+ All Categories
Home > Documents > Tissue metabolite profiling identifies differentiating and prognostic biomarkers for prostate...

Tissue metabolite profiling identifies differentiating and prognostic biomarkers for prostate...

Date post: 15-Dec-2016
Category:
Upload: glen
View: 212 times
Download: 0 times
Share this document with a friend
11
Tissue metabolite profiling identifies differentiating and prognostic biomarkers for prostate carcinoma Klaus Jung 1,2 , Regina Reszka 3 , Beate Kamlage 4 , Bianca Bethan 4 , Carsten Stephan 1,2 , Michael Lein 2,5 and Glen Kristiansen 6 1 Department of Urology, University Hospital Charit e, Schumannstraß 20/21, 10117 Berlin, Germany 2 Berlin Institute for Urologic Research, Schumannstraße 20/21, 10117 Berlin, Germany 3 Metanomics Health GmbH, Tegeler Weg 33, 10589 Berlin, Germany 4 Metanomics GmbH, Tegeler Weg 33, 10589 Berlin, Germany 5 Department of Urology, University Teaching Hospital, Starkenburgring 66, 63069 Offenbach, Germany 6 Institute of Pathology, University Hospital of Bonn, Sigmund-Freud.Straße 25, 53123 Bonn, Germany Metabolomic research offers a deeper insight into biochemical changes in cancer metabolism and is a promising tool for iden- tifying novel biomarkers. We aimed to evaluate the diagnostic and prognostic potential of metabolites in prostate cancer (PCa) tissue after radical prostatectomy. In matched malignant and nonmalignant prostatectomy samples from 95 PCa patients, aminoadipic acid, cerebronic acid, gluconic acid, glycerophosphoethanolamine, 2-hydroxybehenic acid, isopentenyl pyrophos- phate, maltotriose, 7-methylguanine and tricosanoic acid were determined within a global metabolite profiling study using gas chromatography/liquid chromatography-mass spectrometry. The data were related to clinicopathological variables like prostate volume, tumor stage, Gleason score, preoperative prostate-specific antigen and disease recurrence in the follow-up. All nine metabolites showed higher concentrations in malignant than in nonmalignant samples except for gluconic acid and maltotriose, which had lower levels in tumors. Receiver -operating characteristics analysis demonstrated a significant discrimination for all metabolites between malignant and nonmalignant tissue with a maximal area under the curve of 0.86 for tricosanoic acid, whereas no correlation was observed between the metabolite levels and the Gleason score or tumor stage except for gluconic acid. Univariate Cox regression and Kaplan-Meier analyses showed that levels of aminoadipic acid, glu- conic acid and maltotriose were associated with the biochemical tumor recurrence (prostate-specific antigen > 0.2 ng/mL). In multivariate Cox regression analyses, aminoadipic acid together with tumor stage and Gleason score remained in a model as independent marker for prediction of biochemical recurrence. This study proved that metabolites in PCa tissue can be used, in combination with traditional clinicopathological factors, as promising diagnostic and prognostic tools. Prostate cancer continues to be the most frequent cancer in men with 382,000 new cases in Europe and with a predicted incidence of 241,470 for 2012 in the United States. 1,2 How- ever, there is an ongoing controversial debate on the use of prostate-specific antigen (PSA) testing for prostate cancer. 3,4 The limitation of PSA in discriminating between benign and malignant prostate disease and its inability to distinguish between aggressive and less aggressive tumors result in over- diagnosis and over-treatment of rather insignificant tumors with low potential morbidity or death. Consequently, current effort in urologic research has focused on the discovery of new molecular markers to improve both early prostate cancer detection and risk prediction for patients. The developments in molecular biology of “omic” technologies such as genomics, transcriptomics, proteomics and metabolomics provide a firm basis for this identification purpose. 5 Metabolomics is defined as a global quantitative analysis of all endogenous metabolites in a biological system such as an organ, tissue or body fluid. It has become a promising omic approach applying the analytical platforms of nuclear magnetic resonance spectroscopy or mass spectrometry in combination with gas chromatography/liquid chromatogra- phy. 5 Since metabolite profiles closely reflect the total cellular situation, metabolomics has been developed into a more integrative approach in comparison with the other omic approaches. 6 The potential use and application of metabolite profiles in prostate cancer was recently reviewed. 7 The semi- nal study of the prostate cancer metabolome by Sreekumar et al. 8 has raised special interest in this field. The authors Key words: prostate carcinoma, tissue metabolites, recurrence risk prediction, prognostic markers Additional Supporting Information may be found in the online version of this article. Conflict of interest: Regina Reszka, Beate Kamlage, and Bianca Bethan were employees of, and received salary from, Metanomics Health GmbH or Metanomics GmbH during the study. DOI: 10.1002/ijc.28303 History: Received 20 Mar 2013; Accepted 22 Apr 2013; Online 4 Jun 2013 Correspondence to: Klaus Jung, Department of Urology, CCM, University Hospital Charit e, Schumannstr. 20/21, 10117 Berlin, Germany, Tel.: 149-30-450-515041, Fax: 149-30-450-515904, E-mail: [email protected] Early Detection and Diagnosis Int. J. Cancer: 00, 00–00 (2013) V C 2013 UICC International Journal of Cancer IJC
Transcript
Page 1: Tissue metabolite profiling identifies differentiating and prognostic biomarkers for prostate carcinoma

Tissue metabolite profiling identifies differentiatingand prognostic biomarkers for prostate carcinoma

Klaus Jung1,2, Regina Reszka3, Beate Kamlage4, Bianca Bethan4, Carsten Stephan1,2, Michael Lein2,5 and Glen Kristiansen6

1 Department of Urology, University Hospital Charit�e, Schumannstraß 20/21, 10117 Berlin, Germany2 Berlin Institute for Urologic Research, Schumannstraße 20/21, 10117 Berlin, Germany3 Metanomics Health GmbH, Tegeler Weg 33, 10589 Berlin, Germany4 Metanomics GmbH, Tegeler Weg 33, 10589 Berlin, Germany5 Department of Urology, University Teaching Hospital, Starkenburgring 66, 63069 Offenbach, Germany6 Institute of Pathology, University Hospital of Bonn, Sigmund-Freud.Straße 25, 53123 Bonn, Germany

Metabolomic research offers a deeper insight into biochemical changes in cancer metabolism and is a promising tool for iden-

tifying novel biomarkers. We aimed to evaluate the diagnostic and prognostic potential of metabolites in prostate cancer (PCa)

tissue after radical prostatectomy. In matched malignant and nonmalignant prostatectomy samples from 95 PCa patients,

aminoadipic acid, cerebronic acid, gluconic acid, glycerophosphoethanolamine, 2-hydroxybehenic acid, isopentenyl pyrophos-

phate, maltotriose, 7-methylguanine and tricosanoic acid were determined within a global metabolite profiling study using

gas chromatography/liquid chromatography-mass spectrometry. The data were related to clinicopathological variables like

prostate volume, tumor stage, Gleason score, preoperative prostate-specific antigen and disease recurrence in the follow-up.

All nine metabolites showed higher concentrations in malignant than in nonmalignant samples except for gluconic acid and

maltotriose, which had lower levels in tumors. Receiver -operating characteristics analysis demonstrated a significant

discrimination for all metabolites between malignant and nonmalignant tissue with a maximal area under the curve of 0.86 for

tricosanoic acid, whereas no correlation was observed between the metabolite levels and the Gleason score or tumor stage

except for gluconic acid. Univariate Cox regression and Kaplan-Meier analyses showed that levels of aminoadipic acid, glu-

conic acid and maltotriose were associated with the biochemical tumor recurrence (prostate-specific antigen > 0.2 ng/mL). In

multivariate Cox regression analyses, aminoadipic acid together with tumor stage and Gleason score remained in a model as

independent marker for prediction of biochemical recurrence. This study proved that metabolites in PCa tissue can be used, in

combination with traditional clinicopathological factors, as promising diagnostic and prognostic tools.

Prostate cancer continues to be the most frequent cancer inmen with 382,000 new cases in Europe and with a predictedincidence of 241,470 for 2012 in the United States.1,2 How-ever, there is an ongoing controversial debate on the use ofprostate-specific antigen (PSA) testing for prostate cancer.3,4

The limitation of PSA in discriminating between benign andmalignant prostate disease and its inability to distinguish

between aggressive and less aggressive tumors result in over-diagnosis and over-treatment of rather insignificant tumorswith low potential morbidity or death. Consequently, currenteffort in urologic research has focused on the discovery ofnew molecular markers to improve both early prostate cancerdetection and risk prediction for patients. The developmentsin molecular biology of “omic” technologies such asgenomics, transcriptomics, proteomics and metabolomicsprovide a firm basis for this identification purpose.5

Metabolomics is defined as a global quantitative analysisof all endogenous metabolites in a biological system such asan organ, tissue or body fluid. It has become a promisingomic approach applying the analytical platforms of nuclearmagnetic resonance spectroscopy or mass spectrometry incombination with gas chromatography/liquid chromatogra-phy.5 Since metabolite profiles closely reflect the total cellularsituation, metabolomics has been developed into a moreintegrative approach in comparison with the other omicapproaches.6 The potential use and application of metaboliteprofiles in prostate cancer was recently reviewed.7 The semi-nal study of the prostate cancer metabolome by Sreekumaret al.8 has raised special interest in this field. The authors

Key words: prostate carcinoma, tissue metabolites, recurrence risk

prediction, prognostic markers

Additional Supporting Information may be found in the online

version of this article.

Conflict of interest: Regina Reszka, Beate Kamlage, and Bianca

Bethan were employees of, and received salary from, Metanomics

Health GmbH or Metanomics GmbH during the study.

DOI: 10.1002/ijc.28303

History: Received 20 Mar 2013; Accepted 22 Apr 2013; Online 4

Jun 2013

Correspondence to: Klaus Jung, Department of Urology, CCM,

University Hospital Charit�e, Schumannstr. 20/21, 10117 Berlin,

Germany, Tel.: 149-30-450-515041, Fax: 149-30-450-515904,

E-mail: [email protected]

Early

Detection

andDiagn

osis

Int. J. Cancer: 00, 00–00 (2013) VC 2013 UICC

International Journal of Cancer

IJC

Page 2: Tissue metabolite profiling identifies differentiating and prognostic biomarkers for prostate carcinoma

described 37 named metabolites with different concentrationsbetween malignant and nonmalignant tissues. Six metabolites,among them sarcosine, were described as associated withcancer progression. Partly contradictory results of sarcosinein tissue, urine or serum were then reported in subsequentstudies (reviewed in Ref. 7). However, despite these discrep-ancies, the report of Sreekumar et al.,8 together with otherstudies,9–15 suggests it is worth using single metabolites ormetabolite patterns for diagnostic or prognostic purposes inpatients suffering from prostate cancer.

We performed, according to the suggested recommenda-tion of the Early Detection Research Network,16 a retrospec-tive, sample-size based study with matched malignant andnormal adjacent prostate cancer tissue from the samepatients. In contrast to other studies,8 we decided to usetumor tissue and matched normal adjacent tissue to diminishintraindividual variability. The aims were (a) to identifymetabolites with different concentrations in malignant andnonmalignant tissue, (b) to correlate these metabolites withconventional clinicopathological variables (tumor stage andgrade, prostate volume and PSA) and (c) to evaluate thesemetabolites with respect to their potential to discriminatebetween normal and malignant tissue and especially for pre-dicting biochemical recurrence of PSA as indicator of tumorrecurrence after prostatectomy.

Materials and MethodsPatients and tissue samples

For this retrospective study, tumor tissue and matched nor-mal adjacent tissue were taken from prostate specimens afterradical prostatectomy performed between 2001 and 2007 on95 men with untreated prostate carcinoma before surgery. Afull frontal tissue slice of 2–4 mm thickness that was suspi-cious for malignancy was immediately cryopreserved in liquidnitrogen. This tissue slice was only used for research if notnecessary for additional diagnostic purposes. To obtain puretumor tissue (>90%) and matched adjacent normal tissuecompletely free of tumor filtrates and without signs ofinflammation or atrophy, cryosections stained withhematoxylin-eosin were made to identify these regions ofinterest supported by the stained sections from the adjacentin buffered formaldehyde-fixed and in paraffin-embedded tis-sue blocks. Then, a special punch-bioptic technique using abiopsy needle of 3 mm diameter was performed to collect

corresponding tissue samples that were histologically veri-fied.17 The collected samples had a median wet weight of18.1 mg (95% CI: 17.2–19.1 mg) and were stored at 280�Cuntil analysis. Samples of adjacent normal tissue were nottaken from the immediate vicinity of the tumor but asremote from the tumor as technically possible. The selectionof patients was made according to the availability of tissuesamples and the completeness of follow-up data of thepatients on the basis of sample size calculation as mentionedbelow. The study was approved by the local ethical boardand informed patient consent was obtained. For each patient,clinicopathological information on age, prostate volume, pre-operative PSA, tumor classification according to the UICC2002 TNM System,18 tumor Gleason grade based on thewhole specimen and PSA concentrations during postoperativefollow-up were compiled (Table 1).

Measurement of metabolites and PSA

Samples were measured in 2009 within a MxP Broad Profil-ing framework of a global metabolite profiling study by gaschromatography-mass spectrometry (GC-MS) and liquidchromatography-mass spectrometry (LC-MS/MS) analysis(Agilent Technologies, Waldbronn,Germany).19,20 The freeze-dried tissue material was extracted with polar (water) andnonpolar (ethanol/dichloromethane/acetonitrile) solvents. Theextract was fractioned into an aqueous, polar phase (polarfraction) and an organic, lipophilic phase (lipid fraction).

For GC-MS analyses, the lipid fraction was treated withmethanol under acidic conditions to yield the fatty acidmethyl esters derived from both free fatty acids and hydro-lyzed complex lipids. The polar and lipid fractions were fur-ther derivatized with O-methyl-hydroxylamine hydrochloride(20 mg/mL in pyridine, 50 lL) to convert oxo-groups toO-methyloximes and subsequently with N-methyl-N-tri-methylsilyltrifluoracetamid before GC-MS. For LC-MS/MSanalyses, both fractions were reconstituted in appropriatesolvent mixtures. High-performance liquid chromatographywas performed by gradient elution using methanol/water/formic acid on reversed phase separation columns. Anin-house-developed mass spectrometric detection technologywas applied, which allows targeted and high sensitivity“multiple reaction monitoring” profiling in parallel to a fullscreen analysis. Identification of the metabolites was per-formed using an in-house mass spectral library generated

What’s new?

Recent concern about the limitations of prostate-specific antigen (PSA) screening has spurred a search for new molecular bio-

markers for early prostate cancer detection and risk prediction. This global metabolite profiling study contributes to that effort

by identifying metabolites in prostate tumor tissue that are of potential diagnostic and prognostic significance for the disease.

Analysis of tumor tissue and matched normal adjacent tissue from 95 patients revealed that the majority of the metabolites

were present at elevated concentrations in tumor tissue. Nine conspicuous metabolites selected for a more detailed study

were associated with clinicopathological features of disease, and specifically one, aminoadipic acid, was prognostic of tumor

recurrence.

Early

Detection

andDiagn

osis

2 Metabolites in prostate carcinoma tissue

Int. J. Cancer: 00, 00–00 (2013) VC 2013 UICC

Page 3: Tissue metabolite profiling identifies differentiating and prognostic biomarkers for prostate carcinoma

from chromatographic and mass spectrometric measurementsof authentic standard substances.

Tissue samples were analyzed in semirandomized analyti-cal sequence design (samples of each subject analyzed in sub-sequent slots, subjects and sequence of tissue randomized)with pooled samples (5“pool”) generated from extra samplesprovided for this purpose. The raw peak data were normal-ized to the median of pool per analytical sequence to accountfor process variability (so-called ratios versus pool). Thesedata were related to the tissue weight and recentered to themedian of all reference samples to focus on cancer-inducedchanges by retaining variability of control group. Thus, allmetabolite data were presented as relative amounts of

metabolites normalized to a reference pool and that was indi-cated in the figures as “normalized metabolite ratio.” Thus,this approach of relative quantification is to a certain extentsemiquantitative and is generally accepted for nontargetedmetabolomic analyses with an established in-house chemicallibrary.21

A rigorous quality control was performed on peak, analyteand sample level. Typically, the analytical variability, deter-mined as percentage coefficient of variation from repeatedmeasurements of the pool samples, was in the range of 10%.Within each analytical sequence, the signals of the internalstandards were plotted onto control charts. Samples thatdisplayed >30% standard deviation of one of the internal

Table 1. Clinical characteristics of the study group1

Characteristics Total No recurrence Recurrence p2

Number 95 74 21

Age (yr) 0.559

Median (95% CI) 64 (62.5–65) 64 (62.1–65) 64 (61.6–65)

Range 46–73 46–73 46–71

Preoperative PSA (ng/mL) 0.701

Median (95% CI) 7.5 (6.7–8.6) 7.5 (6.5–9.0) 7.5 (5.9–12.4)

Range 1.7–41.9 1.8–41.9 1.7–34.9

Preoperative %fPSA 0.472

Median (95% CI) 8.6 (7.4–10.0) 8.2 (7.0–10.3) 9.1 (7.3–13.9)

Range 2.8–45.3 2.8–45.3 3.2–18.3

Prostate volume (cm3) 0.879

Median (95% CI) 32 (30–35) 32 (28–35) 35 (25–40)

Range 13–98 15–98 13–92

No. pathological stage (%) <0.0003

pT2a 4 (4) 4 (5)

pT2b 15 (16 12 (16) 3 (14)

pT2c 42 (44 40 (54 2 (10)

pT3a 28 (29) 15 (20) 13 (62)

pT3b 6 (6) 3(4) 3 (14)

No. Gleason score (%) <0.0001

5 6 (6) 6 (8) –

6 24 (25) 23 (31) 1 (5)

7 41 (43) 35 (47) 6 (28

8 13 (14) 5 (7) 8 (38)

9 10 (11) 5 (7) 5 (24)

10 1 (1) – 1 (5)

Follow-up (mo) <0.0001

Median (95% CI) 52.1 (46.1–56.8) 55.5 (50.4–66.2) 23.4 (10.9–42.3)

Range 0.6–114 8.3–114 0.6–105.2

1Values are given as medians with 95% CI in parentheses and ranges or numbers of patients with percentages in parentheses.2Significances between no recurrence and recurrence tested by Mann-Whitney U test or v2 test. Abbreviations: CI, confidence interval; PSA, prostate-specific antigen; %fPSA, percentage ratio of free to total PSA.

Early

Detection

andDiagn

osis

Jung et al. 3

Int. J. Cancer: 00, 00–00 (2013) VC 2013 UICC

Page 4: Tissue metabolite profiling identifies differentiating and prognostic biomarkers for prostate carcinoma

standards were invalidated. Outlier peaks on group level(malignant and nonmalignant tissue) were identified by boxplot analysis, manually checked for correct annotation andintegration and, if necessary, manually corrected. Peaks withvery low metabolite abundance that did not allow reliablepeak integration or that did not meet requirements of reten-tion time index were converted to missing values. This proce-dure explains the different missing values for the variousanalytes in the 95 samples evaluated in this study.

PSA was measured on an Elecsys analyzer (RocheDiagnostics, Mannheim, Germany).

Outcome evaluation

The diagnostic validities of metabolites were evaluated regard-ing their capability to discriminate between tumor and normaltissues. Receiver-operating characteristics (ROC) analysis andbinary logistic regression were applied, and the areas underthe curves (AUC), sensitivity, specificity and total correctclassification rates were calculated. The prognostic potential ofthe metabolites in the prostatectomy specimens was assessedto predict tumor recurrence after surgery. The tumor recur-rence was characterized by the conventional criterion ofbiochemical recurrence of PSA defined as the first postopera-tive PSA value of >0.2 ng/mL following an undetectable PSAlevel (<0.04 ng/mL according to the detection limit) after sur-gery and confirmed by consecutive PSA values >0.2 ng/mL.

Statistics

The GraphPad Prism 6.01 (GraphPad Software, San Diego,CA), SPSS Statistics 19 (IBM, Chicago, IL) and MedCalc12.4.0 (MedCalc Software, Mariakerke, Belgium) softwarepackages were used. The use of the different statistical tests(Mann-Whitney U test, Wilcoxon matched pairs test, Spear-man rank correlation, v2 test, Kaplan-Meier analyses andunivariate and multivariate Cox regressions) was mentioned atthe respective places in the text. The SPSS bias-corrected andaccelerated bootstrap method (2,000 bootstrap cycles) wasused as resampling approach and procedure of internal valida-tion of data. ROC analyses were performed with the MedCalcsoftware together with a special bootstrap tool for estimatingoverfitting bias of AUCs.22 The “missing value” module ofSPSS was used to analyze the structure of missing valuesusing Little’s MCAR test. Statistical significance was defined asp < 0.05 (two sided). For multiple testing, p values were cor-rected according to Benjamini and Hochberg23 based on themaximum acceptable false discovery rate (FDR) of 0.05.

Sample size (a 5 5%; b 5 80%) was determined (Graph-Pad Statmate 2.0) as explained in Supporting Information S1.

ResultsPatient characteristics

A total of 95 prostate cancer patients were included in thisretrospective study with matched-paired tissue samples afterradical prostatectomy and complete follow-up data (Table 1).

Within a median follow-up of 52 months, 74 men wererecurrence free and 21 suffered tumor relapse.

Metabolites in malignant and nonmalignant

prostate tissue

A total of 820 spectral features were consistently found in thepaired tissue samples from these 95 patients; 254 of the spec-tral features provided, based on the relative quantificationapproach described in the Methods, robust, semiquantitativedata representing 172 known analytes and 82 unknown spec-tral features. In Supporting Information Table S1, all these254 analytes were listed according to the median ratios ofvalues measured in malignant to nonmalignant samplesincluding their FDR-corrected p values. Taking into accountthe FDR-corrected p values of <0.05 as criterion, 124 weredifferentially expressed. Eighty-six metabolites of these 172known metabolites showed at least a 1.2-fold increasedexpression in the malignant compared with the nonmalignantsamples and only seven metabolites were correspondinglylower expressed in carcinoma tissue.

We compared the known metabolites differentiallyexpressed in the present study with the differential metaboliteslisted in the two mentioned studies of Sreekumar et al.8 andShuster et al.14 Sreekumar et al.8 described 37 and Shusteret al.14 83 differential named metabolites. In total, 186 knownmetabolites were detected in these three studies as differentialmetabolites. An overview with regard to the congruence ofthese differential metabolites between the studies is given inthe Venn diagram (Fig. 1a) and in more detail in theSupporting Information Table S2. Twelve metabolites werecommonly detected in all three studies as differentiallyexpressed compounds. On the other hand, 20 of the 37differential metabolites of the study of Sreekumar et al.8 and34 of the 83 metabolites described by Shuster et al.14 were alsofound differentially expressed in the present study. The dis-tinctly higher number of differential metabolites in our studywas essentially caused by lipid and phospholipid compounds.

Based on these data, we decided to investigate a limitednumber of several known metabolites in more detail to evalu-ate the usefulness of the metabolite changes regarding theirclinical validity (Fig. 1b; Supporting Information Table S1).For this purpose, nine metabolites were selected according tofollowing criteria: (i) highest increase (2-hydroxybehenic acid,cerebronic acid and tricosanoic acid; Supporting InformationTable S1) or (ii) highest decrease (gluconic acid and malto-triose; Supporting Information Table S1) of their levels incarcinoma tissue compared with the adjacent normal tissue,(iii) belonging to different metabolic classes and (iv) exclu-sively found in this study (glycerophosphoethanolamine, iso-pentenyl pyrophosphate and 7-methylguanine). In addition,2-aminoadipic acid, a metabolite of rather unknown function,as one of the 12 metabolites commonly found in all threestudies was included in this validation approach. There werea few missing values of these metabolites due to technicalreasons as explained in the Methods and listed in Supporting

Early

Detection

andDiagn

osis

4 Metabolites in prostate carcinoma tissue

Int. J. Cancer: 00, 00–00 (2013) VC 2013 UICC

Page 5: Tissue metabolite profiling identifies differentiating and prognostic biomarkers for prostate carcinoma

Information Table S3. Little’s MCAR test (p 5 0.140) indicatedthat values were not missing at random and missing com-pletely at random was a reasonable assumption. This test wasalready successfully used in metabolomics.24 The condition offew missing values completely at random allows computationswith a list-wise deletion of data or application of singleimputation methods to replace missing values. We used bothapproaches and mention them at the appropriate places. Signif-icant differences in all the nine metabolites were found betweennonmalignant and malignant tissue samples, with higherconcentrations in tumor samples except for gluconic acid andmaltotriose, which had lower levels in tumors (Fig. 2).

However, no significant differences in the nine metabolitesdependent on Gleason score (<7 vs. �7; Mann-WhitneyU test, p 5 0.089–0.979) or tumor stage (pT2 vs. pT3;Mann-Whitney U test, p 5 0.150–0.887) except for gluconicacid (p 5 0.021) were observed (Supporting Information TableS4). Age, preoperative PSA, prostate volume, tumor stage andGleason score were not significantly correlated (p > 0.05) withany of the nine metabolites in the tumor samples (SupportingInformation Table S5). In contrast, there were strong correla-tions of the metabolites among each other, with particularlydistinctive features: (i) high positive correlations, with rs-valuesfrom 0.89 to 0.93, between cerebronic acid, 2-hydroxybehenic

acid and tricosanoic acid and (ii) negative correlations of thesefatty acids and other metabolites with gluconic acid and mal-totriose that were positively correlated with each other (Sup-porting Information Table S5).

Discriminative potential of metabolites between malignant

and nonmalignant tissue

ROC analysis was used to assess the diagnostic validity of thenine metabolites as discriminative analytes to distinguishmalignant from nonmalignant tissue. AUCs of the ROCanalyses and correct classification rates characterize, in corre-spondence with the data in Figs 2a–2h, the performance ofall metabolites to distinguish malignant from nonmalignanttissue (Table 2). Six metabolites showed AUCs > 0.75.However, the discriminatory power of all nine metabolitescombined was no greater than that of the best metabolite, tri-cosanoic acid, with an AUC value of 0.86, alone. As expectedfrom the metabolite profiles in relation to tumor characteris-tics, the metabolite concentrations were unable to discrimi-nate between Gleason score <7 vs. �7 (AUCs: 0.50–0.61,p 5 0.961–0.099) or tumor stage pT2 vs. pT3 (AUCs: 0.50–0.59, p 5 0.994–0.126) except for gluconic acid with an AUCof 0.64 (p 5 0.017).

Prognostic potential of metabolites predicting

tumor recurrence

The different metabolite profiles led us to evaluate theirpotential for predicting tumor recurrence, defined asbiochemical recurrence with persistent increase of PSA level>0.2 ng/mL following radical prostatectomy.

First, Kaplan-Meier curves were calculated (Figs. 3a–3eand Supporting Information Fig. S1). The recurrence-freeinterval was significantly shorter with increasing tumor stageand Gleason score; this association proved that our studygroup was representative (Figs. 3a and 3b). Kaplan-Meieranalyses were also significant for aminoadipic acid, gluconicacid and maltotriose (Figs. 3c–3e, p 5 0.004–0.030), whilethe other six metabolites were not associated with biochemi-cal relapse (p 5 0.071–0.806). Next, all clinicopathologicalfactors and metabolites were assessed regarding their prog-nostic information in Cox regressions (Table 3). The hazardratios of the metabolites in univariate Cox regression analysesreflected the results of the Kaplan-Meier curves. Then, boththe significant pathological factors (tumor stage and Gleasonscore) and metabolites (aminoadipic acid, gluconic acid andmaltotriose) from the univariate analyses were evaluated inmultivariate Cox regression analyses using full and reducedmodels (Table 3). Taking into account the limited number of21 recurrence events indicated by a biochemical PSA relapse,all these calculations were performed with bootstrappingreplicates. Aminoadipic acid, together with tumor stage andGleason score, remained as independent marker for recurrencerisk prediction in the final model (Table 3). Analyses per-formed either with cases of complete data or with imputedmissing values in the data set gave similar results (Supporting

Figure 1. (a) Venn diagram of the differential metabolites between

prostate cancer and adjacent normal tissue described as known

metabolites in the global metabolite profiling studies by Sreekumat

et al.8 and Shuster et al.14 and in the present study and (b) the

selected metabolites of this study for the analysis of their clinical

validity. The listed nine metabolites used for further analysis in

this study were selected from the 172 known metabolites given in

the Supporting Information Table S1. The selection was based on

the criteria given in the text.

Early

Detection

andDiagn

osis

Jung et al. 5

Int. J. Cancer: 00, 00–00 (2013) VC 2013 UICC

Page 6: Tissue metabolite profiling identifies differentiating and prognostic biomarkers for prostate carcinoma

Information Table S6). Despite the generally acknowledgedbootstrapping approach as the most efficient internal valida-tion approach to obtain a regression model with the lowestpossible bias,25 the limited predictive validity of the results dueto the small number of events is critically discussed later.

DiscussionThe present global metabolite profiling study of prostatecancer tissue samples included, to our best of knowledge, so

far the greatest number of prostate tissue samples measured.This aspect represents the real strength of this study to real-ize successfully the three objectives of this study consisting ofthe identification of differential metabolites on a statisticallysafe basis, the correlation of the metabolites with clinicopa-thological factors and the evaluation of diagnostic and prog-nostic validity of selected particularly striking metabolites.Previous studies often included only a few patients and moreor less assessed the analytical feasibility rather than the

Figure 2. Metabolites in matched malignant and nonmalignant prostatectomy samples. Differences were tested by the Wilcoxon matched

pairs test. Because of missing values as mentioned in the text, there were between 83 and 93 complete matched pairs of samples from

the total cohort of 95 patients as given in Table 3.

Early

Detection

andDiagn

osis

6 Metabolites in prostate carcinoma tissue

Int. J. Cancer: 00, 00–00 (2013) VC 2013 UICC

Page 7: Tissue metabolite profiling identifies differentiating and prognostic biomarkers for prostate carcinoma

diagnostic validity and did not always reveal prognosticinformation (reviewed in Ref. 7).

In consequence of this study design promoted by asophisticated analytical technology that was combined withan in-house library of metabolites based on a longstandingexperience, we detected 124 differential known metabolitesbetween cancer and normal adjacent tissue samples. Thesedifferential metabolites complement the data of two previousstudies that also analyzed metabolites either in tissueextracts8 or in fixative solution of prostate cancer specimensused for histology14 and found 37 and 83 differential namedmetabolites, respectively. To sum up the different analytesfound in these three comparable metabolite profiling studies,more than 180 metabolites could be considered differentiallyexpressed between malignant and nonmalignant prostate tis-sue (Fig. 1; Supporting Information Table S2). Analytical rea-sons (e.g., use of various analytical technologies with differentanalytical sensitivities and availability of different metabolitelibraries) and study design differences (e.g., patient cohortswith different clinicopathological characteristics; number ofsamples; matched/unmatched sample pairs and statisticalcriteria including multiple testing approaches) may explainthe different number of differential metabolites found in thethree studies.

However, the main goal of this study was primarilyfocused on evaluating the relationships between these metab-olites and clinicopathological characteristics as well as theclinical potential of the metabolites. Therefore, we orientatedafter this first explorative step in the further analysis on themost conspicuous metabolites as described. All nine

metabolites clearly discriminated between malignant andnonmalignant tissue with tricosanoic acid showing the high-est AUC value of 0.86. However, the level of these ninemetabolites in the tumor samples did not correlate to theclinical variables age, PSA and prostate volume and only glu-conic acid of the nine metabolites discriminated between thetumor stages pT2 and pT3. A similar observation was madeby Shuster et al.14 showing different levels of alanine and lac-tate between malignant and nonmalignant prostate tissue butnot between tumor stages T2 and T3. On the other hand,these authors demonstrated likewise not only a linearincrease of metabolites (e.g., uracil, leucine and proline)between these tumor stages but also increased levels (putres-cine and spermine) in T2 tumors compared with nonmalig-nant tissue and lower levels of these metabolites in T3 thanin T2 tumors. Furthermore, recent studies using in vivo or exvivo magnetic resonance techniques for measuring choline,creatine, spermine and citrate were able to differentiatebetween Gleason 314 and Gleason 413 tumors reflectingthe tumor aggressiveness.10–12 Thus, it can be concluded thatthe relationships between histopathological variables and dif-ferent metabolites are by no means uniform but ratherdiverse. This fact is hardly surprising since the metabolicalterations due to the numerous disturbed pathways and theirinteractions in cancer are manifold and complex but do notnecessarily reflect morphological characteristics and struc-tures. However, the lack of a correlation of selected metabo-lites with histopathological parameters of tumor progressionmay also indicate an independence of these, should a prog-nostic value of the respective metabolite be demonstrable in

Table 2. Discrimination performance of metabolites between malignant and nonmalignant tissue samples from prostate cancer

Metabolite1 AUC (95% CI)2 Correct classification (%)

Nonmalignant Malignant Overall

Aminoadipic acid 0.69 (0.61–0.76)*** 80 50 65

Cerebronic acid 0.84 (0.78–0.89)*** 89 63 76

Gluconic acid 0.61 (0.53–0.69)** 51 63 57

Glycerophosphoethanol-amine 0.76 (0.69–0.83)*** 75 61 68

2-Hydroxybehenic acid 0.85 (0.79–0.91)*** 86 66 76

Isopentenyl pyrophosphate 0.76 (0.69–0.83)*** 84 60 72

Maltotriose 0.64 (0.54–0.70)** 40 76 58

7-Methylguanine 0.77 (0.69–0.84)*** 79 59 69

Tricosanoic acid 0.86 (0.80–0.91)*** 87 67 77

Combinations

All metabolites3 0.88 (0.82–0.93)*** 90 70 80

Tricosanoic acid4 0.86 (0.80–0.91)*** 87 67 77

1Analyses with 2000 bootstrap cycles were performed with the data sets available for each individual metabolite as given in Supporting InformationTable S3 and only with cases after listwise deletion of cases with incomplete data if binary logistic regression was performed.2Significantly different from AUC 5 0.5: *p < 0.05; **p < 0.001; ***p < 0.0001.3Predicted by full logistic regression.4Predicted by stepwise logistic regression (forward or backward elimination; entry p 5 0.05, elimination p 5 0.10). Abbreviations: AUC, area underthe receiver-operating curve; CI, confidence interval.

Early

Detection

andDiagn

osis

Jung et al. 7

Int. J. Cancer: 00, 00–00 (2013) VC 2013 UICC

Page 8: Tissue metabolite profiling identifies differentiating and prognostic biomarkers for prostate carcinoma

Figure 3. Kaplan-Meier curves for biochemical recurrence-free time after surgery by (a) pathological tumor stage, (b) Gleason score and

levels of (c) aminoadipic acid, (d) gluconic acid and (e) maltotriose. The stratification of the metabolites corresponds to the criteria given in

Table 3.

Early

Detection

andDiagn

osis

8 Metabolites in prostate carcinoma tissue

Int. J. Cancer: 00, 00–00 (2013) VC 2013 UICC

Page 9: Tissue metabolite profiling identifies differentiating and prognostic biomarkers for prostate carcinoma

following studies with larger cohorts and/or in functionalexperiments.

In certain ways, this aspect is supported by the Kaplan-Meier and Cox regression analyses in our study thatconfirmed the prognostic value of aminoadipic acid, malto-triose and gluconic acid. In contrast, the three fatty acidswith the best discriminative power between malignant andnonmalignant tissue were not valid in this respect. Aminoa-dipic acid was revealed as an independent prognosticator fortumor recurrence in a model together with tumor stage andGleason grade. In this respect, metabolites could be used asadjunct indicators for early risk stratification followingprostatectomy to improve the predictive accuracy of theconventional prognostic scoring systems. Using a pattern of

11 metabolite markers, a similarly successful approach wasdescribed for early detection of recurrent breast cancer.26

Thus, also our data raise expectations for successful transla-tion of these results into the application of noninvasivemarkers in urine or serum as already shown in pilotstudies,8,27–29 such a test or test combinations could be usedlike the PCA3 test as a decision tool to perform a biopsy ornot. However, the sarcosine story, with conflicting results innumerous studies, has drawn attention to problems causedby this approach.7

The altered metabolic pattern (Supporting InformationTables S1 and S2) corresponds to well-known metabolic alter-ations in cancer cells and shown in prostate cancer cells.30

These changes are the results of the generally increased

Table 3. Univariate and multivariate Cox proportional hazard regression analyses of tumor stage, Gleason score and tissue metabolites inprostate cancer patients with regard to the recurrence-free interval after radical prostatectomy

Variable1Hazard ratio(95% CI) P

Univariate analysis2

Tumor stage (pT2/pT3) 8.20 (2.77–24.3) 0.0002

Gleason score (516/7/819110) 4.63 (2.21–9.65) 0.0001

Age (<65/�65 yr) 0.82 (0.34–1.96) 0.652

Prostate-specific antigen (<10� ng/mL) 1.07 (0.42–2.69) 0.891

Aminoadipic acid (<1.85�); n 5 93 2.58 (1.10–6.09) 0.031

Cerebronic acid (<5.27�); n 5 87 1.98 (0.75–5.21) 0.168

Gluconic acid (<0.403�); n 5 92 0.38 (0.16–0.94) 0.036

Glycerophosphoethanolamine (<1.15�); n 5 93 2.20 (0.74–6.51) 0.157

2-Hydroxybehenic acid (<5.79�); n 5 83 1.44 (0.57–3.66) 0.441

Isopentenyl pyrophosphate (<1.21�); n 5 92 1.11 (0.47–2.63) 0.806

Maltotriose (<0.129�), n 5 92 0.29 (0.13–0.71) 0.006

7-Methylguanine (<1.00�); n 5 89 1.80 (0.66–4.88) 0.252

Tricosanoic acid (<4.14�); n 5 86 2.31 (0.91–5.85) 0.079

Multivariate analysis, full model3

Tumor stage 3.99 (1.10–14.5) 0.036

Gleason score 2.25 (0.93–5.40) 0.071

Aminoadipic acid 2.37 (0.93–6.02) 0.071

Gluconic acid 0.81 (0.28–2.35) 0.704

Maltotriose 0.62 (0.21–1.81) 0.384

Multivariate analysis, backward elimination4

Tumor stage 3.92 (1.14–13.5) 0.031

Gleason score 2.69 (1.19–6.09) 0.018

Aminoadipic acid 2.51 (1.01–6.23) 0.048

1The categorization criteria are given in brackets. For age and PSA, conventional bifurcation points were used. Metabolites were dichotomized usingtheir values (given in arbitrary units) obtained in the ROC analysis as the point of maximal accuracy to discriminate between recurrence/nonrecurrence.2Univariate analyses using the metabolites were done with the available data for the respective metabolite indicated by “n” (see also SupportingInformation Table S3), while, for the other variables, complete data for all 95 cases were available.3Full model with significant individual variables from the univariate analyses (p < 0.05).4Using the significant individual variables from the univariate analyses (p < 0.05), the multivariate analysis with the backward elimination approachwas made with p 5 0.05 for entry and p 5 0.10 for removal.

Early

Detection

andDiagn

osis

Jung et al. 9

Int. J. Cancer: 00, 00–00 (2013) VC 2013 UICC

Page 10: Tissue metabolite profiling identifies differentiating and prognostic biomarkers for prostate carcinoma

metabolic rates in cancer, affecting especially the reprog-rammed metabolism of amino acids, carbohydrates and com-plex lipids.31 However, the link between the regulation ofmetabolic genes by classical cancer genes and the recent dis-covery of mutations targeting metabolic genes in cancer sup-port the theory that the altered profile of metabolites is notonly the result of cancer but could also promote tumorigene-sis by interactions with several cellular processes.32,33 Thenine metabolites studied belong to different metabolic classes.Four of the seven upregulated metabolites, namely cerebronicacid, hydroxybehenic acid, tricosanoic acid and glycerophos-phoethanolamine, are components of complex lipids. Thethree fatty acids give an indication of the probable cause ofthis alteration. For example, a-hydroxy long-chain fatty acids(hydroxybehenic acid and cerebronic acid) cannot be metabo-lized by b-oxidation but only by a-oxidation, which is local-ized in peroxisomes.34 Cerebronic acid, as a major constituentof cerebrosides, is decarboxylated by a-oxidation to tricosa-noic acid and CO2.

35 The accumulation of these fatty acids inprostate cancer tissue indicates the involvement of peroxiso-mal a-oxidation in these changes.36 Increased levels ofglycerophosphoethanolamine have already been described inprostate cancer samples.37

With regard to the other three increased metabolites, itshould be noted that isopentenyl pyrophosphate as interme-diate in the steroid synthesis pathway is formed from acetyl-CoA in the mevalonate pathway,38 and its increase reflectsthe increased cholesterol values in prostate cancer tissue.15

Aminoadipic acid is an end product of the oxidation oflysine.39 Only recently, the uptake of lysine and secretion ofaminoadipic acid was found to be increased in cell cultureexperiments with tumor suppressor KLF4 deficient cells.40

The metabolism of aminoadipic acid was altered in conse-quence of the changed metabolism of lysine that formsacetyl-CoA for the increased fatty acid synthesis. Therefore,the authors suggested aminoadipic acid as potential bio-marker in cancer indicating differential expression of KLF4that functions as transcription factor.40 In this respect is ofinterest that the expression of KLF4 in prostate cancer tissueis generally decreased and has prognostic value for predictingmetastasis.41 Thus, our results of aminoadipic acid as prog-nosticator of biochemical recurrence, the relation betweenaminoadipic acid and KLF4 and the literature data of theprognostic value of KLF4 in prostate cancer explaineach other.

The 7-methylguanine is a repair product of methylationdamage to DNA.42 Increased methylguanine levels have alsobeen found in other tumors.43,44 A reduced defense againstintracellular reactive oxygen species, either due to enzymaticor zinc-based deficiencies, has been suggested as the cause ofincreased occurrence of methylguanine.43,44

The two downregulated metabolites, gluconic acid andmaltotriose, are related to carbohydrate metabolism. Gluconicacid connects the glucose and pentose phosphate pathways,and changed concentrations have been found under

conditions of oxidative stress.45 Maltotriose, which has beenknown until now only as an intermediate degradation prod-uct of ingested starch by the digestive enzyme amylase, wasrecently discovered to be synthesized in chorioncarcinomacells exhibiting distinct immunoregulatory activities.46 It wasalso shown that maltotriose represents one of the essentialnuclear localization signal sequence motifs.47 These signalmolecules govern the transport of proteins between thenuclear and cytoplasmic compartments by the so-callednuclear protein import cycle in order to maintain their dis-tinctive composition.48 These brief comments draw attentionto cellular processes possibly altered, as indicated by thechanged metabolite concentrations not previously identified,but the actual genetic and proteomic changes responsible forthese alterations still remain unclear. An elucidation of thesephenomena might contribute to the discovery of newbiomarkers for diagnosis and prediction of outcome and newtherapeutic targets.

Some apparent limitations of this study with regard to itsdesign, multiple test problems and number of patientsincluding the number of events used for the prognosticevaluation of metabolites need to be addressed. It was aretrospective study, but to avoid a selection bias, the sampleswere used according to the availability of cryopreserved tissuein consecutive order. Multiple testing in the first step ofvalidating the metabolites as biomarkers of malignancy wascompensated by the FDR approach according to Benjaminiand Hochberg.23 To avoid Type I and II errors, we examinedat least 10% more patients than the sample size calculationsuggested both for ROC and Kaplan-Meier analyses (Sup-porting Information Methods S1) and applied the bootstrap-ping approach for internal validation of all data. A limitedvalidity of the final prognostic model of the multivariate Coxregression analysis has to be considered due to the smallnumber of recurrence events of biochemical relapse. Five to10 events per variable in a Cox regression analysis have beensuggested in order to obtain reliably stable prognosticmodels.49 On the other hand, the regression coefficientscalculated by the bootstrapping replicates supported the accu-racy of the regression calculations as shown in other regres-sion analyses. Thus, although the study design considered allthese limitations, the significance of our results requiresfurther evaluations.

In conclusion, both the diagnostic and prognostic validityof the exemplarily selected metabolites could be demon-strated in this study. The data support the concept that theinclusion of single metabolites or metabolite combinationsfrom different metabolic classes, either in analyses of biopsiesor resected prostates, could improve conventional prognosticmodels based on clinicopathological variables.9,14 In future,metabolomics, as part of an integrated network withgenomics, transcriptomics proteomics, and concurrent histo-logical analysis, will decisively contribute to a deeper under-standing of cancer processes and to improved diagnostic andprognostic results.50

Early

Detection

andDiagn

osis

10 Metabolites in prostate carcinoma tissue

Int. J. Cancer: 00, 00–00 (2013) VC 2013 UICC

Page 11: Tissue metabolite profiling identifies differentiating and prognostic biomarkers for prostate carcinoma

References

1. Ferlay J, Parkin DM, Steliarova-Foucher E.Estimates of cancer incidence and mortality inEurope in 2008. Eur J Cancer 2010;46:765–81.

2. Siegel R, Naishadham D, Jemal A. Cancerstatistics, 2012. CA Cancer J Clin 2012;62:10–29.

3. Catalona WJ, D’Amico AV, Fitzgibbons WF,et al. What the U.S. Preventive Services TaskForce missed in its prostate cancer screeningrecommendation. Ann Intern Med 2012;157:137–8.

4. Moyer VA. Screening for prostate cancer: U.S.Preventive Services Task Force recommendationstatement. Ann Intern Med 2012;157:120–34.

5. Aboud OA, Weiss RH. New opportunities fromthe cancer metabolome. Clin Chem 2013;59:138–46.

6. Claudino WM, Goncalves PH, Di LA, et al.Metabolomics in cancer: a bench-to-bedsideintersection. Crit Rev Oncol Hematol 2012;84:1–7.

7. Trock BJ. Application of metabolomics to pros-tate cancer. Urol Oncol 2011;29:572–81.

8. Sreekumar A, Poisson LM, Rajendiran TM, et al.Metabolomic profiles delineate potential role forsarcosine in prostate cancer progression. Nature2009;457:910–4.

9. Poisson LM, Sreekumar A, Chinnaiyan AM, et al.Pathway-directed weighted testing procedures forthe integrative analysis of gene expression andmetabolomic data. Genomics 2012;99:265–74.

10. Nagarajan R, Margolis D, Raman S, et al. MRspectroscopic imaging and diffusion-weightedimaging of prostate cancer with Gleasonscores. J Magn Reson Imaging 2012;36:697–703.

11. Dittrich R, Kurth J, Decelle EA, et al. Assessingprostate cancer growth with citrate measured byintact tissue proton magnetic resonancespectroscopy. Prostate Cancer Prostatic Dis 2012;15:278–82.

12. Selnaes KM, Gribbestad IS, Bertilsson H, et al.Spatially matched in vivo and ex vivo MR meta-bolic profiles of prostate cancer—investigation ofa correlation with Gleason score. NMR Biomed2013;26:600–6.

13. Maxeiner A, Adkins CB, Zhang Y, et al. Retro-spective analysis of prostate cancer recurrencepotential with tissue metabolomic profiles.Prostate 2010;70:710–17.

14. Shuster JR, Lance RS, Troyer DA. Molecularpreservation by extraction and fixation, mPREF: amethod for small molecule biomarker analysisand histology on exactly the same tissue. BMCClin Pathol 2011;11:14.

15. Thysell E, Surowiec I, Hornberg E, et al.Metabolomic characterization of human prostatecancer bone metastases reveals increased levels ofcholesterol. PLoS One 2010;5:e14175.

16. Pepe MS, Etzioni R, Feng Z, et al. Phases ofbiomarker development for early detection ofcancer. J Natl Cancer Inst 2001;93:1054–61.

17. Jentzmik F, Stephan C, Lein M, et al. Sarcosinein prostate cancer tissue is not a differentialmetabolite for prostate cancer aggressiveness andbiochemical progression. J Urol 2011;185:706–11.

18. Sobin LH, Wittekind C. TNM classification ofmalignant tumours, 6th edn. New York: Wiley-Liss, 2002. 184–7.

19. Roessner U, Wagner C, Kopka J, et al. Technicaladvance: simultaneous analysis of metabolites inpotato tuber by gas chromatography-mass spec-trometry. Plant J 2000;23:131–42.

20. van Ravenzwaay B, Cunha GC, Leibold E, et al.The use of metabolomics for the discovery ofnew biomarkers of effect. Toxicol Lett 2007;172:21–8.

21. Evans AM, DeHaven CD, Barrett T, et al. Inte-grated, nontargeted ultrahigh performance liquidchromatography/electrospray ionization tandemmass spectrometry platform for the identificationand relative quantification of the small-moleculecomplement of biological systems. Anal Chem2009;81:6656–67.

22. Schr€oder B. Evaluating the predictive perform-ance of species distribution models. Available at:http://lec.wzw.tum.de/index.php?id567. AccessedMarch 15, 2013.

23. Benjamini Y, Hochberg Y. Controlling the falsediscovery rate: a practical and powerful approachto multiple testing. J R Stat Soc Series B StatMethodol 1995;57:289–300.

24. Hrydziuszko O, Viant MR. Missing values inmass spectrometry based metabolomics: anundervalued step in the data processing pipeline.Metabolomics 2012;8:S161–S174.

25. Steyerberg EW, Harrell FE Jr, Borsboom GJ, et al.Internal validation of predictive models: efficiencyof some procedures for logistic regressionanalysis. J Clin Epidemiol 2001;54:774–81.

26. Asiago VM, Alvarado LZ, Shanaiah N, et al. Earlydetection of recurrent breast cancer usingmetabolite profiling. Cancer Res 2010;70:8309–18.

27. Koochekpour S, Majumdar S, Azabdaftari G,et al. Serum glutamate levels correlate with Glea-son score and glutamate blockade decreases pro-liferation, migration, and invasion and inducesapoptosis in prostate cancer cells. Clin Cancer Res2012;18:5888–901.

28. Lokhov PG, Dashtiev MI, Moshkovskii SA, et al.Metabolite profiling of blood plasma of patientswith prostate cancer. Metabolomics 2010;6:156–63.

29. Stabler S, Koyama T, Zhao Z, et al. Serum methi-onine metabolites are risk factors for metastaticprostate cancer progression. PLoS One 2011;6:e22486.

30. Teahan O, Bevan CL, Waxman J, et al. Metabolicsignatures of malignant progression in prostateepithelial cells. Int J Biochem Cell Biol 2011;43:1002–9.

31. Munoz-Pinedo C, El MN, Ricci JE. Cancermetabolism: current perspectives and futuredirections. Cell Death Dis 2012;3:e248.

32. Teicher BA, Linehan WM, Helman LJ. Targetingcancer metabolism. Clin Cancer Res 2012;18:5537–45.

33. Oermann EK, Wu J, Guan KL, et al. Alterationsof metabolic genes and metabolites in cancer.Semin Cell Dev Biol 2012;23:370–80.

34. Singh I. Biochemistry of peroxisomes in healthand disease. Mol Cell Biochem 1997;167:1–29.

35. Sandhir R, Khan M, Singh I. Identification of thepathway of alpha-oxidation of cerebronic acid inperoxisomes. Lipids 2000;35:1127–33.

36. Zha S, Ferdinandusse S, Hicks JL, et al. Peroxiso-mal branched chain fatty acid beta-oxidationpathway is upregulated in prostate cancer.Prostate 2005;63:316–23.

37. Swanson MG, Keshari KR, Tabatabai ZL, et al.Quantification of choline- and ethanolamine-containing metabolites in human prostate tissuesusing 1H HR-MAS total correlation spectroscopy.Magn Reson Med 2008;60:33–40.

38. Lombard J, Moreira D. Origins and earlyevolution of the mevalonate pathway of isopre-noid biosynthesis in the three domains of life.Mol Biol Evol 2011;28:87–99.

39. Sell DR, Strauch CM, Shen W, et al. 2-aminoa-dipic acid is a marker of protein carbonyl oxida-tion in the aging human skin: effects of diabetes,renal failure and sepsis. Biochem J 2007;404:269–77.

40. Bellance N, Pabst L, Allen G, et al. Oncosecre-tomics coupled to bioenergetics identifies alpha-amino adipic acid, isoleucine and GABA aspotential biomarkers of cancer: differentialexpression of c-Myc, Oct1 and KLF4 coordinatesmetabolic changes. Biochim Biophys Acta 2012;1817:2060–71.

41. Wang J, Place RF, Huang V, et al. Prognosticvalue and function of KLF4 in prostate cancer:RNAa and vector-mediated overexpressionidentify KLF4 as an inhibitor of tumor cellgrowth and migration. Cancer Res 2010;70:10182–91.

42. Ames BN. Endogenous DNA damage as relatedto cancer and aging. Mutat Res 1989;214:41–6.

43. Saad AA, O’Connor PJ, Mostafa MH, et al.Bladder tumor contains higher N7-methylguaninelevels in DNA than adjacent normal bladderepithelium. Cancer Epidemiol Biomarkers Prev2006;15:740–3.

44. Newberne PM, Broitman S, Schrager TF. Esopha-geal carcinogenesis in the rat: zinc deficiency,DNA methylation and alkyltransferase activity.Pathobiology 1997;65:253–63.

45. Liu XR, Zheng XF, Ji SZ, et al. Metabolomic anal-ysis of thermally injured and/or septic rats. Burns2010;36:992–8.

46. Zhu A, Romero R, Huang JB, et al. Maltooligo-saccharides from JEG-3 trophoblast-like cellsexhibit immunoregulatory properties. Am JReprod Immunol 2011;65:54–64.

47. Niikura K, Sekiguchi S, Nishio T, et al. Oligosac-charide-mediated nuclear transport of nanopar-ticles. Chembiochem 2008;9:2623–7.

48. Stewart M. Molecular mechanism of the nuclearprotein import cycle. Nat Rev Mol Cell Biol 2007;8:195–208.

49. Kawada T. The appropriate number of endpointsto keep validity for Cox proportional hazard anal-ysis. Int J Cardiol 2011;153:110–1.

50. Brown MV, McDunn JE, Gunst PR, et al. Cancerdetection and biopsy classification usingconcurrent histopathological and metabolomicanalysis of core biopsies. Genome Med2012;4:33.

Early

Detection

andDiagn

osis

Jung et al. 11

Int. J. Cancer: 00, 00–00 (2013) VC 2013 UICC


Recommended