced glucose uptake and glycolytic activity (with concomitant lactate production), increased de novo biosynthesis of nucleotides via the pentose phosphate pathway, increased glutaminolysis and a shift in citrate metabolism from oxidation to lipogenesis [7–10]. Direct evidence of alterations in these pathways has been provided through metabolic profiling of tumor cells and tissues [11]. Moreover, as metabonomics enables the holistic non-selective assessment of metabolites, new hypothesis concerning cancer metabolic reprogramming could emerge. Tissue metabonomics, employing mainly 1 H High Resolution Magic Angle Spinning (HRMAS) NMR for direct sample analysis, has enabled general and specific metabolic markers of malignancy to be defined for a range of different cancer types [12–14]. This approach has also been explored for cancer characterization, for instance concerning tumor grading [15] and correlation to prognostic factors [16], both aspects being crucial in disease evaluation and treatment planning. While analysis of tumor tissues and cells represents a direct window to cancer cellular altered metabolism, metabolic profiling of biofluids has the potential to assess the complex dynamic interaction between tumor and host, a relationship which is likely to play a critical role in defining clinical outcomes and response to therapy. Furthermore, biofluids such as blood or urine are easily obtained through minimally-invasive or non-invasive collection, thus representing increasingly attractive sources for biomarkers. Indeed, in recent years, there has been a substantial upsurge of studies focused on the metabonomic investigation of biofluids, mostly blood serum/plasma and urine, aimed at achieving various goals such as early cancer screening, improved diagnostic accuracy or prediction of response to therapy. A significant part of these studies has made use of high resolution NMR spectroscopy for characterizing the metabolic composition of biofluids. Among all available analytical platforms in metabonomics [17,18], NMR stands out as the most robust and reliable technique, presenting unparalleled analytical reproducibility [19] and generally requiring minimal sample preparation, thus preserving the native form of biofluids and even allowing for their recovery, as measurements are non-destructive. Another distinct advantage of NMR is the ability to simultaneously provide structural and quantitative information. Direct and unequivocal identification of metabolites, including those which are unexpected or unknown, can often be attained, thus assisting the biochemical interpretation of the results. Moreover, NMR is intrinsically quantitative with peak integrals directly reflecting concentrations. The greatest limitation of NMR, however, lies in its inherent low sensitivity, which precludes the detection of metabolites present in levels below a few micromolar, at least in the 500–600 MHz spectrometers commonly used in metabonomics studies. Although not yet routinely employed, advances in NMR instrumentation, including the introduction of cryogenically cooled probes and microcoil probes, together with specific methods, such as isotope labeling and 1D selective Total Correlation Spectroscopy (TOCSY),

53

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

such as the Human Metabolome Database (HMDB) [22] and the Biological Magnetic Resonance Data Bank (BMRB) [23]. For some sample types (e.g. blood plasma or serum), broader signals arising from larger molecules, e.g. lipids and proteins, may also be overlapped with sharper resonances, further increasing the intricacy of spectral assignment. In these instances, spectral editing according to the different relaxation and diffusion properties of metabolites and larger molecules is a useful approach. The Carr–Purcell– Meiboom–Gill (CPMG) experiment is often used to attenuate the broad signals of large, fast-relaxing molecules (e.g. serum/plasma proteins and lipoproteins) and improve the visibility of sharp signals from small metabolites, whereas diffusion-edited experiments allow the clearer observation of macromolecular profiles. Besides signal assignment, the other critical task in a metabonomics study is the detection of consistent patterns of changes in the dataset, and then correlation to the pathology or perturbation

are being implemented to reduce the sensitivity problem and expand the array of NMR-detected metabolites in biofluids [20]. Further information on instrumental and experimental aspects of NMR spectroscopy of biofluids can be found in several other reviews [17,21]. One-dimensional (1D) 1H NMR spectra of biofluids typically show complex profiles, comprising many overlapped signals (Fig. 2). The assignment of these signals and translation of the spectrum into a list of metabolites relies necessarily on twodimensional (2D) NMR experiments, such as TOCSY, heteronuclear correlation experiments (e.g. 1H–13C HSQC – Heteronuclear Single Quantum Correlation and HMBC – Heteronuclear Multiple Bond Correlation) and J-resolved spectroscopy, to increase signal dispersion and reveal molecular connectivities. The chemical shift and multiplicity patterns retrieved from 1D and 2D experiments are then matched to those of reference spectra in available databases,

(a) Blood plasma

9

8

7

6

5

4

3

2

1

ppm

8

7

6

5

4

3

2

1

ppm

8

7

6

5

4

3

2

1

ppm

7

6

5

4

3

2

1

ppm

(b) Urine

9

(c) Hepatic bile

9

(d) Amniotic fluid

9 1

8

Fig. 2. Typical H NMR spectra of biofluid samples: (a) blood plasma, (b) urine, (c) hepatic bile, (d) amniotic fluid. All spectra were recorded at 500 MHz for 1H observation, with the exception of the bile spectrum, which was recorded at 600 MHz.

54

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

under study, while assessing their classification/predictive value. As each spectrum comprises thousands of variables (spectral intensities) and tens to hundreds of spectra are usually collected within a study, multivariate statistical analysis (MVA) is required. A wide range of MVA methods is available today [24,25], with linear methods being the most routinely applied. Initial data exploration is usually carried out through unsupervised methods, which do not include information about sample class membership, thus providing an unbiased overview of the variability in the dataset. Common examples of such methods are Hierarchical Clustering Analysis (HCA) and Principal Component Analysis (PCA), the most widely used unsupervised approach in metabonomics. Once clustering behavior has been detected, supervised methods are usually employed, to maximize class separation, build classification models and validate biomarker profiles. These methods incorporate class membership information, usually aiming at describing the maximum covariance between the predictor variables and some response variable or classifier. Partial least squares (PLS) and Orthogonal Projection to Latent Structures (OPLS) are amongst the supervised methods most commonly employed, often being combined with Discriminant Analysis (DA) to establish an optimal position to trace a discriminant surface that best separates classes. Further details about these and other MVA methods can be found elsewhere [24,25]. This review intends to give a critical appraisal of NMR-based metabonomics studies of biofluids aimed at unveiling metabolic signatures related to cancer. This work is motivated by the substantial growth of such studies in recent years, which propels the need to systematize and integrate the reported results in order to grasp their achievements and limitations and to accomplish a comprehensive view of the metabolic pictures captured for different cancer types. Other previous and very useful broader-scope reviews have provided accounts on metabonomic applications in oncology employing different analytical platforms (essentially NMR and MS-based methods) and biological matrices (cells, tissues and biofluids) [26–29]. A particularly comprehensive work, recently published, compiled the information on metabonomics-derived cancer marker metabolites, discussing them in the context of altered cancer metabolism [29]. By focusing on NMR studies of biofluids, the present review is expected to complement the existing literature by specifically addressing the following goals: (i) to highlight both common and specific metabolite variations in biofluids (particularly urine and blood serum/plasma) across different cancer types; (ii) to address relevant influential factors and pitfalls in study design, biofluid analysis and data analysis; (iii) to help in evaluating the current status of biofluid NMR-based metabonomics in the oncology field; (iv) to point future directions towards clinical validation of the metabonomics methodology. Literature searching was based primarily on the Web of Science (WOS) and PubMed databases, being confined to human biofluids (thus excluding samples from animal models). The search has also focused on the direct analysis of biofluids, although a few examples of NMR analysis of extracts are mentioned. It should additionally be clarified that, in the face of the current trend to use the terms metabonomics and metabolomics interchangeably, both terms were considered in the literature search, although metabonomics has been selected for use throughout this paper.

Table 1 Overview of studies employing NMR spectroscopy to search for cancer-related marker metabolites in human biofluids. Cancer type

Published studies [ref. no.] Blood serum or plasma

a

Biliary tract Bladder Brain Breast Colorectal Esophageal Kidney Leukaemia Liver Lung Oral Ovarian Pancreatic Prostate

[33]a [36–39] [40,44] [45] [46]a, [47,48] [49]a, [50] [51] [55,56] [58,59] [61,62] [64,65] [69]

Thyroid

[71]a

Urine

Other fluids [30] – bile

[32]a [33]a – cerebrospinal fluid [34] [43]

[41,42] – fecal extracts

[52,53] [43,57]

[54]a – bile

[34]

[60]a – ovarian cyst fluid [66–68]a – seminal fluid/ prostatic secretions

Studies not employing multivariate statistical analysis.

ogy. Since they are easily collected through minimally invasive means, these biofluids represent a convenient, readily accessible, source to search for biomarkers of disease onset, progression, response to therapy or other pathophysiological responses that reflect the complex disease–host interaction. Other biofluids which have been profiled by NMR in the context of cancer include bile, ovarian fluid, prostatic secretions and cerebrospinal fluid. Their use has, however, been limited to specific cases in which they are in direct contact with the affected organ or are especially relevant in the specific organ system under study. In the following sections, the studies listed in Table 1 are reviewed within the biological context and clinical needs pertaining to different cancer types. To facilitate systematization and comparison of those studies, Tables 2 and 3 summarize, respectively, the experimental conditions used and the MVA methods employed, while Table 4 shows an overview of the main metabolic findings and their proposed biochemical interpretation. 2.1. Biliary tract cancer Aiming at contributing to the diagnosis of biliary tract cancer and differentiating it from benign biliary duct diseases, bile samples from two groups (cancer n = 17, benign disease n = 21) have been analyzed by 1H NMR spectroscopy and their profiles subjected to multivariate modeling [30]. By using a leave-one-out predictive test, the OPLS-DA model obtained exhibited a sensitivity of 88% and a specificity of 81%, leading the authors to suggest a better performance compared to conventional serum markers (cancer antigens CA 19-9 and CEA) or bile cytology. The contributing NMR signals were analyzed using a statistical TOCSY (STOCSY) approach [31]. Differences in bile acid composition and in the amount of citric acid (found to be higher in cancer patients) were shown to play an important role in class discrimination. 2.2. Bladder cancer

2. NMR-based metabolic signatures of different cancer types Table 1 provides an overview of studies reporting the NMR analysis of human biofluids to characterize the metabolic phenotype (or related conditions) of different cancer types. Similarly to other clinical metabonomics research, blood serum (or plasma) and urine are the most extensively investigated biofluids in oncol-

Motivated by the expectation that urine might directly mirror metabolic perturbations in urinary bladder cancer (UBC) cells, 1H NMR spectroscopy has been used to analyze urine samples from UBC patients (n = 33), healthy controls (n = 37) and subjects with benign pathological conditions (urinary tract infection n = 31 or bladder stone n = 2) [32]. The authors performed absolute quantification of selected metabolites, using 3-(trimethylsilyl)-propionic

Table 2 Subject groups and experimental conditions used in NMR cancer-related studies of human biofluids. m: male, f: female. Cancer type

Subject groups

Sample collection, storage and preparation

NMR field/ nucleus

30m,2f 8m,2f

 Patients’ sera collected before surgery, stored 80 °C  ‘Normal’ sera obtained from blood bank  CHCl3/MeOH lipids extraction, extracts re-dissolved in CDCl3

400 MHz/ 1H and

40–75

f

 Commercial sera in dry ice, stored 80 °C  D2O/TSP added, vortexed, centrifuged (7000 rpm, 10 min)

500 MHz/ 1H

Recurrent breast cancer n 20 No evidence of disease n 36

53 (37–75) 55 (36–69)

f

 Sera stored 80 °C  D2O and NaN3 added, vortexed, centrifuged (8000 rpm, 5 min)

500 MHz cryoprobe/ 1H

[38]

Early breast cancer (BC) n 44 Metastatic breast cancer n 51 Early BC (validation) n 45

58 (34–76) 59 (38–86) 54 (30–85)

f

 Blood (fasting) into serum tubes, centrifuged (1500g, 10 min, RT), stored 80 °C  Samples shaken, phosphate buffer pH 7.4 added

600 MHz/ 1H

[39]

Breast cancer n 21 (Weight gain n 10, No gain n 11)

–

f

 Blood centrifuged (12000g, 10 min, 4 °C), sera 20 °C, transferred 24h  D2O saline added, centrifuged (12000g, 5 min), stored 40 °C

600 MHz/ 1H

CRC n 38 Non-cancer controls n 19 Adenoma n 8

67 ± 13 63 ± 10

–

 Blood clotting at 41 °C 1–2h, centrifuged (2500g, 10 min), stored

CRC patients undergoing capecitabine treatment n 52

79 (42–86)

18m,34f

 Sera stored 80 °C  D2O saline added, centrifuged (16000g, 5 min), stored -40 °C

600 MHz/ 1H

EAC n 68 Barrett’s esophagus n 5 High-grade dysplasia n 11 Healthy n 34

65.6

57m,11f

 Blood (fasting) clotting 45 min, centrifuged (2000 rpm, 10 min), sera stored 80 °C  D2O added, TSP (capillary)

500 MHz/ 1H

RCC n 29 Healthy n 19

62 (37–88) 51 (38–60)

21m,8f 8m,11f

 Blood (fasting) into EDTA tubes, plasma separated by centrifugation (3000g, 10 min), standard solutiona added, stored 20 °C

300 MHz/

[47]

RCC low grade n 49 RCC advanced n 25 Healthy n 55

39–68 41–76 52 ± 13

34m,15f 15m,10f 31m,24f

 Blood (fasting) clotting 1h RT, centrifuged (1024g, 10 min, 4 °C), sera stored 80 °C  Phosphate buffer pH 7.4 added, centrifuged (12000g, 10 min, 4 °C)

600 MHz/ 1H

[48]

RCC n 32 Benign urological problems n 13

63 (45–80) 54 (56–74)

–

 Blood (fasting) into EDTA tubes, plasma separated by centrifugation (1000g, 10 min, RT), stored 80 °C  D2O added

400 MHz/ 1H

Acute leukaemia n 30 Healthy n 21

46 (22–72) 32 (22–55)

18m,12f 11m,10f

 Blood (fasting) into heparinized tubes, centrifuged (no details)  Plasma PL extraction by Folch’s method

300 MHz/

[ref. no.]

Control-patient matching Age

Gender

Brain cancer n 32 Healthy n 10

35.6 ± 14.9 31.4 ± 13.9

Breast cancer n 27 Healthy n 30

[37]

Blood serum or plasma Brain [33]

Breast [36]

[44]

Esophageal (EAC) [45]

Kidney (RCC) [46]

Leukaemia [49]

80 °C

P

800 MHz cryoprobe/ 1H

31

P

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

Colorectal (CRC) [40]

80 °C in

31

31

P

55

(continued on next page)

56

Table 2 (continued) Cancer type

Subject groups

[ref. no.] [50]

Liver (HCC) [51]

[56]

Age

Gender

Chronic lymphocytic leuk. n 29 (M-IGHV n 19, UM-IGHV n 10) Healthy n 9

71 ± 2 63 ± 4

18m,11f 5m,4f

HCC n 39 Liver Cirrhosis n 36 Healthy n 63

–

Lung cancer n 14 (SCC n 5, AC n 9) Healthy n 7

Lung cancer n 85 Healthy n 78

Sample collection, storage and preparation

600 MHz/ 1H



Blood (fasting) into vacuum tubes, clotting 45 min RT, centrifuged (1000g, 10 min), sera stored 80 °C D2O saline added, centrifuged (12000g, 5 min, 4 °C)

–

 

Blood clotting 1h RT, sera stored 80 °C Phosphate buffer pH 7.4 added, centrifuged (12000g, 10 min, 4 °C)

600 MHz/ 1H

–

–

 

Sera collected just before day of surgery, stored frozen D2O added

600 MHz HRMAS/ 1H

63 (30–85) 41 (22–60)

55m,30f 38m,40f



Blood (fasting) into heparinized tubes, collected pre-surgery for patients, centrifuged (1500g, 10 min), stored 80 °C D2O saline added, centrifuged (8000 rpm, 5 min)

500 MHz/ 1H

Blood (nonfasting) centrifuged (3500 rpm, 5 min), sera frozen in liquid nitrogen and stored 80 °C Filtered 4 °C by centrifugation (10000 rpm), diluted with water, phosphate buffer (with NaN3, TMSP/D2O) added, pH 7.0 ± 0.1

500 MHz cryoprobe/ 1H

Blood (fasting) into lithium heparin tubes, centrifuged (3000g, 10 min, 4 °C), plasma stored 20 °C D2O added

600 MHz/ 1H

Sera separated by centrifugation (no details) within 2h after collection, stored 80 °C D2O added

600 MHz/ 1H

Sera separated by centrifugation (no details) within 2h after collection, stored 80 °C D2O saline (with formate) added, filtered by centrifugation (5 min 5 °C)

600 MHz microflow probe/1H

Blood (nonfasting) into heparinized tubes, plasma separated by centrifugation (no details), stored 80 °C Lipids extraction, stored 80 °C, extracts reconstituted in CDCl3/MeOD

600 MHz cryoprobe/ 1H



 Oral (OSCC) [58]

OSCC n 15 Healthy n 10

64 (46–84) 57 (40–77)

5m,10f 6m,4f

 

[59]

Ovarian (EOC) [61]

[62]

Pancreatic [64]

OSCC n 33 OLK n 5 Healthy n 28

55 (24–80) 55 (44–61) 49 (39–69)

18m,15f 2m,3f 21m,7f



EOC n 38 Benign ovarian disease n 12 Healthy premenopausal n 19 Healthy postmenopausal n 32

61 50 28 57

(46–86) (22–68) (22–44) (51–69)

f



First set EOC n 120 Healthy n 132 Second set EOC n 50 Healthy n 50 Third set RCC n 30

49 (25–67)

f

49 (27–63)

Pancreatic cancer n 91 Healthy n 90

60.3 ± 10.6 60.5 ± 11.5





 

66 (34–92)

-

 

[65] Prostate (PCa) [69]

NMR field/ nucleus

Pancreatic cancer n 56 Benign disease n 43

65 ± 10 58 ± 15

31m,25f 21m,22f



Blood collected (mostly nonfasting) before surgery (after anaesthesia), sera separated within 6h RT, stored 20 °C

600 MHz/ 1H

PCa Gleason 5 n 20 PCa Gleason 7 n 22 Benign hyperplasia n 14

61.3 62.7 69.5

m



Blood clotting 30 min, centrifuged (3000 rpm, 15 min, 20 °C), sera stored 80 °C D2O/TSP added

500 MHz/ 1H



I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

Lung [55]

Control-patient matching

Thyroid [71]

31

Hypothyroid-cancer n 16 Hypothyroid-remission n 17 Euthyroid-remission n 14 Healthy n 23

41 ± 18 46 ± 16 43 ± 13 30 ± 10

9m,7f 10m,7f 10m,4f 9m,14f

 Blood (fasting) into EDTA tubes, centrifuged (3000g, 10 min), standard solutionb added, stored 10 °C

300 MHz/

Bladder cancer patients n33 Healthy n 37 Urinary tract infection (UTI) n 31 Bladder stone (BS) n 2

45 (20–70) 35 (20–50) UTI+BS 33 (18–48)

m 15m,22f

 Urine (fasting, vegetarian diet 2 days before collection) into vials with NaN3, snap-frozen, stored 80 °C  Centrifuged (12000 rpm, 10 min, 4 °C)

400 MHz/ 1H

Breast and Ovarian [34] Breast cancer n 48 EOC n 50 Healthy n 62

56 (30–86) 56 (21–83) 56 (19–83)

f

 Urine frozen -20 °C within 1h after collection, stored  DSS, NaN3 added, pH 6.8 ± 0.1

600 MHz/ 1H

Colorectal and Lung [43] Patients with muscle loss n 44 Patients with muscle gain n 29

62 ± 10 64 ± 11

29m,15f 14m,15f

 Urine collected randomly, NaN3 added, stored  DSS, NaN3 added, pH 6.75 ± 0.05

HCC n 18 LC n 10 Healthy n 14

46 (25–85) 37 (23–62) 37 (27–80)

12m,6f 8m,2f 7m,7f

 Urine collected randomly, centrifuged (13000g, 10 min), stored 80 °C  Phosphate buffer pH 7.4, TSP/D2O added, centrifuged (13000g, 10 min)

500 MHz/ 1H

HCC n 16 LC n 14 Healthy n 17

51.5 54 41

15m,1f 11m,3f 9m,8f

 Urine collected randomly, stored 80 °C  Phosphate buffer pH 7.4, TSP/D2O added, centrifuged (13000g, 10 min)

500 MHz/ 1H

Lung cancer n 71 Healthy n 54

64 (43–85) 42 (25–57)

49m,22f 23m,31f

 Urine (fasting) stored 80 °C  Centrifuged (8000 rpm, 5 min), phosphate buffer, TSP, NaN3 added, pH 7.00 ± 0.02, centrifuged (8000 rpm, 5 min)

500 MHz/ 1H

Biliary tract cancer n 17 Benign disease n 21

70.4 ± 10.6 59.4 ± 15.5

14m,3f 12m,9f

 Bile (various collection procedures) frozen 80 °C, freeze-dried in vacuum  Dry samples solubilised in D2O/CD3OD with sodium phosphate pH 6.0, centrifuged (no details), TSP added

500 MHz/ 1H

HCC n 11 Cholangiocarcinoma n 7 Non-malignant liver disease n 9 Non-liver disease n 17

56.8 52.7 47 44

7m,4f 5m,2f 5m,4f 4m,13f

 Bile collected at surgery, into vials with NaN3, stored 80 °C  Aqueous samples: bile diluted with water, pH 6.0 ± 0.5, TSP/D2O (capillary)  Organic samples: bile mixed with DMSO, TSP/D2O (capillary)

500 MHz/ 1H

EOC n 12 Benign cysts n 28

–

f

 Cyst fluid collected during laparoscopy or from removed cystic ovaries, centrifuged (3000g, 5 min, within 30–60 min), stored 35 °C  Deproteinized by centrifugation (3000g, 2h, over 10 kDa filter), freeze-dried, redissolved in D2O (pD 2.5 ± 0.10)

600 MHz/ 1H

55 (46–64) 47 35 (23–60)

m

 Seminal fluid and prostatic secretions frozen and stored -20 °C  DSS/D2O added

500 MHz/ 1H

P

Urine Bladder [32]

[53]

Lung [57]

80 °C

600 MHz/ 1H

Bile Biliary tract [30]

Liver (HCC) [54]

Ovarian fluid Ovarian (EOC) [60]

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

Liver (HCC) [52]

80 °C

Seminal fluid/prostatic secretions Prostate (PCa) [66]

PCa n 2 Benign hyperplasia n 1 Healthy, n 3

57

(continued on next page)

58

Table 2 (continued) Subject groups

[ref. no.] [67]

Sample collection, storage and preparation

NMR field/ nucleus

Gender

55 ± 6 59 ± 2 63 ± 3 67 ± 5

m

 Seminal fluid and prostatic secretions frozen and stored  DSS/D2O added

58.0 ± 7.0 52.2 ± 12.1

m

 Prostatic secretions centrifuged (13000g, 5 min, 4 °C), stored days)  TMSP/D2O added, centrifuged (4000g, 5 min, 4 °C)

Brain cancer n 63 Neurological disorders n 27 Healthy n 20

–

–

 CSF stored 80 °C  CHCl3/MeOH lipids extraction, extracts re-dissolved in CDCl3

400 MHz/ 1H and

CRC n 21 Healthy n 11

63 (32–78) 58 (31–73)

8m,13f 2m,9f

 Stool collected before surgery/endoscopy, diluted with water, homogenized, freeze-thaw cycles, centrifuged (10000 rpm, 15 min, 4 °C), supernatants stored 80 °C  D2O added, centrifuged (10000 rpm, 10 min, 20 °C)

600 MHz/ 1H

CRC n 111 Normal (non-cancer) n 412

–

–

 Stool collected before surgery/colonoscopy, refrigerated (24–48h), stored 70 °C, PBS/D2O added after thawing, mixed and centrifuged (3200 rpm, 3 min), supernatants stored 70 °C  D2O/TSP added

400 MHz flow probe/ 1H

Seminal fluid Prostatic sec.

[68]

Control-patient matching Age

PCa n 21 Noncancer n 16 PCa n 7 Noncancer n 17

PCa n 52 Healthy n 26

20 °C

80 °C (up to 21

500 MHz/ 1H

500 MHz microprobe/1H

Cerebrospinal fluid (CSF) Brain [33]

31

P

Fecal extracts Colorectal (CRC) [41]

[42]

Abbreviations: AC adenocarcinoma, DMSO dimethyl sulfoxide, DSS 4,4-dimethyl-4-silapentane-1-sulfonic acid, EDTA ethylenediaminetetraacetic acid, M-/U-IGHV mutated/unmutated immunoglobulin heavy chain variable region, OLK oral leucoplakia, PBS phosphate buffer saline, PL phospholipids, RT room temperature, SCC squamous cell carcinoma, TSP or TMSP 3-(trimethylsilyl)-propionic acid. a Standard solution: sodium cholate (232 mmol/l), Na2EDTA (26.87 mmol/l), NaOH (25 mmol/l), N-(phosphonomethyl)-glycine (5.12 mmol/l) in D2O. b Standard solution: glycerophosphate (5 mmol/l), sodium cholate (430 mmol/l), Na2EDTA (30 mmol/l), NaOH (24 mmol/l).

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

Cancer type

Table 3 Multivariate analysis (MVA) methods and outputs reported in NMR cancer-related studies of human biofluids. Cancer Type

Data pre-treatment

MVA methods and validation

Summary of MVA outputs

Biofluid [ref. no.]

R2Y

Q2

%Sens.

%Spec.

0.95

0.91

88

81

0.75 0.77

0.57 0.60

100 98

93 99

AUC

 Bucketing 0.04 ppm  Normalization to TA and then to TSP

 OPLS-DA  Y-scrambling validation (200 permutations)  Prediction by LOV analysis

Cancer vs benign OPLS-DA (1 + 4 LVs)

Breast/Ovarian Urine [34]

 67 Metabolites quantified  Probabilistic quotient normalization  Mean-centering and UV scaling

 PCA, PLS-DA and OPLS-DA of log10-transformed normalized concentrations  Permutation analysis  OPLS-DA prediction in a test set

Cancer vs healthy OPLS-DA (breast) OPLS-DA (ovarian)

Breast Serum [36]

 Bucketing 0.02 ppm  Normalization to TA  Mean-centering

 PCA, (OSC)-PLS-DA of DART-MS, NMR  (OSC)-PLS-DA of DART-MS against scores from PCAa/PLS-DAb of NMR  4-Fold cross-model validation

Cancer vs healthy PLS-DA (NMR) OSC-PLS-DA (MS/NMR)

81.5 88.9a 96.3b

83.3 96.7a 93.3b

0.83 0.99a

Breast Serum [37]

 Integration of selected metabolite signals

 Logistic regression for marker selection  PLS-DA on 11 marker metabolites  aPLS-DA with LOV CV; bPLS-DA of training set, validated in independent set

Recurrent vs no evidence of disease PLS-DAa PLS-DAb

86 78

86 85

0.88 0.84

Breast Serum [38]

 Not stated

Early vs metastatic disease OPLS

75

69

Breast Serum [39]

 Bucketing 0.01 ppm  Mean-centering

 PLS-DA  7-Fold internal CV  Permutation testing

Weight gain vs no gain PLS-DA (2 LVs) 0.235 (P < 0.002 by random permutation)

Colorectal (CRC) Serum [40]

 1D CPMG spectra log-transformed, mean-centered  Selected signals in 1D slices of Hadamard-TOCSY spectra integrated and median-centered

 PCA applied to: 1D CPMG spectra; selected signals from Hadamard-TOCSY spectra

CRC vs non-cancer PCA (Hadamard-encoded data)

70

95

Colorectal (CRC) Fecal extracts [41]

   

 OSC-PCA

Not stated

Colorectal (CRC) Fecal extracts [42]

 Normalization of magnitude spectra to unit area  First derivative

 Feature selection by a genetic algorithm approach employing LDA with LOV CV

CRC vs normal Training set Test set

87.5 85.2

86.9 86.9

Colorectal/Lung Urine [43]

 71 Metabolites quantified (63 for further analysis)  Normalization to creatinine/ TA/probability quotient normalization

 8 different statistical approaches, including PLS-DA and SVM  5-Fold CV  Permutation testing

Muscle loss vs no loss PLS-DA SVM

Normalization to TA Bucketing 0.01 ppm Log-transformation Mean-centering

   

Data reduction by OPLS SVM applied to OPLS scores Double CV scheme to assess predictive ability Independent validation set tested

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

Biliary tract Bile [30]

0.932 0.923

Average accuracy 76.7 82.2 (continued on next page) 59

60

Table 3 (continued) Cancer Type

Data pre-treatment

MVA methods and validation

Summary of MVA outputs R2Y

Biofluid [ref. no.]  Normalization (median fold change to median spectrum)  Variable size bucketing  UV scaling

 PLS-DA  Permutation testing

Esophageal (EAC) Serum [45]

 Bucketing 1.5 Hz  Normalization to TSP  Auto-scaling

   

 Bucketing 0.04 ppm  Normalization to TA  Mean-centering

 PCA

Kidney (RCC) Serum [47]

Kidney (RCC) Plasma [48]

Leukaemia (CLL) Serum [50]

Liver (HCC) Serum [51]

Liver (HCC) Urine [52]

Liver (HCC) Urine [53]

 Bucketing 0.04 ppm  Normalization to TA  Centering and Pareto scaling

 Variable-sized bucketing of 109 selected signals  Normalization to TA  UV scaling

 Bucketing 0.04 ppm  Normalization to TA

 Intelligent bucketing (0.02±0.1 ppm)  Normalization to TA  Mean-centering

 Intelligent bucketing (0.02±0.1 ppm)  Normalization to TA  Mean-centering

PLS applied to bucketed spectra Univariate selection of variables PLS-DA applied to 8 metabolites, LOV CV Visual predictions: Y-predicted scatter plot

 (OSC)-PCA, (OSC)-PLS-DA

 PCA, PLS-DA  Permutation testing

 PCA, PLS-DA

 PCA, PLS-DA (with OSC)  LOV CVa  External validationb (training set 70% samples + independent set)

 PCA, PLS-DA (with OSC)  LOV CVa  External validationb (training set 70% samples + independent set)

%Sens.

%Spec.

AUC

88 82

82 92

0.875 0.888

Grade 0 (no toxicity) vs grade 3 (severe toxicity) PLS-DA (2 LVs) 0.92 0.42 (P = 0.098 by random permutation)

EAC vs healthy PLS-DA Training set Test set

RCC vs healthy PCA T1a RCC vs healthy PCA

T1 RCC vs controls PLS-DA T2 RCC vs controls PLS-DA T3 RCC vs controls PLS-DA T1 vs T3 RCC PLS-DA

0.792

0.636

0.865

0.766

0.96

0.90

0.98

0.70

0.91

0.84

0.98

0.60

CLL vs healthy PLS-DA 0.80 UM-IGHV vs M-IGHV PLS-DA 0.80 (P < 0.025 by permutation test)

HCC vs healthy PLS-DA Liver cirrhosis vs healthy PLS-DA HCC vs liver cirrhosis PLS-DA

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

Colorectal (CRC) Serum [44]

Q2

0.50 0.30

0.52

0.80

0.51

0.88

0.35

0.46

0.78a

0.72a

100a 100b

93.3a 87.5b

0.40a

0.62a

89.5a 62.5b

88.9a 100b

HCC vs healthy PLS-DA

0.80a

0.74a

100a 100b

94a 93b

HCC vs liver cirrhosis PLS-DA

0.54a

0.25a

81a 75b

71a 67b

HCC vs healthy PLS-DA (1+3 LVs) HCC vs liver cirrhosis PLS-DA (1+3 LVs)

 21 spectral regions (S/N>4) in tissue and serum spectra selected

   

PCA Linear regression analysis applied to PC’s Canonical correlation analysis (CCA) Nominal logistic regression

Adenocarcinoma vs squamous cell carcinoma CCA tissue CCA serum

Lung Plasma [56]

 No bucketing  Normalization to TA  UV scaling

  

PCA, PLS-DA, OPLS-DA 7-Fold internal CVa MCCVb (500 iterations), permutation testing

Cancer vs healthy PLS-DA (2 LVs)

0.73a

0.57a 0.64b

Lung Urine [57]

 Variable-sized bucketing  Normalization to TA  UV scaling

  

PCA, PLS-DA, OPLS-DA 7-Fold internal CVa MCCVb (500 iterations), permutation testing

Cancer vs healthy PLS-DA (3 LVs)

0.89a

0.68a 0.74b

Oral (OSCC) Serum [58]

 Bucketing 0.005 ppm  Generalized log-transformation

 

PCA, PLS-DA (1D spectra, 1D J-res projections) CV using the ‘‘Venetian blind’’ algorithm

OSCC vs healthy PLS-DA

Oral (OSCC) Plasma [59]

 Bucketing 0.04 ppm  Normalization to unit area  Mean-centering

 

PCA, PLS-DA (training and test sets) Permutation testing

OSCC vs healthy PLS-DA (training set)

Ovarian (EOC) Serum [61]

 Bucketing 0.04 ppm  Normalization to TA  Mean-centering and Pareto scaling

 

PCA, SIMCA Univariate ROC analysis for each of 219 regions

Ovarian (EOC) Serum [62]

 Bucketing 0.004 ppm  Normalization to TA  Mean-centering and Pareto scaling

 

Pancreatic Plasma extracts [64]

 Intelligent bucketing (0.02-0.06 ppm)

0.96 0.89

91.5b

89.2

93.2b

94.0b

>95

>95

EOC vs healthy ROC (2 NMR regions)

100

100

EOC (stage I/II) vs healthy Logistic regression (‘joint model’)

63a 68b

80a 95b



PCA Logistic regression on PCs (standard 1D, CPMG and ‘joint model’) Prediction on test seta and independent setb

  

PCA, PLS-DF Leave-25%-out CV Permutation testing

Cancer vs healthy PLS-DF (5-bin model) Training set Test set

97.5 96.3

94.1 93.0

0.71

b

0.63

0.80a 0.95b

Pancreatic Serum [65]

 Metabolites quantified, normalized to total concentration  Log-transformation, centering and scaling  Metabolites selection (t-tests)

 

PCA, OPLS-DA 7-Fold CV

Cancer vs benign disease OPLS-DA

0.837

Prostate (PCa) Serum [69]

 Bucketing 0.04 ppm  Normalization to TA  Mean-centering and Pareto scaling

  

PCA Random Forests 10-Fold CV

PCa vs benign prostate hyperplasia Random Forests

0.876

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

Lung Serum/tissue [55]

Note that a and b are used for cross referencing within one same table row (or study). Abbreviations: AUC area under curve, CV cross validation, LOV leave-one-out, MCCV Monte Carlo cross validation, OPLS orthogonal projection to latent structures, OSC orthogonal signal correction, PCA principal component analysis, PLS-DA partial least squares discriminant analysis, PLS-DF partial least squares discriminant function, ROC receiver operating characteristic, Sens. sensitivity, Spec. specificity, S/N signal-to-noise, SVM support vector machines, TA total area, TSP 3-(trimethylsilyl)-propionic acid, UV unit variance.

61

62

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

Table 4 Main cancer-related metabolic findings unveiled by NMR of human biofluids and proposed perturbed pathways. Metabolic changes in cancer vs controlsa

Proposed metabolic pathway perturbations

(+) citrate Changes in bile acids (BA)

 High citrate: decreased TCA cycle activity; expected to affect acetyl CoA concentration (CHOL precursor) and hence BA biosynthesis.

(+) taurine (-) citrate, hippurate

 High taurine: role against cell damage, assisting unhindered tumour proliferation.  Low citrate: active citrate uptake from extra-cellular medium (urine) into tumour cells to sustain lipid synthesis for tumour proliferation.

Brain Serum/CSF lipid extracts [33]

Serum:(+) CHOL, PL; (-) CHOLest CSF: CHOL, PL, CHOLest only in patients

 Increased CHOL and PL: higher utilization in brain tumour tissue, leading to their enhanced synthesis in liver and, hence, increased levels in blood.  Decreased CHOLest: higher transportation rate from blood to cancerous tissue.

Breast Urine [34]

(-) 26 metabolites

 Decreases in TCA cycle intermediates: suppressed TCA cycle.  Decreases in glc and amino acids: enhanced use by tumours, resulting in overall energy metabolism decrease, hindering other pathways e.g. urea cycle (low urea, creatine) and affecting gut microflora.

Serum [36]

Changes in glc, lactate, taurine, others

 No interpretation proposed.

Serum [37]

Recurrent vs no evidence of disease: changes in formate, His, Pro, cho, 3-hydroxybutyrate, lactate, Tyr

 Changes in several pathways e.g. amino acid metabolism, glycolysis, phospholipids and fatty acid metabolisms.

Serum [38]

Metastatic vs early disease: (+) N-acetylcysteine, glc, Lys, Phe, Pro; (-) lipids

 No interpretation proposed.

Serum [39]

Weight gain vs no gain: (+) several metabolites (especially lactate)

 Metabolic profile of patients with weight gain consistent with excess of energy expenditure relatively to oxidative capacity.

(+) acetate, acetoacetate, 3-hydroxybutyrate, lactate, pyruvate

 No interpretation proposed.

Fecal water extracts [41]

(+) Cys, Pro (-) acetate, butyrate

 Increased amino acids: altered production/function of mucins in colonic epithelium.  Decreased short chain fatty acids: protective role against colorectal cancer.

Fecal water extracts [42]

Changes in Glu, Ile, Val, n-butyric acid, lipids

 No interpretation proposed.

Urine [43]

Several metabolites related to muscle loss (colorectal and lung cancers)

 Creatine, creatinine, 3-hydroxyisovalerate: muscle catabolism.  Increased flux of amino acids and amino acid carbon through intermediary metabolism (succinate, trans-aconitate) and 1-carbon metabolism (betaine, trigonelline).  Increased glc: early sign of insulin resistance.

Serum [44]

Grade 3 (severe toxicity) vs grade 0 (no toxicity): (+) LDL-like lipids

 Increased serum lipids: inflammation proposed as possible cause.  Subtle differences in lipid-based metabolism may affect drug-protein binding, resulting in altered clearance of capecitabine (or its metabolites).

(+) citrate, creatine, glc, Gln, 3-hydroxybutyrate, lactate, Lys

 Glc and lactate accumulation: high energy demand of tumour malignancy.  Increased 3-hydroxybutyrate: ketogenesis due to exceeding acetyl-CoA (Cori cycle not able to convert abundant lactate back into glc).  High levels of creatine, Gln, Lys: consistent with effects on TCA cycle and lactate accumulation.

(-) LPC1, LPC2, total PL

 Decrease in LPCs: changed enzyme activities, e.g. inhibition of enzymes converting PC into LPC; disturbance of LPCs catabolism.

(+) acetate, N-acetylglycoproteins, Ala, glycerol, Ile, lactate, LDL, VLDL, Leu, pyruvate, Val (-) acetoacetate, glc, Gln, PCho/cho

 Increased lactate and decreased glc: active glycolysis.  Increased pyruvate: reduced utilization in TCA cycle.  Increased branched chain amino acids: reduced utilization of succinyl-CoA synthesis due to impaired TCA cycle.  Increased glycerol: increased peroxisome b-oxidation activity, in turn activating gluconeogenesis-related conversion from triglycerides.  Decreased Gln: impaired synthesis, due to defective a-ketoglutarate transport from mitochondria into cytosol.  Decreased PCho/cho: increased PL metabolism (rapidly replicating cancer cells).

Cancer Type Biofluid [ref. no.] Biliary tract Bile [30] Bladder Urine [32]

Colorectal Serum [40]

Esophageal Serum [45]

Kidney Plasma [46]

Serum [47]

63

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74 Table 4 (continued) Metabolic changes in cancer vs controlsa

Proposed metabolic pathway perturbations

(+) cho, Glu, Ile, Leu, Val (-) acetate, acetoacetate, Gln, lipids (LDL/VLDL)

 Increased branched chain amino acids: increased muscle protein breakdown.  Decreased Gln and increased Glu: enhanced glutaminase activity; role of Gln as a substrate for nucleotide synthesis and as a respiratory fuel.  Decreased LDL/VLDL: abnormal CHOL metabolism.  Decrease of acetate and acetoacetate: lesser catabolism of lipids (increased utilization in cell proliferation).

(-) PC + PCLAS, LPC, SM, PI+PE

 Decreased PL: increased PL breakdown; PL utilization by rapidly proliferating cells.

(+) Glu, Pro, pyridoxine, pyruvate (-) Ile UM- vs M-IGHV: (+) CHOL, VLDL+LDL, lactate, fumarate, uridine; (-) pyridoxine, glycerol, 3hydroxybutyrate, Met

 Increased pyruvate and glutamate: increased pyruvate kinase type M2 activity; decreased pyruvate transaminase activity; thiamine deficiency (as thiamine pyrophosphate is coenzyme in pyruvate decarboxylation).

(+) acetate, N-acetylglycoproteins, Gln, glycerol, aketoglutarate, 1-methylhistidine, Phe, pyruvate, Tyr (-) acetoacetate, cho, LDL, VLDL, Val

 Increased glutamine: accumulation of a-ketoglutarate, fluxed out of mitochondria and converted into glutamine in cytosol.  Increased pyruvate: lesser utilization in TCA or increased anaerobic cell respiration.  Increased acetate and reduced lipid: enhanced lipid metabolism.  Decreased acetoacetate: impairment of TCA cycle and energy metabolism in liver mitochondria.  Reduced amino acid levels: reduced translation from succinyl-CoA because of impaired TCA cycle.

Urine [52]

(+) carnitine, creatine (-) acetone, creatinine

 High carnitine: tumour overproduction to fuel mitochondrial activity and maintain rapid growth.  Increased creatine: role in energy transfer (increased in rapidly growing cells).  Decreased creatinine: reduced muscle mass.  Decreased acetone: altered lipid metabolism, reflecting increased cell turnover.

Urine [53]

(+) carnitine, creatine (-) citrate, creatinine, Gly, hippurate, TMAO

    

Bile [54]

Liver diseases (including HCC) vs non-liver disease (controls): (-) total BA, Gly-conjugated BA, CHOL, PL HCC vs non-malignant liver disease: (+) total BA, Glyconjugated BA, CHOL, PL; (-) urea

 Decrease of bile metabolites in liver diseases: impairment of bile synthesis and aberrant enterohepatic circulation (related to bile duct obstruction).

Cancer Type Biofluid [ref. no.] Plasma [48]

Leukaemia Plasma extracts [49] Serum [50]

Liver Serum [51]

Lung Serum/tissue [55]

As in [52] and additionally: Decreased Gly: role in chromosomal methylation. Low TMAO: derived mainly from diet; suppressed gut microbial activity Low hippurate: decreased hepatic function (less efficient benzoate conjugation) Decreased citrate: physiological stress and increased energy demand.

Changes in several spectral regions

 No interpretation proposed.

Plasma [56]

(+) lactate, VLDL+LDL, pyruvate (-) acetate, Ala, citrate, formate, glc, Gln, HDL, His, methanol, Tyr, Val

 Increased lactate and decreased glc, together with increased pyruvate and decreased citrate: increased glycolytic activity and reduced TCA cycle activity.  Decrease in Ala and other amino acids: utilization as major gluconeogenic precursors.  Reduced glutamine: increased glutaminolysis to sustain TCA cycle and energy regeneration and to provide precursors for nucleic acid synthesis.  Decreased acetate: reduced lipid catabolism to sustain accelerated cell proliferation.  Decreased HDL: increased uptake of circulating lipoproteins to supply CHOL for membrane build-up by fast-growing tumor cells.

Urine [57]

(+) N-acetylglutamine, citrate, creatinine, 3hydroxyisobutyrate, 3-hydroxyisovalerate (-) hippurate, trigonellinamide, trigonelline

 Decreased hippurate: effects on intestinal microflora; diet influence not excluded.  Decreased trigonelline and trigonellinamide: impairment of nicotinate and nicotinamide metabolism.

(+) acetoacetate, acetone, betaine, cho, glc, 2hydroxybutyrate, 3-hydroxybutyrate, pyruvate (-) acetate, Ala, citrate, ethanol, formaldehyde, formic acid, lactate, methanol, Phe, Tyr, Val

 High glc: possible unique behaviour of oral cancer; interference with insulin ability to modulate glc uptake.  Increased ketone bodies: lipolysis (as backup for energy production).  Decreased citrate, succinate and formate: suppressed TCA cycle.  Increased 2-hydroxybutyrate: protein and amino acid catabolism (also supported by vastly altered amino acid profiles).

(+) creatinine, glc Changes in myo-inositol, taurine, chocontaining compounds, etc.

 Altered myo-inositol and taurine: role in osmoregulation and tumour volume regulation.

Oral Serum [58]

Plasma [59]

(continued on next page)

64

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

Table 4 (continued) Cancer Type Biofluid [ref. no.]

Metabolic changes in cancer vs controlsa

Proposed metabolic pathway perturbations  Altered cho-containing compounds: membrane synthesis/degradation (apoptotic activity in cancer).  Increased glc: decreased glc anabolism in tumour tissue.

Ovarian Ovarian cyst fluid

(+) cho, Gln, Ile, Val, lactate, Lys, Met, Thr

 High lactate and Ala (with low glc): preference for anaerobic metabolism.

[60]

(-) glc

 High cho: increased cell membrane synthesis for proliferation.

Serum [61]

(+) 3-hydroxybutyrate

 No interpretation proposed.

Serum [62]

(+) acetoacetate, acetone, 3-hydroxybutyrate (-) Ala, cho of PL, creatine/creatinine, LDL, VLDL, unsaturated lipid, Val

 Increased ketone bodies: lipolysis (alternative route for energy production).

Urine [34]

(-) 30 metabolites

 The same as proposed for breast cancer.

Changes in lipids (e.g. lower phosphatidylinositols)

 Altered pancreatic function (altered levels of pancreatic enzymes important in fat digestion e.g. lipase, phospholipase A and cholesterol esterase).

22 metabolites found to be altered in cancer vs benign disease, e.g.: (+) acetone, glc, Glu, 3-hydroxybutyrate (-) creatine, Gln

 Increased acetone and 3-hydroxybutyrate (end products of ketogenesis): latent diabetogenic changes resulting from cancer.

(-) citrate

 Altered secretion characteristics in prostate.

Prostatic secretion [68]

(-) citrate, myo-inositol, spermine

 Decreased citrate: transformation to citrate-oxidizing cancer cells that have lost ability to accumulate zinc.  Decreased myo-inositol: altered osmoregulatory function.  Spermine: endogenous inhibitor of prostate cancer growth.

Serum [69]

(+) formate, Glu, lipids

 Increase in lipids: down-regulation of Apolipoproteins A-IV and A-I.

Remission vs cancer: (+) PE+SM, PC Hypo vs euthyroid (in remission) – hormonal status: (+) all PL except LPC1

 Increased PL metabolism (due to faster replication of cancer cells).

Pancreatic Plasma extracts [64] Serum [65]

Prostate Seminal fluid/ prostatic secretion [66,67]

Thyroid Serum [71]

Abbreviations: BA bile acids, Cho choline, CHOL cholesterol, CHOLest cholesterol esters, CSF cerebrospinal fluid, Glc glucose, HDL high density lipoprotein, LDL low density lipoprotein, LPC lysophosphatidylcholine, PC phosphatidylcholine, PCho phosphocholine, PCLAS acetylo-alkylo- phosphatidylcholine, PE phosphatidylethanolamine, PI phophatidylinositol, PL phospholipids, SM sphingomyelin, TCA tricarboxylic acids cycle, TMAO trimethylamine N-oxide, VLDL very low density lipoprotein. a (+) increased in cancer and (-) decreased in cancer relatively to control; Other comparisons indicated when pertinent.

acid (TSP) as internal reference, and found statistically different levels of citrate, taurine and hippurate in UBC samples compared to those from either benign disease or healthy controls. In particular, taurine was increased in the urine of cancer patients, possibly to assist unhindered tumor proliferation, given its role as a free radical scavenger and osmoregulator, while citrate and hippurate were decreased. Whereas the latter could relate to exogenous sources, the reduced citrate levels were suggested to arise from its active uptake from extra-cellular medium (urine) into the tumor cell, where it is used in lipogenesis to support rapid proliferation. 2.3. Brain cancer In order to assess the value of biofluid lipid composition in detecting brain malignancies and differentiating tumor subtypes or grades, 1H and 31P NMR spectroscopies have been employed to characterize lipid extracts of cerebrospinal fluid (CSF) and serum from patients with primary brain tumors and normal controls, complementary to tissue extracts analysis [33]. Increased levels of cholesterol and phospholipids in the sera of tumor patients (n = 32), especially those with higher grade gliomas, compared to

normal individuals (n = 10), correlated well with the observed increase in corresponding tumor tissues. A high turnover of membrane constituents (to support fast proliferation of tumor cells) was proposed to account for the enhanced synthesis of lipid components in liver and hence their increased concentrations in blood. The low levels of cholesterol esters in the sera of cancer patients were suggested to relate to the rapid transport of these compounds from blood to cancerous tissue. In regard to CSF extract analysis, whereas cholesterol, cholesterol esters and choline-containing phospholipids were absent in normal individuals (n = 20), as well as in patients with neurological disorders (meningitis, motor neuron disease and mitochondrial myopathies) (n = 27), they were present in the CSF of brain tumor patients (n = 63), increasing with tumor grade. These compounds were thus identified as possible specific markers for primary brain tumors. 2.4. Breast cancer Although mammography is widely used as a screening test for breast cancer, it has some disadvantages such as the relatively low sensitivity (<75%) and the discomfort caused to women. NMR urinary profiling has been explored as an alternative method

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

for breast cancer detection, in a study comprising urine samples from 48 women with breast cancer and 62 healthy subjects [34]. Multivariate analysis was applied to the log-transformed normalized concentrations of 67 metabolites. The OPLS-DA scores scatter plot of the first component against the orthogonal component (Fig. 3a) showed a good differentiation between the two classes, which was further validated through permutation testing (Fig. 3b). Within the metabolites quantified, 26 were found to be significantly reduced in the urine of cancer patients. The authors proposed that the decreased levels of several tricarboxylic acid (TCA) cycle intermediates were suggestive of a suppressed TCA cycle and, together with decreased glucose levels, could be indicative of the Warburg effect (increased glucose consumption and lactate production even in aerobic conditions) [35]. Moreover, decreases in circulating glucose and amino acids (due to enhanced use by tumors) were proposed to result in an overall decrease in energy metabolism, hindering other metabolic pathways such as the urea cycle (resulting in reduced urea and creatine) and potentially affecting gut microbial metabolism. In a subsequent study, Gu et al. have proposed the combined multivariate modeling of NMR and direct analysis in real time (DART)-MS data from blood serum to differentiate between women with breast cancer (n = 27) and healthy controls (n = 30) [36]. PLS regression with Orthogonal Signal Correction (OSC) of MS data against the score values resulting from either PCA or PLS-DA of the NMR data enabled class

Fig. 3. (a) Scores scatter plot of the first component (tPS[1]) against the orthogonal component (toPS[1]) corresponding to the OPLS-DA model based on 67 urinary metabolites from 38 breast cancer subjects (s) and 62 healthy female subjects (j) (Explained variance, R2 = 0.75; Predictive power, Q2 = 0.57), (b) Statistical validation of the corresponding PLS-DA model by permutation analysis. Adapted and reprinted by permission from the American Association for Cancer Research: [34].

65

differentiation with 82–96% sensitivity and 83–97% specificity levels, as assessed by 4-fold cross-model validation. Although metabolite identification was limited, this study illustrated the advantages of combining datasets obtained through different analytical methods and of expressing classification on a continuum rather than using a binary scale (disease vs no disease), as more usually observed. Disease recurrence following initial treatment is another critical issue in breast cancer, which would clearly benefit from more accurate estimation of risk (in order to allow better selection of patients for adjuvant therapy), as well as from improved follow-up surveillance (in order to allow the detection of recurrent disease at early stages). NMR and bidimensional Gas Chromatography coupled to Mass Spectrometry (GC  GC–MS) were used to characterize the metabolic profiles of over 250 retrospective serum samples from 20 recurrent breast cancer patients and 36 subjects showing no evidence of disease (NED), over a 6-year follow up period [37]. The 116 sera from recurrent patients were divided into samples collected more than 3 months before clinically determined recurrence (Pre), samples collected within 3 months before or after recurrence (Within) and samples collected more than 3 months

Fig. 4. (a) Receiver Operating Characteristic (ROC) curve generated from the 11marker PLS-DA model (built using Post and Within recurrence vs NED serum samples) and the performance of the cancer antigen CA 27.29 on the same samples; (b) Percentage of recurrence patients correctly identified using the 11-marker model (red squares) and CA 27.29 (blue triangles) as a function of time using a cutoff threshold of 48. Adapted and reprinted by permission from the American Association for Cancer Research: [37]. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

66

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

after recurrence diagnosis (Post). Making use of logistic regression of NMR and GC  GC–MS detected metabolites, a set of 11 markers (7 from NMR and 4 from MS) was selected for PLS-DA modeling. The Receiver Operating Characteristic (ROC) curve for the predictive model derived from PLS-DA analysis using Post and Within vs NED samples showed relatively high area under curve (AUC 0.88), sensitivity (86%) and specificity (84%), outperforming the cancer antigen CA 27.29, a common tumor molecular marker (Fig. 4a). Moreover, when evaluating the power of the model for early recurrence detection, it was found that, notably, more than half of the patients could be correctly predicted to have recurrence 13 months before clinical diagnosis (Fig. 4b). As stated by the authors, these results unveil a possible significant improvement for early detection and potentially better treatment options for recurring patients. The performance of the model with respect to the initial stage of breast cancer, estrogen (ER) and progesterone (PR) receptor status and site of recurrence was also assessed. In another study focused on recurrence risk assessment, Oakman et al. have analyzed blood serum from breast cancer patients with early (n = 44) and metastatic disease (n = 51), and used the metabolic differentiation achieved between these classes to estimate the relapse risk of another group of early stage patients (n = 45) [38]. Preoperative early patients were discriminated from metastatic cases with an overall accuracy of 72%, with the presence of micrometastasis possibly accounting for the misclassification of some early patients. Based on the clustering of these two classes, the ‘metabolomic risk’ for each early stage cancer patient was measured as its inverse Eucledian distance from the barycenter of the metastatic population and showed reasonable agreement with disease stage prediction accuracy. However, some discrepancy was found between the ‘metabolomic risk’ and the risk assessed by Adjuvantionline (a well-established prognostic tool), with the former assigning more patients to low risk. In face of these results, the authors highlight the need to further investigate the prognostic ability of metabonomics by applying it to an enlarged cohort of patients with at least 10-year follow up data. Weight gain in women receiving chemotherapy for breast cancer is usually associated to a higher risk of recurrence. In a study correlating serum metabolic profiles to body mass measurements, several metabolites were found to be associated with weight gain, revealing a profile consistent with an excess of energy expenditure relative to oxidative capacity [39]. Furthermore, increased baseline lactate and alanine, together with high body fat, showed prognostic value for chemotherapy-related weight gain. 2.5. Colorectal cancer In face of the high incidence and high mortality of colorectal cancer (CRC), largely related to late detection, there has been a continuous search for minimally invasive cost-effective screening methods suitable for CRC early detection. Ludwig et al. have used 1 H NMR of blood serum to attempt the discrimination between CRC patients (n = 38), patients with adenomas (n = 8) and non-cancer controls (n = 19), i.e. individuals with indicative symptoms who turned out not to have CRC. By using Hadamard-encoded TOCSY spectra, based on eight preselected metabolites, an enhanced separation between sample groups in PCA was achieved compared to the multivariate analysis of 1D CPMG spectra. A clear anticorrelation was found between glucose and lactate, indicating a pronounced Warburg effect. Moreover, increased levels of acetate, acetoacetate, 3-hydroxybutyrate and pyruvate were observed in the serum of cancer patients relative to controls [40]. The NMR analysis of stool samples has also shown promise as an alternative screening method for CRC. In a study involving 32 subjects (21 CRC and 11 controls), Monleón et al. found lower levels of acetate and butyrate and higher levels of proline and cysteine in fecal extracts

of CRC patients [41]. Some of these findings were corroborated in a larger scale study involving more than 500 subjects (111 CRC, 412 controls), in which a statistical classification strategy, involving a genetic algorithm-driven feature selection approach, employing linear discriminant analysis with leave-one-out cross validation, led to 85–88% average sensitivity and specificity for both training and monitoring sets [42]. Cancer-related muscle depletion is associated with poor functional status, treatment toxicity and reduced life expectancy. In a study comprising colorectal and lung cancer patients, the metabolic signature for muscle mass loss has been investigated through NMR-based metabonomics of urine [43]. Eight different standard statistical and machine learning approaches, including PLS-DA and Support Vector Machines (SVM), were applied to the NMRdetermined concentrations of 63 urinary metabolites, in order to identify cancer patients likely to loose muscle mass. Of all algorithms tested, SVM provided the best accuracy (82.2%), as assessed by 5-fold cross-validation. Bivariate statistics showed that several metabolites were related to muscle loss, including muscle catabolic products (creatine, creatinine, 3-hydroxy-isovalerate) and a number of amino acids. The authors concluded that the rapid, accurate and minimally invasive test developed can detect a physiologically relevant rate of muscle loss and envisaged possible improvements through the use of more sensitive analytical methods, namely Mass Spectrometry, serial urine sampling and Computed Tomography (CT) image analysis over time, and the reduction of uncontrolled sources of variation. Recently, the potential of pharmacometabonomic profiling, based on the 1H NMR analysis of pre-treatment serum samples collected from CRC patients, to predict toxicity caused by the chemotherapeutic agent capecitabine has been assessed [44]. This study revealed a putative metabolic signature associated with capecitabine toxicity and, although the PLS-DA model built (with only 12 samples: 3 corresponding to no toxicity and 9 to severe toxicity) showed poor predictive ability, several resonances arising mainly from low density lipoproteins (LDL) presented statistically significant differences between patients, in extremes of toxicity. Hence, although the authors recognize the need to further validate their findings in a larger patient group, this work demonstrates the possible role of metabonomics in individualizing chemotherapy regimens to avoid toxic side effects. 2.6. Esophageal cancer In a study combining univariate integration and multivariate modeling of serum spectra, Zhang et al. attempted to highlight metabolic profiles discriminating esophageal adenocarcinoma (EAC) patients (n = 68) from healthy subjects (n = 34) and patients with Barret’s esophagus (BE) (n = 5) or high-grade dysplasia (HGD) (n = 11), which are widely regarded as precursors of EAC [45]. Eight metabolites (glutamine, b-hydroxybutyrate, citrate, lysine, creatinine, lactate, glucose and an unidentified compound) showed significantly increased levels in cancer patients relatively to controls and were used for PLS-DA modeling of these two sample classes. Receiver Operating Characteristic (ROC) curve analysis was used to assess the sensitivity and specificity of the PLS-DA model, resulting in values of AUC greater than 0.85 for both training and validation sets. In spite of the small sample numbers, HGD and BE conditions were also evaluated by the same model, the first being classified primarily as cancer samples and the latter presenting mixed classification. 2.7. Kidney cancer Renal cell carcinoma (RCC) is the most prevalent malignancy of the kidney, with a favorable prognosis if detected at an early stage,

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

when nephrectomy (kidney removal) can be performed. RCC diagnosis has been approached through NMR spectroscopy of blood serum/plasma. In an early work, the distribution of phospholipids (lysophosphatidylcholines), assessed by 31P NMR spectroscopy, was shown to deviate considerably in RCC patients (n = 29) relatively to healthy volunteers (n = 19), approaching normality for patients in remission, and to correlate to tumor stage as well as the presence of metastases [46]. Later on, by applying PCA to serum 1 H NMR profiles, Gao et al. showed good unsupervised separation between RCC patients (n = 74) and healthy subjects (n = 25), even when considering only early stage (T1a) disease [47]. The metabolic pattern characterizing RCC samples comprised alterations in lipids and several metabolites and was found to be somewhat reversed after nephrectomy. Furthermore, low and advanced grade RCC patients could be clearly distinguished, suggesting that the serum NMR profiles could be useful in patient stratification and prognosis. In another study applying PCA and PLS-DA to plasma 1 H NMR profiles, Zira et al. showed a clear discrimination between RCC patients at different disease stages (n = 32) and subjects with benign urologic disease (n = 13), and proposed a metabolic signature for cancer patients comprising mainly decreased levels of lipoproteins (LDL/VLDL), acetate, acetoacetate and glutamine, and increased levels of glutamate, branched chain amino acids and choline [48]. These results did not fully agree with those previously reported by Gao et al. [47], and the authors suggested that the discrepancy could be attributed to the different genetic background and metabolic phenotype of the subjects enrolled in each case (Asians vs Caucasians). In addition, the lack of specificity of the major biomarkers (lipoproteins and choline) was noted as a possible limitation of the method to monitor RCC. 2.8. Leukaemia In a study of lipid extracts by 31P NMR spectroscopy, the plasma phospholipids composition has been proposed to be altered in acute leukaemia (AL) patients (n = 30) compared to healthy subjects (n = 21) [49]. In particular, the concentrations of a number of phospholipids (phosphatidylcholine, plasmalogen of phosphatidylcholine, lysophosphatidylcholine, sphingomyelin, phosphatidylethanolamine and phosphatidylinositol) were found to be significantly reduced relatively to either controls or AL patients presenting complete remission. More recently, the 1H NMR metabolic profile of serum from patients with chronic lymphocytic leukaemia (CLL) at initial stages (n = 29) was shown to differ from that of healthy controls (n = 9), with increased pyruvate, glutamate, proline and pyridoxine and decreased isoleucine in the serum of CLL patients [50]. Furthermore, the potential for differentiating between patients with different prognosis, as classified by the mutational status of the immunoglobulin heavy chain variable region (IGHV) genes, has been assessed. The authors found that the patients with leukaemic cells carrying unmutated IGHV (UM-IGHV) genes, which relate to poor prognosis and response to chemotherapy, could be differentiated from M-IGHV subjects through multivariate modeling of their 1H NMR serum profiles. Among the metabolic differences highlighted through PLS-DA loadings inspection, the most significant, as confirmed by univariate analysis, were reported for cholesterol and lactate (elevated in UM-IGHV) and for methionine and pyridoxine (reduced in UM-IGHV). 2.9. Liver cancer A number of studies have been undertaken to search for metabolic markers of hepatocellular carcinoma (HCC), the most common primary malignant tumor of the liver, in different biofluids (blood serum, urine and bile). Given that HCC often develops as a complication of a previous liver disease, such as liver cirrhosis

67

(LC) and hepatitis B or C, differentiation of HCC from these conditions as well as from healthy controls has been assessed. Gao and colleagues have found several alterations in the sera of LC (n = 36) and HCC (n = 39) patients compared to healthy subjects (n = 63), reflecting impairment of the TCA cycle and altered lipid metabolism [51]. Moreover, the separation of LC and HCC samples in the PLS-DA scores plot led the authors to suggest that patients could be classified according to their serum profiles, although recognizing that long-term studies of larger numbers of patients would be required to confirm these results. The urinary 1H NMR profiles have also been explored for the diagnosis of HCC amongst Nigerian and Egyptian populations, both with high prevalence of hepatitis viruses and hence increased risk for HCC [52,53]. In both studies, HCC samples could be differentiated from controls with 100% sensitivity and 87–93% specificity, as determined by OSC-PLS-DA modeling of a training set, comprising 70% of randomly selected samples, and prediction of class membership in a test set with the remaining 30% of samples, a procedure which was repeated three times. The main urinary metabolites accounting for this discrimination within the Nigerian population (18 HCC, 10 LC and 14 healthy controls) were creatinine and acetone (reduced in HCC) together with creatine and carnitine (elevated in HCC) [52]. The study on the Egyptian population (16 HCC, 14 LC and 17 healthy controls) corroborated most of these findings and revealed further decreases in glycine, trimethylamine N-oxide, hippurate and citrate in HCC compared to healthy controls [53]. Similarly to what has been noted for serum [51], the key metabolites accounting for the differentiation between HCC and cirrhotic patients were the same as those explaining the HCC vs healthy controls discrimination, thus showing the need to further assess the specificity of HCC urinary profiles relative to other liver diseases. A 1H NMR study of gallbladder bile showed altered bile homeostasis among different liver diseases, including HCC (n = 11) and a different type of liver cancer, cholangiocarcinoma (CC) (n = 7), as well as non-malignant liver diseases (NMLD) (n = 9), and control subjects undergoing surgery for non-liver disease (NLD) (n = 17) [54]. In particular, concentrations of major bile metabolites, namely phospholipids, bile acids and cholesterol, were found to be significantly reduced in liver diseases compared to controls, the most striking decreases having been noted for CC and NMLD. In addition, HCC patients could be differentiated from NMLD based on the amounts of phospholipids, total bile acids, glycineconjugated bile acids and cholesterol, whereas differentiation from CC was weaker. Based on these results, the authors proposed the NMR analysis of bile as a simple and highly informative way of assessing liver status and exploring biomarkers for liver diseases. 2.10. Lung cancer Lung cancer is one of the most prevalent and fatal types of cancer, its poor prognosis being related to asymptomatic development and late detection. The feasibility of using NMR-based metabonomics of biofluids to develop new lung cancer screening methods and improve early diagnosis has recently been evaluated. As a first step to establish lung cancer serum profiles, Jordan et al. have investigated correlations between paired tissue and serum samples retrieved from patients with either lung squamous cell carcinoma (n = 5) or adenocarcinoma (n = 9), considering also a small group (n = 7) of sera from healthy volunteers [55]. All samples were analyzed by 1H HRMAS NMR and the spectral data subjected to PCA and canonical correlation analysis (CCA), involving selected PCs and the % volume of cancer in tissue samples. It was shown that, by applying to serum data the discriminant structure discovered from tissue analysis, the resulting differentiating power between cancer groups and between cancer and controls held statistical

68

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

significance. In a later study comprising a significantly larger group of subjects, PLS-DA modeling of plasma 1H NMR spectra allowed cancer patients (n = 85) to be differentiated from healthy subjects (n = 78) with sensitivity and specificity levels of about 90%, as determined by Monte Carlo cross validation (MCCV) [56]. The predictive ability of the model was further confirmed by permutation testing and illustrated by plotting the prediction results for the real (true classes assigned) and permuted models in the ROC space (Fig. 5a), accompanied by the distribution of Q2 values (expressing the predictive power) (Fig. 5b). Through the inspection of OPLS-DA loadings colored according to Variable Importance in the Projection (VIP), relatively lower HDL and higher VLDL + LDL could be detected in the patients’ plasma, together with increased lactate and pyruvate and decreased glucose, citrate, formate, acetate, several amino acids (alanine, glutamine, histidine, tyrosine, valine) and methanol (Fig. 5c). Notably, these changes were found to be present at initial disease stages and could be related to known cancer biochemical hallmarks, such as enhanced glycolysis, glutaminolysis and gluconeogenesis, together with suppressed TCA cycle and reduced lipid catabolism, thus supporting the hypothesis of a systemic metabolic signature for lung cancer. An equally good discrimination between lung cancer (n = 71) and healthy subjects (n = 54) was obtained by NMR profiling of urine [57], highlighting a number of consistently altered metabolites, such as hippurate and trigonelline (reduced in patients), and b-hydroxyisovalerate,

2.11. Oral cancer In relation to oral cancer, more specifically oral squamous cell carcinoma (OSCC), two accounts of NMR metabonomics studies have been reported on blood samples [58,59]. One of the studies considered the blood sera of OSCC patients (n = 15) and healthy donors (n = 10) and characterized them by 1D and 2D J-resolved spectra (500 MHz) as well as by 2D Correlation Spectroscopy (COSY) NMR at 800 MHz. PCA and PLS-DA of both 1D spectra and 1D projections of the J-resolved spectra indicated a clear distinction, not only between controls and OSCC patients, but also between early and late stages of the disease [58]. Cross validation of PLS-DA results showed significantly high sensitivity and specificity and the importance of these results was highlighted in view of the small size of the tumors (<0.005% of total body mass for earlier stages) and the fact that some patients showed no metastasis or only small locoregional metastasis. An interesting observation relied on the high levels of glucose and low levels of lactate in the sera of OSCC patients, suggesting a possible unique behavior of oral cancer,

1

(b) 40%

0.9

35%

True Classes

Permuted Classes

0.8

Frequency

TPR (sensitivity)

(a)

a-hydroxyisobutyrate, N-acetylglutamine and creatinine (elevated in patients relatively to controls). Both studies have assessed the influence of possible confounders (age, gender, smoking habits) and concluded on their lower importance compared to the presence of the disease.

0.7 0.6 0.5 0.4 0.3

True Classes

30% 25% 20% 15% 10%

0.2

5%

0.1

0%

0 0

0.2

0.4

0.6

0.8

-1

1

-0.7

-0.4

0.2

0.5

0.8

5e-08

3.31

Lipids (mainly LDL+VLDL)

0e+00 -5e-08 -1e-07

LV1 loadings

-2e-06 -1e-06 0e+00 1e-06 2e-06 3e-06 4e-06 5e-06

-0.1

Q 2 values

FPR (1-specificity)

(c)

Permuted Classes

Unsaturated lipids (mainly LDL+VLDL)

*

*

Lactate

Formate Histidine

Tyrosine

Glyceryl of lipids

1.66

Lipids 9.0

8.5

8.0

Increased in cancer Decreased in cancer

8

7.5

7.0

6.5

6.0

Pyruvate ppm

Valine Unsaturated lipids (mainly HDL) Glucose 6

Methanol Lipids (Choline) 4

Citrate Acetate Glutamine

Lysine+ Arginine 2

Lipids (mainly HDL) Alanine

0

0

ppm Fig. 5. Multivariate modeling of 500 MHz 1H CPMG NMR plasma spectra of healthy subjects (n = 78) and lung cancer patients (n = 85). (a) Receiver Operating Characteristic (ROC) space, where each point represents a prediction result (sensitivity and 1-specificity) of the confusion matrices obtained from MCCV (500 iterations) of the PLS-DA models: full symbols – prediction results in the original model (class membership correctly assigned), open symbols – prediction results in the permuted model (class membership randomly permuted); (b) distribution of Q2 values obtained by MCCV (500 iterations) of the same PLS-DA models (original and permuted); (c) OPLS-DA LV1 loadings plot colored as a function of Variable Importance in Projection (VIP). Assignment of main signals is indicated (unassigned signals with high VIP are marked with an asterisk).

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

interfering with insulin metabolism and glucose regulation and, subsequently, leading to altered energy metabolism. Other observations comprised elevated ketone bodies (lipolysis for energy production), decrease of several TCA cycle metabolites (suppressed TCA cycle), elevated 2-hydroxybutyrate, decreased ornithine and most amino acids (possible altered protein and amino acid metabolism) (Fig. 6). Regarding differentiation of early and late disease stages, the main metabolites observed to change were choline, betaine, dimethylglycine, carnitine and acetyl carnitine. This study also addressed possible confounding variables such as age and gender, suggesting their low relevance in the observed patient/control separation. A second study addressed the levels of metabolites in blood plasma of OSCC patients (n = 33), compared to subjects affected by oral leucoplakia (OLK), a pre-malignant condition involving lesions in the oral mucosa (n = 5), and controls (n = 28) [59]. PCA and PLS-DA were applied to the NMR data (standard 1D and CPMG) of plasma divided into a training set and two test sets (comprising samples also included in the training set). Although distinct metabolic signatures were apparent for both OSCC and OLK groups, the limitation posed by low sample numbers was recognized, as well as the importance of the subjects’ age. In this study, alterations were noted for myo-inositol and taurine (connected to osmoregulation and tumor volume regulation), choline-containing compounds (reflecting membrane synthesis and degradation, possibly indicative of apoptotic activity in OSCC), creatinine and glucose (found elevated and interpreted as a reflection of decreased glucose anabolism in cancer tissues). 2.12. Ovarian cancer Epithelial ovarian cancer (EOC) has been studied through NMR metabonomics applied to several biofluids, one initial study having

69

addressed ovarian cyst fluid [60] and more recent work considering blood serum [61,62] and urine [34], thus envisaging less invasive approaches to study this type of cancer. A 600 MHz 1H NMR study of 40 samples of ovarian cyst fluid, corresponding to 12 malignant and 28 benign tumors has been reported [60]. A total of 36 metabolites were identified, comprising some compounds not expected in this context: 5-oxoproline and N-acetylaspartic acid. Compound quantification was carried out by integration vs the TSP reference signal and significant differences in concentration were found between malignant and benign cysts for several metabolites. In particular, higher concentrations of isoleucine, valine, threonine, lactic acid, lysine, methionine, glutamine and choline were noted in malignant cyst fluids and the authors advanced the possibility of using selected spectral regions where these compounds resonate for in vivo spectroscopy of ovarian tumors. Regarding biochemical interpretation, the high lactic acid and alanine contents, together with the lower glucose content, suggested a preference for anaerobic metabolism whereas higher choline related to increased cell proliferation. Other changes were discussed and the authors acknowledged the possible importance of medication as a confounding variable. Regarding blood characterization, the sera of 38 women affected by EOC were characterized by 1H NMR, as well as samples from 12 patients with benign ovarian disease and 53 healthy (both pre- and post-menopausal) women [61]. The data was analyzed by PCA and Soft Independent Modeling of Class Analogy (SIMCA) using the Cooman’s plot to assess the classification performance [63]. This plot, where class distances are plotted against each other, demonstrated that sera from the three classes compared did not share multivariate space, providing validation for class separation. Furthermore, a 2-variable model (consisting of 1H NMR descriptors at d 2.77 and 2.04) was shown by ROC analysis to provide 100% sensitivity and specificity. A more recent study has built on these observations by considering a significantly enlarged subjects

Fig. 6. Schematic representation of the most relevant metabolic differences between oral cancer patients and healthy controls, as assessed by NMR analysis of blood serum. Reprinted with permission from: [58].

70

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

universe (a first set with n = 120 EOC and n = 132 age matched controls and a validation set with n = 50 for each group) and by focusing on early stages of the disease (I/II), in order to evaluate the potential of the approach in ovarian cancer screening [62]. Both standard 1D and CPMG spectra of serum, as well as concatenated data, were considered. A predictive model for early stage EOC was developed with the first set of samples through PCA and logistic regression on Principal Components. Independent validation was subsequently carried out with a second set of samples, the results showing good separation between early stage EOC and healthy women in PCA and yielding 95% specificity, 68% sensitivity and an AUC of 0.949. Furthermore, the predictive model together with an a priori probability of EOC (‘prevalence’ in a population) was used to calculate the a posteriori probability, p-EOC, of early stage disease, the results showing that patients clearly exhibited higher p-EOC values than healthy controls. In terms of altered metabolites, identified with STOCSY and 2D experiments, decreased levels of valine, alanine, creatinine/creatine, phospholipids and lipoproteins (mainly VLDL and LDL) and increased levels of ketone bodies (acetone, acetoacetate and b-hydroxybutyrate) were found in the serum of EOC patients compared to controls. Finally, the specificity of the results for EOC was discussed based on a comparison with a small set of sera samples from patients with RCC and the potential for a cancer-type specific screening test was advanced. The urine of ovarian cancer patients (n = 40) has also been characterized by 1H NMR, in comparison with healthy donors (n = 62) [34]. The study also included samples obtained from breast cancer patients, thus allowing specificity in regard to cancer type to be assessed. Application of PCA, PLS-DA and OPLS-DA, in tandem with sensitivity and specificity evaluation and significance testing through the Wilcoxon’s test, revealed that a distinct and validated urine metabolite signature seems to characterize ovarian cancer. The few but clear misclassifications noted were discussed based on patient age, cancer stage and other conditions. The majority of the metabolites excreted seemed to decrease in cancer patients, with decreased glucose and amino acids suggesting TCA cycle suppression. Based on the performance of the PLS models, the authors envisaged promising applications of this work as a potential screening tool. 2.13. Pancreatic cancer An initial NMR metabonomics study on pancreatic cancer has focused on plasma lipophilic extracts obtained from approximately 100 patients and 90 age-matched controls, in tandem with Liquid Chromatography coupled to Mass Spectrometry (LC-MS) profiling of the lipids [64]. PCA and partial least squares discriminant function (PLS-DF) were applied to the binned NMR spectra of the extracts. The PLS-DF models were built using forward regression on the bins (integrals of spectral regions ranging between 0.02 and 0.06 ppm in width), a bin being added to the model if it increased its F-score by 2.0. The subsets of spectral bins that gave a model its optimal performance were determined based on prediction accuracy, and the results were reported for models built with either 4 or 5 bins. High levels of sensitivity and specificity were attained for these models, as assessed by leave-25%-out cross validation, and permutation testing further validated the results. Complementary to NMR findings, MS profiling of phospholipids found three phosphatidylinositols at significantly lower levels in cancer patients than in controls. A more recent NMR study on blood sera aimed at exploring the possibilities of improving diagnostic accuracy by studying the metabolite profile differences between pancreatic cancer (n = 56) and benign hepatobiliary disease (n = 43), a lesion that mimics the former condition [65]. Close to 60 metabolites were identified and changes in the concentration

of many of these were evaluated, followed by OPLS-DA applied to the most significant metabolites. The differences observed suggested higher glutamate and glucose levels in the cancer group, along with lesser contents of glutamine and creatine. However, the authors identified the need to improve the specificity of the study, namely in regard to possible confounding factors such as age or other disorders e.g. jaundice and diabetes. 2.14. Prostate cancer Regarding prostate cancer, NMR studies of seminal fluid and/or prostatic secretions [66–68], and of blood serum [69] have been carried out. An initial study on human seminal fluid aimed at determining the variation in citrate concentration in relation to the development of prostate adenocarcinoma [66]. Samples comprised fluid from three control subjects with low prostate specific antigen (PSA) levels, one subject with benign prostatic hyperplasia and two patients with prostate cancer, as well as two additional samples of prostatic fluid. Once the effect of sample storage (at 20 °C) and thawing was determined not to influence the samples, quantification of citrate was carried out using 4,4-dimethyl-4-silapentane-1-sulfonic acid (DSS) as internal reference. Results showed a clear decrease of citrate accompanying the development of cancer, as viewed through PSA levels, thus opening up the possibility of rapid citrate measurements in seminal fluid as a screening method. The results were confirmed in a subsequent study [67], where citrate was again quantified by NMR, in both seminal fluid and expressed prostatic fluid, considering a total of 61 subjects. The performances of diagnostic tests were compared by ROC curve analysis, having established that citrate measurements in both fluids outperformed PSA measurements. In addition, a further study described a more thorough analysis of prostatic secretions as to their metabolic profile, along with absolute quantification of several metabolites [68]. The subject groups included 52 men with prostate cancer and 26 healthy donors. Quantitative 500 MHz 1H NMR spectra were recorded and analyzed by logistic regression (LG), confirming the importance of citrate and further revealing myo-inositol and spermine as good predictors for prostate cancer (their concentrations being inversely related to it). Furthermore, these variations seemed to be age-independent reinforcing their potential as prostate cancer markers. Following a previous study of a human tumor xenograft mouse model of prostate cancer [70], which included proteomic analysis of the samples and identified specific correlations between certain metabolites and proteins, a similar multi-omic approach was followed in a more recent study applied to human blood serum [69]. In this study of 14 patients with benign prostatic hyperplasia (BPH) and 42 patients with prostate cancer at different stages, the patients’ blood sera were analyzed not only by 1H NMR but also by 2D differential gel electrophoresis (DIGE), for evaluation of relevant proteins. PCA and Random Forest Classification were applied in order to unveil relevant features in both datasets. Analysis of NMR data revealed changed levels of lipids in cancer patients (suggested to relate to the lower levels of apolipoproteins Apo-IV and ApoA-I), together with increased levels of glutamate and formate. Furthermore, the predictive ability of 2D DIGE selected features was found to be significantly high, namely in comparison to the usual PSA marker, although the need for further validation using a larger sample universe was acknowledged. 2.15. Thyroid cancer A 31P NMR study of phospholipids in the blood plasma of 33 patients with thyroid carcinoma was carried out, in comparison to a group of 23 healthy volunteers [71]. By mixing the blood plasma samples with a standard solution containing glycerophosphate as

71

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

a reference compound, the major plasma phospholipids were quantified: phosphatidylethanolamine + sphingomyelin, 1- and 2acyl-lysophosphatidylcholine, phosphatidylinositol and phosphatidylcholine. The authors related the levels of these compounds with an increased rate of phospholipid metabolism and faster cell replication, in relation to thyroid cancer progression and remission status. However, an important interference between the disease and the specific hypothyroid hormonal status was acknowledged, rendering this an important factor in future studies. 3. Integration of published results Many of the above mentioned studies aimed at finding metabolic patterns able to discriminate between cancer and control subjects, envisaging the development of new methods for cancer screening and diagnosis. Interestingly, while some changes have been consistently reported across the literature, pointing to common altered pathways in cancer, the metabolic patterns unveiled show high variability between cancer types, and, in some cases, within the same cancer type. This variability is illustrated in Fig. 7, which presents an overview of NMR-detected metabolic alterations in the blood serum or plasma of cancer patients compared to healthy controls or subjects with benign conditions. Increased glycolytic activity, as reflected by reduced glucose [47,56] and elevated lactate and/or pyruvate levels [40,47,50,51,56], has often been associated with cancer, CRC [40]

EAC [45]

RCC [46]

RCC [47]

RCC [48]

corroborating the well known Warburg effect [35]. A few studies, however, have reported increased glucose levels, namely in the serum/plasma of patients with esophageal [45], oral [58,59] and pancreatic [65] cancers. Perturbations in the TCA cycle are another feature commonly observed. Citrate has been found to be reduced in the blood plasma/serum of lung and oral cancer patients [56,58], possibly arising from its increased utilization in macromolecule synthesis and/or from reduced TCA cycle activity, but has been reported to be increased in the serum of esophageal cancer subjects [45]. A large number of studies have found altered levels of several amino acids in cancer biofluids. However, the specific amino acids varying as well as the direction of their variation in cancer vs control are not always consistent between cancer types. For instance, whereas glutamine has been found to be reduced in the serum/ plasma of kidney [47,48], lung [56] and pancreatic [65] cancer patients, increased levels of this amino acid were reported for esophageal [45] and liver [51] cancers. Branched chain amino acids also showed discrepant variations, with decreased valine in liver [51], lung [56], oral [58] and ovarian [62] cancers and increased valine, leucine and isoleucine in kidney cancer [47,48]. Fluctuations in blood lipid levels have also been frequently associated with different cancer types, although, once more, the detected patterns of change are variable. Total phospholipids, for instance, have been found to be increased in subjects with brain [33] and thyroid [71] tumors, compared to healthy controls, and decreased in renal cell carcinoma patients [46]. Lipoproteins also have shown CLL [50]

HCC [51]

Lung OSCC OSCC EOC [56] [58] [59] [62]

Panc. [65]

PCa [69]

acetate acetoacetate acetone N-acetylglycoproteins alanine arginine asparagine betaine choline citrate creatine creatinine dimethylamine ethanol formaldehyde formate glucose glutamate glutamine glycerol HDL histidine 2-hydroxybutyrate 3-hydroxybutyrate isoleucine α-ketoglutarate lactate LDL, VLDL leucine lipids / phospholipids lysine mannose methanol 3-methyl-2-oxovalerate 1-methylhistidine phenylalanine proline 4-pyridoxate pyridoxine pyruvate threonine tryptophan tyrosine urea valine

Fig. 7. Overview of metabolites found to be increased (in dark grey) or decreased (in light grey) in cancer patients relatively to healthy controls or subjects with benign conditions, as assessed by direct NMR analysis of blood serum or plasma. CRC colorectal, EAC esophageal, RCC renal cell carcinoma, CLL chronic lymphocytic leukaemia, HCC hepatocellular carcinoma, EOC epithelial ovarian cancer, OSCC oral squamous cell carcinoma, Panc. pancreatic, PCa prostate. Reference numbers in square brackets.

72

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

discrepant variations, even within the same cancer type, as is the case of the LDL/VLDL variations in kidney cancer [47,48]. Other metabolites related to lipid metabolism, such as acetate and ketone bodies (acetoacetate, acetone and 3-hydroxybutyrate, deriving mainly from fatty acid oxidation), also showed varying patterns in blood for different cancer types (Fig. 7). On one hand, the observed variability may reflect the different biological behavior and adaptive mechanisms of different cancer types, which impart distinct, eventually specific, signatures to the body fluids analyzed. On the other hand, a number of possible influential factors that may hamper direct comparison of reported studies ought to be considered. The different genetic background and dietary/lifestyle habits of studied subjects may be one of such factors. Indeed, previous reports have highlighted the variability in the metabolic profile of biofluids (particularly urine) between populations of different continents (America and Asia) [19] and even of different countries within Europe (Britain and Sweden) [72]. It seems reasonable to propose that if the subjects show a distinct baseline profile in health, they may also respond differently to disease and therefore show discrepant metabolic patterns in their biofluids. This could be at the basis of some of the inconsistencies noted, for instance, in what concerns altered lipid metabolism in renal cell carcinoma patients of Chinese [47] and European [48] origins, or hepatocellular carcinoma in two African populations [52,53]. Moreover, due to the complex and heterogenous nature of cancer, great care should be taken when comparing results pertaining to the same cancer type but considering different histological sub-types or grades. Differences in sample history may also account importantly for discrepant metabolic results between inter-laboratory studies. Although general guidelines for preparing biofluids for NMR-based metabonomics have been published [73,74], a large variability in experimental protocols remains, as clearly seen in Table 2. The significance of systematic bias due to differential sample collection, handling and storage has been well documented for urine and blood plasma [75,76]. It was found that different pre-analytical conditions and steps, such as centrifugation speed, filtration, and addition of preservatives in the case of urine, or elapsed time between collection, processing and freezing in the case of blood derivatives (serum and plasma), have major influences on the metabolic profiles. For instance, glucose, lactate and pyruvate, often indicated as key markers of altered cancer metabolism, have been found to be very prone to variations as a function of time and temperature from blood collection through processing [76]. Therefore, possible pitfalls arising from non-reproduced sample collection and handling procedures across the universe of studies should be considered. Also, the conditions used for NMR analysis, namely temperature and acquisition parameters (e.g. relaxation delay, spin-echo time in the CPMG experiment), may influence spectral profiles and quantitative measurements and should, therefore, be taken into account when comparing different studies. Finally, the studies reported are also very heterogeneous in terms of the methodology followed for data pre-treatment and multivariate analysis (Table 3). For instance, in some studies, the variables are absolute concentrations of selected metabolites, whereas in others, binned (bucketed) spectral regions (with fixed or variable widths), or even full resolution data, are computed. Also, significant variability occurs in the normalization and scaling procedures employed. As highlighted in several studies [77–79], the ways in which these steps are performed may significantly affect data quality, both in terms of the classification power of multivariate models and the accuracy of univariate quantitative information. Furthermore, the comparative analysis of studies concerning the robustness of the models and the classification power is hampered by the different output parameters shown and the various methods through which they are obtained, as critically discussed by the Data

Analysis Working Group [80]. Therefore, at this stage, it is difficult to achieve solid conclusions regarding the consistency and specificity of metabolic findings, and we believe that there is still a long way to go before establishment of definitive clinically valuable signatures for different cancer types, as discussed below. 4. Progress into the clinic: needs and future directions The research conducted so far provides compelling evidence that NMR-based metabonomics of biofluids offers great promise for the minimally-invasive screening of cancer related perturbations, at several stages of disease management, ranging from early detection and differentiation from benign conditions to identification of metastatic disease and surveillance of recurrence. However, in spite of the encouraging results presented, many studies were exploratory in nature and the proposed cancer metabolic biomarkers have not yet progressed into the clinic. As discussed by others [81,82], the requisites for a potential biomarker to become a clinically approved test are vast and demanding and many factors related to study design and execution have previously contributed to prevent the delivery of validated and clinically useful results. Therefore, we believe that efforts should now be intensified to overcome limitations and avoid pitfalls, in order to move towards the effective clinical establishment of metabonomics in oncology. Enlarged sample cohorts from well-characterized groups of subjects are a first mandatory condition to achieve this goal, in order to account for the ‘real world’ variability and enable statistically validated conclusions to be achieved. Also, the groups compared should be adequately matched in terms of a number of intrinsic and extrinsic factors (e.g. gender, age, body mass index, circadian rhythm, diet, etc.), which are firmly known to modulate the metabolic composition of body fluids [83–89], otherwise the observed variations may arise from differences in any of these factors rather than from the disease itself. Since it is often not possible to compare perfectly matched groups, for instance because some of the patients suffer from co-morbidities or are under specific medication, the influence of the unmatched variable on the metabolic profiles should be checked and possible bias discarded. As already pointed out, standardization of sample collection, handling and analysis, as well as of data treatment and reporting procedures is another crucial issue. Significant efforts are being made in this direction [76,80] and adherence to common operating procedures should greatly facilitate inter-laboratory data comparison, thus increasing the level of confidence on data quality and reproducibility and speeding up the discovery of clinically useful biomarker profiles. At present, a particularly fragile step in this process is the statistical validation of the multivariate models developed for classification purposes. Most studies published to date performed internal validation to optimize model parameters but lack appropriate full external validation (i.e. prediction of class membership for a set of new samples, not used at all in model development). As argued in a number of papers [90,91], this may lead to overoptimistic results and should be avoided through the use of adequate statistical strategies to validate the classification models (e.g. cross model validation and permutation testing, amongst others). Finally, it is expected that increased knowledge on the identity and biological significance of marker metabolites will arise from advances in available databases and assignment strategies, as well as from the combination of NMR with more sensitive MS methods to cover a wider range of the metabolome. While the challenges are numerous and should not be underestimated, the research so far developed supports a strong belief that metabolic profiling of biofluids through NMR spectroscopy will, in time, provide a unique and important tool in the clinical management of cancer, and have major impacts in areas such as screening, diagnosis and monitoring of disease progression. Furthermore,

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

metabonomics studies have great potential in the emerging area of personalized cancer therapy, namely through the identification of metabolic markers of treatment efficacy or early toxicity, thus assisting in the development of novel, targeted therapeutics. Acknowledgments Funding is acknowledged from the European Regional Development Fund through the Competitive Factors Thematic Operational Programme and from the Foundation for Science and Technology (FCT), Portugal (Research Project FCT/PTDC/QUI/68017/2006; FCOMP-01-0124-FEDER-007439). The Portuguese National NMR Network (RNRMN), supported with FCT funds, is also acknowledged. I.F.D. further acknowledges L’Oréal Portugal, FCT and the National UNESCO Committee for funding through the L’Oréal Medals of Honor for Women in Science 2007, and the Portuguese League Against Cancer (LPCC). References [1] [2] [3] [4] [5] [6] [7] [8]

[9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19]

[20] [21] [22]

[23]

[24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36]

J.K. Nicholson, J.C. Lindon, E. Holmes, Xenobiotica 29 (1999) 1181. O. Fiehn, Plant Mol. Biol. 48 (2002) 155. N. Vinayavekhin, E.A. Homan, A. Saghatelian, ACS Chem. Biol. 5 (2009) 91. R. Madsen, T. Lundstedt, J. Trygg, Anal. Chim. Acta 659 (2010) 23. A. Nordström, R. Lewensohn, J. Neuroimmune Pharm. 5 (2010) 4. M. Mamas, W. Dunn, L. Neyses, R. Goodacre, Arch. Toxicol. 85 (2011) 5. P. Vizan, S. Mazurek, M. Cascante, Metabolomics 4 (2008) 1. D.R. Wise, R.J. DeBerardinis, A. Mancuso, N. Sayed, X.-Y. Zhang, H.K. Pfeiffer, I. Nissim, E. Daikhin, M. Yudkoff, S.B. McMahon, C.B. Thompson, Proc. Natl. Acad. Sci. USA 105 (2008) 18782. G. Kroemer, J. Pouyssegur, Cancer Cell 13 (2008) 473. X. Tong, F. Zhao, C.B. Thompson, Curr. Opin. Genet. Dev. 19 (2009) 32. J.L. Griffin, J.P. Shockcor, Nat. Rev. Cancer 4 (2004) 551. B. Sitter, T.F. Bathen, M.-B. Tessem, I.S. Gribbestad, Prog. Nucl. Magn. Reson. Spectrosc. 54 (2009) 239. E.M. DeFeo, L.L. Cheng, Technol. Cancer Res. Treat. 9 (2010) 381. S. Moestue, B. Sitter, T.F. Bathen, M.B. Tessem, I.S. Gribbestad, Curr. Top. Med. Chem. 11 (2011) 2. G. Erb, K. Elbayed, M. Piotto, J. Raya, A. Neuville, M. Mohr, D. Maitrot, P. Kehrli, I.J. Namer, Magn. Reson. Med. 59 (2008) 959. G.F. Giskeødegård, M.T. Grinde, B. Sitter, D.E. Axelson, S. Lundgren, H.E. Fjøsne, S. Dahl, I.S. Gribbestad, T.F. Bathen, J. Proteome Res. 9 (2010) 972. E.M. Lenz, I.D. Wilson, J. Proteome Res. 6 (2007) 443. J.C. Lindon, J.K. Nicholson, Annu. Rev. Anal. Chem. 1 (2008) 45. M.E. Dumas, E.C. Maibaum, C. Teague, H. Ueshima, B.F. Zhou, J.C. Lindon, J.K. Nicholson, J. Stamler, P. Elliott, Q. Chan, E. Holmes, Anal. Chem. 78 (2006) 2199. S. Zhang, G.A. NaganaGowda, T. Ye, D. Raftery, Analyst 135 (2010) 1490. M. Malet-Martino, U. Holzgrabe, J. Pharm. Biomed. Anal. 55 (2011) 1. D.S. Wishart, D. Tzur, C. Knox, R. Eisner, A.C. Guo, N. Young, D. Cheng, K. Jewell, D. Arndt, S. Sawhney, C. Fung, L. Nikolai, M. Lewis, M.-A. Coutouly, I. Forsythe, P. Tang, S. Shrivastava, K. Jeroncic, P. Stothard, G. Amegbey, D. Block, D.D. Hau, J. Wagner, J. Miniaci, M. Clements, M. Gebremedhin, N. Guo, Y. Zhang, G.E. Duggan, G.D. MacInnis, A.M. Weljie, R. Dowlatabadi, F. Bamforth, D. Clive, R. Greiner, L. Li, T. Marrie, B.D. Sykes, H.J. Vogel, L. Querengesser, Nucleic Acids Res. 35 (2007) D521. E.L. Ulrich, H. Akutsu, J.F. Doreleijers, Y. Harano, Y.E. Ioannidis, J. Lin, M. Livny, S. Mading, D. Maziuk, Z. Miller, E. Nakatani, C.F. Schulte, D.E. Tolmie, R. Kent Wenger, H. Yao, J.L. Markley, Nucleic Acids Res. 36 (2008) D402. J. Trygg, E. Holmes, T. Lundstedt, J. Proteome Res. 6 (2007) 469. T.M.D. Ebbels, R. Cavill, Prog. Nucl. Magn. Reson. Spectrosc. 55 (2009) 361. J.L. Spratlin, N. Serkova, S.G. Eckhardt, Clin. Cancer Res. 15 (2009) 431. V.W. Davis, O.F. Bathe, D.E. Schiller, C.M. Slupsky, M.B. Sawyer, J. Surg. Oncol. 103 (2011) 451. D. Nagrath, C. Caneba, T. Karedath, N. Bellance, Biochim. Biophys. Acta, Bioenerg. 1807 (2011) 650. D. Ng, K. Pasikanti, E. Chan, Metabolomics 7 (2011) 155. H. Wen, S.S. Yoo, J. Kang, H.G. Kim, J.-S. Park, S. Jeong, J.I. Lee, H.N. Kwon, S. Kang, D.-H. Lee, S. Park, J. Hepatol. 52 (2010) 228. O. Cloarec, M.E. Dumas, A. Craig, R.H. Barton, J. Trygg, J. Hudson, C. Blancher, D. Gauguier, J.C. Lindon, E. Holmes, J.K. Nicholson, Anal. Chem. 77 (2005) 1282. S. Srivastava, R. Roy, S. Singh, P. Kumar, D. Dalela, S.N. Sankhwar, A. Goel, A.A. Sonkar, Cancer Biomarkers 6 (2010) 11. N.K. Srivastava, S. Pradhan, G.A.N. Gowda, R. Kumar, NMR Biomed. 23 (2010) 113. C.M. Slupsky, H. Steed, T.H. Wells, K. Dabbs, A. Schepansky, V. Capstick, W. Faught, M.B. Sawyer, Clin. Cancer Res. 16 (2010) 5835. O. Warburg, Science 123 (1956) 309. H.W. Gu, Z.Z. Pan, B.W. Xi, V. Asiago, B. Musselman, D. Raftery, Anal. Chim. Acta 686 (2011) 57.

73

[37] V.M. Asiago, L.Z. Alvarado, N. Shanaiah, G.A.N. Gowda, K. Owusu-Sarfo, R.A. Ballas, D. Raftery, Cancer Res. 70 (2010) 8309. [38] C. Oakman, L. Tenori, W.M. Claudino, S. Cappadona, S. Nepi, A. Battaglia, P. Bernini, E. Zafarana, E. Saccenti, M. Fornier, P.G. Morris, L. Biganzoli, C. Luchinat, I. Bertini, A. Di Leo, Ann. Oncol. 22 (2011) 1295. [39] H.C. Keun, J. Sidhu, D. Pchejetski, J.S. Lewis, H. Marconell, M. Patterson, S.R. Bloom, V. Amber, R.C. Coombes, J. Stebbing, Clin. Cancer Res. 15 (2009) 6716. [40] C. Ludwig, D.G. Ward, A. Martin, M.R. Viant, T. Ismail, P.J. Johnson, M.J.O. Wakelam, U.L. Gunther, Magn. Reson. Chem. 47 (2009) S68. [41] D. Monleón, J.M. Morales, A. Barrasa, J.A. Lopez, C. Vazquez, B. Celda, NMR Biomed. 22 (2009) 342. [42] T. Bezabeh, R. Somorjai, B. Dolenko, N. Bryskina, B. Levin, C.N. Bernstein, E. Jeyarajah, A.H. Steinhart, D.T. Ruben, I.C.P. Smith, NMR Biomed. 22 (2009) 593. [43] R. Eisner, C. Stretch, T. Eastman, J.G. Xia, D. Hau, S. Damaraju, R. Greiner, D.S. Wishart, V.E. Baracos, Metabolomics 7 (2011) 25. [44] A. Backshall, R. Sharma, S.J. Clarke, H.C. Keun, Clin. Cancer Res. 17 (2011) 3019. [45] J.A. Zhang, L.Y. Liu, S.W. Wei, G.A.N. Gowda, Z. Hammoud, K.A. Kesler, D. Raftery, J. Thorac. Cardiov. Sur. 141 (2011) 469. [46] F. Sullentrop, D. Moka, S. Neubauer, U. Engelmann, J. Hahn, H. Schicha, NMR Biomed. 15 (2002) 60. [47] H. Gao, B. Dong, X. Liu, H. Xuan, Y. Huang, D. Lin, Anal. Chim. Acta 624 (2008) 269. [48] A.N. Zira, S.E. Theocharis, D. Mitropoulos, V. Migdalis, E. Mikros, J. Proteome Res. 9 (2010) 4038. [49] M. Kuliszkiewicz-Janus, M.A. Tuz, S. Baczyn´ski, Biochim. Biophys. Acta, Spec. Sect. Lipids 1737 (2005) 11. [50] D.A. MacIntyre, B. Jiménez, E.J. Lewintre, C.R. Martín, H. Schäfer, C.G. MBallesteros, J.R. Mayans, M. Spraul, J. García-Conde, A. Pineda-Lucena, Leukemia 24 (2010) 788. [51] H.C. Gao, Q. Lu, X. Liu, H. Cong, L.C. Zhao, H.M. Wang, D.H. Lin, Cancer Sci. 100 (2009) 782. [52] M.I.F. Shariff, N.G. Ladep, I.J. Cox, H.R.T. Williams, E. Okeke, A. Malu, A.V. Thillainayagam, M.M.E. Crossey, S.A. Khan, H.C. Thomas, S.D. Taylor-Robinson, J. Proteome Res. 9 (2010) 1096. [53] M.I. Shariff, A.I. Gomaa, I.J. Cox, M. Patel, H.R. Williams, M.M. Crossey, A.V. Thillainayagam, H.C. Thomas, I. Waked, S.A. Khan, S.D. Taylor-Robinson, J. Proteome Res. 10 (2011) 1828. [54] G.A.N. Gowda, Biomarkers Med. 4 (2010) 299. [55] K.W. Jordan, C.B. Adkins, L. Su, E.F. Halpern, E.J. Mark, D.C. Christiani, L.L. Cheng, Lung Cancer 68 (2010) 44. [56] 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, J. Proteome Res. 10 (2011) 4314. [57] 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, J. Proteome Res. 10 (2011) 221. [58] S. Tiziani, V. Lopes, U.L. Günther, Neoplasia 11 (2009) 269. [59] J. Zhou, B. Xu, J. Huang, X. Jia, J. Xue, X. Shi, L. Xiao, W. Li, Clin. Chim. Acta 401 (2009) 8. [60] E.A. Boss, S.H. Moolenaar, L.F.A.G. Massuger, H. Boonstra, U.F.H. Engelke, J.G.N. Jong, R.A. Wevers, NMR Biomed. 13 (2000) 297. [61] K. Odunsi, R.M. Wollman, C.B. Ambrosone, A. Hutson, S.E. McCann, J. Tammela, J.P. Geisler, G. Miller, T. Sellers, W. Cliby, F. Qian, B. Keitz, M. Intengan, S. Lele, J.L. Alderfer, Int. J. Cancer 113 (2005) 782. [62] E. Garcia, C. Andrews, J. Hua, H.L. Kim, D.K. Sukumaran, T. Szyperski, K. Odunsi, J. Proteome Res. 10 (2011) 1765. [63] D. Coomans, I. Broeckaert, M.P. Derde, A. Tassin, D.L. Massart, S. Wold, Comput. Biomed. Res. 17 (1984) 1. [64] R.D. Beger, L.K. Schnackenberg, R.D. Holland, D. Li, Y. Dragan, Metabolomics 2 (2006) 125. [65] O.F. Bathe, R. Shaykhutdinov, K. Kopciuk, A.M. Weljie, A. McKay, F.R. Sutherland, E. Dixon, N. Dunse, D. Sotiropoulos, H.J. Vogel, Cancer Epidem. Biomar. 20 (2011) 140. [66] T.A. Averna, E.E. Kline, A.Y. Smith, L.O. Sillerud, J. Urol. 173 (2005) 433. [67] E.E. Kline, E.G. Treat, T.A. Averna, M.S. Davis, A.Y. Smith, L.O. Sillerud, J. Urol. 176 (2006) 2274. [68] N.J. Serkova, E.J. Gamito, R.H. Jones, C. O’Donnell, J.L. Brown, S. Green, H. Sullivan, T. Hedlund, E.D. Crawford, Prostate 68 (2008) 620. [69] Y. Fan, T.B. Murphy, J.C. Byrne, L. Brennan, J.M. Fitzpatrick, R.W. Watson, J. Proteome Res. 10 (2011) 1361. [70] M. Rantalainen, O. Cloarec, O. Beckonert, I.D. Wilson, D. Jackson, R. Tonge, R. Rowlinson, S. Rayner, J. Nickson, R.W. Wilkinson, J.D. Mills, J. Trygg, J.K. Nicholson, E. Holmes, J. Proteome Res. 5 (2006) 2642. [71] K. Raffelt, D. Moka, F. Sullentrop, M. Dietlein, J. Hahn, H. Schicha, NMR Biomed. 13 (2000) 8. [72] E.M. Lenz, J. Bright, I.D. Wilson, A. Hughes, J. Morrisson, H. Lindberg, A. Lockton, J. Pharm. Biomed. Anal. 36 (2004) 841. [73] O. Beckonert, H.C. Keun, T.M.D. Ebbels, J. Bundy, E. Holmes, J.C. Lindon, J.K. Nicholson, Nat. Protoc. 2 (2007) 2692. [74] M. Lauridsen, S.H. Hansen, J.W. Jaroszewski, C. Cornett, Anal. Chem. 79 (2007) 1181. [75] O. Teahan, S. Gamble, E. Holmes, J. Waxman, J.K. Nicholson, C. Bevan, H.C. Keun, Anal. Chem. 78 (2006) 4307. [76] P. Bernini, I. Bertini, C. Luchinat, P. Nincheri, S. Staderini, P. Turano, J. Biomol. NMR 49 (2011) 231.

74

I.F. Duarte, A.M. Gil / Progress in Nuclear Magnetic Resonance Spectroscopy 62 (2012) 51–74

[77] B.-J.M. Webb-Robertson, D.F. Lowry, K.H. Jarman, S.J. Harbo, Q.R. Meng, A.F. Fuciarelli, J.G. Pounds, K.M. Lee, J. Pharm. Biomed. Anal. 39 (2005) 830. [78] S. Zhang, C. Zheng, I.R. Lanza, K.S. Nair, D. Raftery, O. Vitek, Anal. Chem. 81 (2009) 6080. [79] T. De Meyer, D. Sinnaeve, B. Van Gasse, E.-R. Rietzschel, M. De Buyzere, M. Langlois, S. Bekaert, J. Martins, W. Van Criekinge, Anal. Bioanal. Chem. 398 (2010) 1781. [80] R. Goodacre, D. Broadhurst, A. Smilde, B. Kristal, J. Baker, R. Beger, C. Bessant, S. Connor, G. Capuani, A. Craig, T. Ebbels, D. Kell, C. Manetti, J. Newton, G. Paternostro, R. Somorjai, M. Sjöström, J. Trygg, F. Wulfert, Metabolomics 3 (2007) 231. [81] E.P. Diamandis, J. Natl. Cancer Inst. 102 (2010) 1462. [82] H.J. Issaq, T.J. Waybright, T.D. Veenstra, Electrophoresis 32 (2011) 967. [83] S. Kochhar, D.M. Jacobs, Z. Ramadan, F. Berruex, A. Fuerhoz, L.B. Fay, Anal. Biochem. 352 (2006) 274. [84] N.G. Psihogios, I.F. Gazi, M.S. Elisaf, K.I. Seferiadis, E.T. Bairaktari, NMR Biomed. 21 (2008) 195.

[85] D.S. Freedman, J.D. Otvos, E.J. Jeyarajah, I. Shalaurova, L.A. Cupples, H. Parise, R.B. D’Agostino, P.W. Wilson, E.J. Schaefer, Clin. Chem. 50 (2004) 1189. [86] E.J. Saude, D. Adamko, B.H. Rowe, T. Marrie, B.D. Sykes, Metabolomics 3 (2007) 439. [87] Y. Park, S.B. Kim, B. Wang, R.A. Blanco, N.-A. Le, S. Wu, C.J. Accardi, R.W. Alexander, T.R. Ziegler, D.P. Jones, Am. J. Physiol. Regul. Integr. Comp. Physiol. 297 (2009) R202. [88] C. Stella, B. Beckwith-Hall, O. Cloarec, E. Holmes, J.C. Lindon, J. Powell, F. van der Ouderaa, S. Bingham, A.J. Cross, J.K. Nicholson, J. Proteome Res. 5 (2006) 2780. [89] E. Peré-Trepat, A.B. Ross, F.-P. Martin, S. Rezzi, S. Kochhar, A.L. Hasselbalch, K.O. Kyvik, T.I.A. Sørensen, Chemom. Intell. Lab. Syst. 104 (2010) 95. [90] D. Broadhurst, D. Kell, Metabolomics 2 (2006) 171. [91] J. Westerhuis, H. Hoefsloot, S. Smit, D. Vis, A. Smilde, E. Velzen, J. Duijnhoven, F. Dorsten, Metabolomics 4 (2008) 81.