Review

Metabolomic Biomarkers of Prostate Cancer:Prediction, Diagnosis, Progression, Prognosis,and RecurrenceRachel S. Kelly1,2, Matthew G. Vander Heiden3,4,5, Edward Giovannucci1,2,6,and Lorelei A. Mucci1,2

Abstract

Metabolite profiling is being increasing employed in the studyof prostate cancer as ameans of identifying predictive, diagnostic,and prognostic biomarkers. This review provides a summary andcritique of the current literature. Thirty-three human case–controlstudies of prostate cancer exploring disease prediction, diagnosis,progression, or treatment response were identified. All but onedemonstrated the ability of metabolite profiling to distinguishcancer from benign, tumor aggressiveness, cases who recurred,and those who responded well to therapy. In the subset of studieswhere biomarker discriminatory ability was quantified, highAUCs were reported that would potentially outperform the cur-rent gold standards in diagnosis, prognosis, and disease recur-

rence, including PSA testing. There were substantial similaritiesbetween the metabolites and the associated pathways reported assignificant by independent studies, and important roles for abnor-mal cell growth, intensive cell proliferation, and dysregulation oflipid metabolism were highlighted. The weight of the evidencetherefore suggests metabolic alterations specific to prostate car-cinogenesis and progression that may represent potential meta-bolic biomarkers. However, replication and validation of themost promising biomarkers is currently lacking and a numberof outstanding methodologic issues remain to be addressed tomaximize the utility of metabolomics in the study of prostatecancer. Cancer Epidemiol Biomarkers Prev; 25(6); 887–906. �2016 AACR.

Aims and MethodsThe aim of this review is to summarize the existing literature

wheremetabolomics has been used to evaluate prostate cancer, toexplore potential metabolite biomarkers that could augmentcurrent diagnostic, prognostic, or screening strategies. The meta-bolites and pathways implicated by these studies are also dis-cussed to considerwhatmechanistic information theymay impartabout prostate cancer.

PubMed was searched to identify studies of prostate cancer inhumans that defined themselves as "metabolomics" studies, orwhich reported metabolic fingerprints, profiles, or signatures.Studies considering disease prediction, diagnosis, progression ortreatment response, and studies using any biologic media wereconsidered. The references of each identified study were screenedfor further qualifying manuscripts. Studies in animal models andin cell lines were not included. Where the full texts were notavailable the authors were contacted.

IntroductionProstate cancer represents the second leading cause of cancer

mortality in many western countries and accounted for morethan 28,000 deaths in the United States in 2013 (1). However,key questions remain on how best to diagnose and managethis cancer. In particular, the identification of the men atgreatest risk from lethal prostate cancer, the prediction oftreatment response, and the prediction of recurrence remainchallenging.

Most prostate cancers are first found during screening with aprostate-specific antigen (PSA) blood test, alone or in combi-nation with a digital rectal exam, followed by a diagnosticbiopsy and potentially imaging if there is a suspicion of cancerspread. When prostate cancer is diagnosed, PSA levels areutilized in tumor staging and tracking cancer progression.Initial treatment may be in the form of a radical prostatectomy,radiotherapy, and/or androgen deprivation therapy (ADT),which is based on the fact that prostate cancer growth isdependent on the predominant male hormone testosterone.After treatment, PSA testing may again be employed to assessdisease recurrence. However, there are well documented limita-tions at each of these stages. Only around 25% of men with anelevated PSA level, defined as > 4.0 ng/mL, are diagnosed withprostate cancer at biopsy, and conversely false-negatives arealso common (2). Biopsies frequently miss cancer due to tumorheterogeneity, necessitating the need for multiple repeat biop-sies which are potentially hazardous to patients and technicallychallenging (3). Radical prostatectomies are associated withfrequent comorbidities, including erectile dysfunction andincontinence, but are associated with better survival outcomesthat radiotherapy (4). Similarly, ADT has a number of adverseside effects and, on average, is only effective for two to threeyears before the emergence of castration-resistant prostate

1Department of Epidemiology, Harvard T.H. Chan School of PublicHealth, Boston, Massachusetts. 2Channing Division of Network Med-icine, Department of Medicine, Brigham and Women's Hospital andHarvard Medical School, Boston, Massachusetts. 3Koch Institute forIntegrativeCancerResearch atMassachusetts Instituteof Technology,Cambridge, Massachusetts. 4Department of Medical Oncology, Dana-Farber Cancer Institute, Boston, Massachusetts. 5Broad Institute ofHarvard and Massachusetts Institute of Technology, Cambridge, Mas-sachusetts. 6Department of Nutrition, Harvard T.H. Chan School ofPublic Health, Boston, Massachusetts.

Corresponding Author: Rachel S. Kelly, Channing Division of Network Medicine,Brigham and Women's Hospital and Harvard Medical School, 181 LongwoodAvenue, Boston, MA 02115. Phone: 617-525-0065; Fax: 617-432-1643; E-mail:hprke@channing.harvard.edu

doi: 10.1158/1055-9965.EPI-15-1223

�2016 American Association for Cancer Research.

CancerEpidemiology,Biomarkers& Prevention

www.aacrjournals.org 887

on June 16, 2020. © 2016 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

Published OnlineFirst April 6, 2016; DOI: 10.1158/1055-9965.EPI-15-1223

cancer (CRPC), an incurable and often fatal disease. There arecurrently no known methods to predict duration of treatmentresponse or to identify those men who will respond mosteffectively (5, 6). Furthermore, due to the aforementionedissues with both screening and biopsy, overtreatment repre-sents a significant problem in the management of prostatecancer; the majority of diagnoses will not prove fatal, but thereis a subset of men with aggressive and lethal disease. Identifyingthese men at the earliest stage possible is of paramount impor-tance to reduce to mortality from prostate cancer.

A number of additional biomarkers are now being exploredto try and address these questions and challenges. This has beenfacilitated, to a large extent, by developments in high-through-put sequencing technologies. Gene expression signatures havebeen demonstrated to have the ability to group patients withCRPC into high- and low-risk group (7) and the quantificationof circulating tumor DNA in the peripheral blood has beenshown to determine prognosis and monitor treatment effects(8). However, the development of novel prostate cancer bio-markers is still in its infancy. Increasingly, metabolomics isbeing explored to address this need. This high-throughputmethodology has the dual benefit of identifying biomarkerswhile also enabling a better understanding of the underlyingdisease mechanisms.

Metabolomics has been applied as an interdisciplinary"omics" science combining pattern recognition approachesand bioinformatics with epidemiology, analytical biochemis-try, and biology in the study of the "metabolome" (9), whichhas been defined as "the quantitative complement of all of thelow molecular weight molecules present in cells in a particularphysiologic or developmental state" (10). Metabolomics pro-vides a downstream measure of a whole system's activity thatreflects the genome, epigenome, transcriptome and proteome,and their interactions with the environment (11). The meta-bolome, as a measure of systemic activity, has also beenshown to be indicative of the disease state (12). In particular,neoplastic cells are known to possess unique metabolic sig-natures (13), including aerobic glycolysis (also known as theWarburg effect characterized by increased glucose uptake andlactate production) and production of choline-containingcompounds (9, 14–16). Accordingly, a number of studieshave attempted to capture metabolomic biomarkers of pros-tate cancer.

Prostate cancer represents a particularly attractive model formetabolite profiling. First, there is strong evidence to suggestthat dysregulation of metabolism plays an important role in thedevelopment and progression of this malignancy. One of themost consistently cited risk factors is metabolic syndrome; acollection of pathophysiologic entities including visceral obe-sity, insulin resistance, low HDL cholesterol, high triglycerides,elevated C-reactive protein, and low adiponectin levels (17).This syndrome is associated with chronic inflammation andhigh concentrations of inflammation-related markers, whichare thought to enhance tumor growth (18). Second, the healthyprostate is known to exhibit a unique metabolism to producethe components of prostatic fluid: PSA, spermine, myo-inosi-tol, and citrate. In fact, the levels of citrate in the prostate areorders of magnitude higher than anywhere else in the body(19, 20). In addition to the metabolic features common to allmalignancies, neoplastic prostate cells also lose the capacity toaccumulate zinc which is thought to inhibit the ability to

accumulate citrate. The metabolomic alterations reflecting thisphenomenon, which is unique to prostate cancer cells, arehypothesized to result in a distinctive and specific metabolomethat can be captured through metabolomic profiling.

Selected StudiesA search of the peer-reviewed literature identified a total of

33 qualifying studies (Table 1). The term "metabolomics"was not coined until 2002 by Fiehn and colleagues (21).However, researchers have been exploring the measurementsof multiple metabolites as potentially useful biomarkers ofprostate cancer for many years. Therefore, three studies pre-dating this specific terminology which attempted to charac-terize a "metabolomic profile" are also included. Since 2002,there has been a steady increase in the number of metabolo-mics studies, with more than half of the selected studiespublished from 2011 onwards.

Study Designs and MethodsThe included studies fell into five broad groups based on their

stated aims; studies attempting to identify predictive biomarkersprior to diagnosis (n ¼ 2; refs. 22, 23), studies aiming to distin-guish malignant from benign (n ¼ 19; refs. 24–42), studiesconsidering biomarkers of tumor aggressiveness (n ¼ 1; ref. 43),studies investigating the effect of therapy (n¼ 3; refs. 44–46), andstudies consideringmultiple outcomes (n¼ 8; refs. 16, 47–53). Incommonwith themajority of the existingmetabolomics literature(54), all studies were case–control in design, including two nestedwithin cohorts (22, 23). Twenty-two studies used external con-trols who were either healthy (n¼ 13 studies; refs. 22, 23, 34, 37–42, 49–52), suffering frombenignprostatic hyperplasia (BPH;n¼3; refs. 26, 29, 53), non-recurrent cases (n¼ 2; refs. 45, 48), or theyincluded multiple control groups (n ¼ 4; refs. 25, 35, 36, 43). Inthe remainder, the prostate cancer cases acted as their own con-trols through the use of matched biologic samples or the mea-surement of metabolites pre-and post-therapy. The most com-monly utilized biologic sample was prostate tissue (n ¼ 14),followed by blood (n¼ 10), urine (n¼ 5), and expressed prostaticsecretions (n¼1 study). In addition, three studies used a variety ofmedia (Table 1).

The two most commonly used methods for performingmetabolomics studies are mass spectrometry (MS) and nuclearmagnetic resonance spectroscopy (1H-NMR). MS involves ion-ization of chromatographically separated chemical compoundsand detection based on their mass-to-charge ratio and retentiontime during chromatography. Consequently, detection of meta-bolites by MS is biased towards those that ionize most effi-ciently. Nevertheless, MS it is the more sensitive of the twotechniques and can be performed with smaller amounts of abiologic sample, an important consideration in large-scalehuman studies. The second method, NMR, subjects a sampleto an electromagnetic field and measures the characteristicradio waves displayed by each compound in the sample inresponse to changes in the magnetic field. For a mixture ofmetabolites in a biologic sample the different patterns of energyrelease are represented as peaks in a chromatogram, and thearea of the peaks is proportional to the concentration of eachmetabolite (19). NMR is a more analytically reproduciblemethod, does not require sample separation, and provides

Kelly et al.

Cancer Epidemiol Biomarkers Prev; 25(6) June 2016 Cancer Epidemiology, Biomarkers & Prevention888

on June 16, 2020. © 2016 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

Published OnlineFirst April 6, 2016; DOI: 10.1158/1055-9965.EPI-15-1223

Table

1.Metab

olomicsstud

iesofprostatecancer

Biologic

med

iaAutho

rsSa

mple

Primaryaim/o

utco

me

Tech

nology

Comparisongroup

Population

Tissue

Hallid

ayet

al.(1988;ref.24

)Surgically

remove

dprostatetissue

Disting

uish

neoplasticfrom

non-

neoplastic

tissue

13C-N

MRspectroscopy

Adjacent

hyperplastic

tissue

7prostatecancer

patients,USA

Schiebleret

al.(1993;

ref.25

)Prostatetissue

from

prostatectomy

Disting

uish

aden

ocarcinoma,

BPHan

dno

rmal

periphe

ral

zone

tissue

1 H-N

MRspectroscopy

BPHan

dno

rmal

prostatetissue

13prostatecancer

patients,USA

Hah

net

al.(1997;

ref.26

)Prostatetissue

from

TURPorRP

Disting

uish

maligna

ntfrom

ben

igntissue

1 H-M

RS

BPHpatients

50prostatecancer

patients,Can

ada

Men

ardet

al.(20

01;ref.44)

Prostatebiopsies

Disting

uish

maligna

ntfrom

ben

igntissue

follo

wing

radiotherap

y

1 H-M

RS

Ben

ignprostatetissue

35prostatecancer

patients,Can

ada

Swan

sonet

al.(20

03;

ref.27

)Postsurgical

prostate

tissue

Disting

uish

maligna

ntfrom

ben

igntissue

(1HHR-M

AS)NMR

spectroscopy

Ben

ignprostatetissue

26prostatecancer

patients,USA

Che

nget

al.(20

05;

ref.28

)Prostatetissue

from

RP

Diagno

stic

biomarkers

of

prostatemaligna

ncy

1 HHR-M

ASMRS

Ben

ignprostatetissue

82prostatecancer

patients,USA

Maxeine

ret

al.(20

10;ref.4

5)Prostatetissue

from

need

lebiopsy

Biochem

icalrecurren

ceafterRP

1 HHR-M

ASMRS

Tissuefrom

casesthat

did

not

recur

16recurren

tprostatecancer

cases

and32

non-recurren

tcases,USA

Komoroskiet

al.(20

11;ref.2

9)

Prostatetissue

from

TURPorRP

Disting

uish

maligna

ntfrom

ben

igntissue

31PNMR

BPHpatients

8prostatecancer

patientsan

d13

BPHpatients,USA

Shu

ster

etal.(20

11;ref.3

0)

Postoperativebiopsy

tissue

Diagno

stic

biomarkers

of

prostatemaligna

ncy

GC-M

San

dLC

/MS-M

SBen

ignprostatetissue

8prostatecancer

patients,USA

Brownet

al.(20

12;ref.3

1)Postoperativebiopsy

tissue

Diagno

stic

biomarkers

of

prostatemaligna

ncy

GC-M

San

dLC

/MS

Ben

ignprostatetissue

Rea

nalysisofShu

ster

etal.(20

11;

ref.30

)population

Giskeødeg

a� rdet

al.(20

13;ref.4

7)Prostatetissue

from

RP

Diagno

stic

biomarkers

and

biomarkers

oftumor

aggressiven

ess

HR-M

ASMRS

Norm

alad

jacent

tissue

48prostatecancer

patients,Norw

ay

Kam

ietal.(20

13;ref.3

2)Surgically

remove

dprostatetissue

Disting

uish

maligna

ntfrom

norm

alprostatetissue

CE-TOF/M

SNorm

alprostatetissue

7prostatecancer

patients,Japan

Liet

al.(20

13;ref.33

)Prostatetissue

from

RP

Iden

tificationofprostate

carcinogen

esisrelevant

pathw

ays

LC/G

C-M

SBen

ignad

jacent

tissue

Rea

nalysisofS

reekum

aret

al(2009;

ref.52

)population

McD

unnet

al.(20

13;ref.16)

Prostatetissue

from

RP

Diagno

stic

biomarkers

of

prostatemaligna

ncyan

dbiomarkers

ofbiological

recurren

ce

GC-M

San

dLC

/MS

Ben

ignprostatetissue

331prostatecancer

patientsfrom

twoindep

enden

tco

horts,USA

Blood

Oslet

al.(20

08;ref.4

9)

Serum

Diagno

stic

biomarkers

of

prostatemaligna

ncyan

dbiomarkers

oftumor

aggressiven

ess

FIAMS/M

SorLC

/MS-M

SBloodserum

from

healthy

controls

206prostatecancer

patientsan

d114

healthyco

ntrols,A

ustria

Lokh

ovet

al.(20

10;ref.3

4)

Plasm

aDiagno

stic

biomarkers

of

prostatemaligna

ncy

MicrO

TOF-Q

MS

Bloodfrom

healthyco

ntrols

40prostatecancer

patientsan

d30

healthyco

ntrols,R

ussia

Fan

etal.(20

11;ref.5

3)Serum

Diagno

stic

biomarkers

of

prostatemaligna

ncyan

dbiomarkers

oftumor

aggressiven

ess

13CNMRspectroscopy

Bloodserum

from

BPHpatients

42Prostatecancer

patientsan

d14

BPHpatients,Irelan

d

Miyag

iet

al.(20

11;ref.5

0)

Plasm

aDiagno

stic

biomarkers

of

prostatemaligna

ncyan

dbiomarkers

oftumor

aggressiven

ess

HPLC

-ESI-MS

Bloodfrom

gen

der

andag

ematched

controls

134prostatecancer

patientsan

d666

cancer-freeco

ntrols,Jap

an

Say

loret

al.(20

12;ref.46)

Plasm

aMetab

olic

effectsofAnd

rogen

dep

rivationtherap

yGC-M

San

dLC

/MS

Pre-the

rapybloodsamples

36prostatecancer

patientstrea

ted

withADT,U

SA

(Con

tinu

edon

thefollowingpag

e)

Metabolomic Biomarkers of Prostate Cancer

www.aacrjournals.org Cancer Epidemiol Biomarkers Prev; 25(6) June 2016 889

on June 16, 2020. © 2016 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

Published OnlineFirst April 6, 2016; DOI: 10.1158/1055-9965.EPI-15-1223

Table

1.Metab

olomicsstud

iesofprostatecancer

(Cont'd)

Biologic

med

iaAutho

rsSa

mple

Primaryaim/o

utco

me

Tech

nology

Comparisongroup

Population

Blood

Zho

uet

al.(20

12;ref.40)

Plasm

aDiagno

stic

biomarkers

of

prostatemaligna

ncy

ESI-MS/M

SBloodplasm

afrom

prostate-

cancer

free

controls

105prostatecancer

patientsan

d36

maleprostate-cancer

free

controls,U

SA

Hua

nget

al.(20

14;ref.5

1)Serum

Progno

stic

biomarkers

and

biomarkers

oftherap

eutic

ben

efit

LC/M

SBloodfrom

age-matched

healthyco

ntrols

18un

trea

tedprostatecancer

patients,36

prostatecancer

patientsreceivingen

docrine

therap

y,18

healthymen

,China

Mond

ulet

al.(20

14;ref.2

2)Serum

Predictive

biomarkers

of

prostatemaligna

ncy

LC/M

San

dGC-M

SBloodfrom

agean

ddateof

baselinebloodsample

matched

controls

74prostatecancer

casesan

d74

controlsselected

from

theAlpha

-Toco

phe

rol,Beta-Carotene

Can

cerP

reve

ntionStudyco

hortof

malesm

okers,F

inland

Mond

ulet

al.(20

15;ref.23

)Serum

Predictive

biomarkers

of

prostatemaligna

ncyan

dag

gressiven

ess

LC/M

San

dGC-M

SBloodfrom

agean

ddateof

baselinebloodsample

matched

controls

200prostatecancer

casesan

d20

0co

ntrolsselected

from

theAlpha

-Toco

phe

rol,Beta-Carotene

Can

cerP

reve

ntionStudyco

hortof

malesm

okers,F

inland

Zan

get

al.(20

14;ref.4

1)Serum

Diagno

stic

biomarkers

of

prostatemaligna

ncy

LC/M

SBloodfromag

ematched

healthy

controls

64prostatecancer

casesan

d50

healthyindividua

ls,U

SA

Urine

Cho

etal.(20

09;ref.3

5)Urine

Disting

uish

prostatecancer

patients,BPHpatientsan

dhe

althyco

ntrols

LC/M

S-M

SUrine

from

patientswithben

ign

prostatic

hyperplasiaan

dhe

althyco

ntrols

18Prostatecancer

patients,16

BPH

patients,an

d18

healthymen

,Korea.

Wuet

al.(20

11;ref.3

6)

Urine

Diagno

stic

biomarkers

of

prostatemaligna

ncy

GC-M

SUrine

from

patientswithBPH

andhe

althyco

ntrols

20Prostatecancer

patients,8BPH

patients,20

healthyco

ntrols,

China

Gam

aged

araet

al.(20

12;ref.37

)Urine

Diagno

stic

biomarkers

of

prostatemaligna

ncy

LC/M

S-M

SUrine

from

healthyco

ntrols

63prostatecancer

patientsan

d68

healthyco

ntrols,U

SA

Zha

nget

al.(20

13;ref.38

)Urine

Diagno

stic

biomarkers

of

prostatemaligna

ncy

LC/H

RMS

Urine

from

healthyco

ntrols

60prostatecancer

patientsan

d30

healthyco

ntrols,H

ong

Kong

Struck-Le

wicka

etal.(20

15;ref.42)

Urine

Diagno

stic

biomarkers

of

prostatemaligna

ncy

GC-M

San

dLC

/MS

Urine

from

agean

dBMIm

atched

healthyco

ntrols

32prostatecancer

casesan

d32

healthyvo

luntee

rs,P

oland

Prostatic

secretions

Serko

vaet

al.(20

08;ref.3

8)

Exp

ressed

prostatic

secretions

Diagno

stic

biomarkers

of

prostatemaligna

ncy

1 H-N

MRspectroscopy

Prostatic

secretions

from

healthyco

ntrols

52Prostatecancer

patientsan

d26

healthyco

ntrols,U

SA

Multiple

Sreekum

aret

al.(20

09;ref.52

)Prostatetissue

,urine

andplasm

aDiagno

stic

biomarkers

of

prostatemaligna

ncyan

dbiomarkers

oftumor

aggressiven

ess

GC-M

San

dLC

/MS

Ben

ignad

jacent

prostatetissue

,an

durine/plasm

afrom

healthyco

ntrols

59prostatecancer

patientsan

d51

cancer-freeco

ntrols,U

SA

Thy

sellet

al.(20

10;ref.4

3)Prostatetissue

,plasm

aan

dbone

Biomarkers

oftumor

aggressiven

ess

GC-M

SNon-maligna

nttissue

and

plasm

asamplesfrom

men

withben

ignprostatedisea

sean

dno

rmal

bone

13prostatecancer

patientswith

metastases,13

prostatecancer

caseswitho

utmetastasesan

d30

men

withben

ignprostatedisea

se,

Swed

enStableret

al.(20

11;ref.4

8)

Urine

andserum

Biomarkers

oftumor

aggressiven

essan

dbiological

recurren

ce

GC-M

SUrine

andbloodfrom

non-

recurren

tcases

Urine

:25recurren

tprostatecancer

casesan

d29

non-recurren

tcases.

Blood:2

8recurren

tprostate

cancer

casesan

d30

non-recurren

tcases,USA

Abbreviations:A

DT,and

rogen

dep

rivationtherap

y;CE-TOF,cap

illaryelectropho

resis-timeoffl

ight;ESI,electrosprayionization;FIA,flowinjectionan

alysis;G

C,gas

chromatography

;HP,highperform

ance;H

R,highresolution;

LC,liquidchromatography

;MAS,m

agican

glespinning

;MS,m

assspectrometry;M

S/M

S,tan

dem

massspectrometry;N

MR,nuclearmag

neticresona

nce;TURP,transurethralresectionofthe

prostate;RP,rad

icalprostatectomy.

Kelly et al.

Cancer Epidemiol Biomarkers Prev; 25(6) June 2016 Cancer Epidemiology, Biomarkers & Prevention890

on June 16, 2020. © 2016 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

Published OnlineFirst April 6, 2016; DOI: 10.1158/1055-9965.EPI-15-1223

both quantitative and structural information (55). However,because sensitivity is limited, it is only able to detect a smallerrange of metabolites (2, 19). The majority (n ¼ 21) of theselected studies performed metabolomic profiling using massspectrometry (MS), with the remainder employing nuclearmagnetic resonance (NMR)–based techniques.

Metabolite identificationMetabolomic studies generally fall into two classes; (1) targeted

studies which focus on the, often quantitative, measure of specificmetabolites, sometimes selectedon thebasis of biologic rationale,and (2) untargeted studies with no a priori hypothesis whichmeasure a much larger number of metabolites, but the exactidentity of many metabolites detected by untargeted MS is notapparent without further study. Seventeen studies took an untar-geted approach to the search for discriminatory metabolites, withonly 12 employing a targeted search. The targeted approachintroduces an inherent bias as it is only able to detect the specificmetabolites or metabolite classes, such as methionine metabo-lites (48), corticoids (35), or lipids (29, 40), which have beenpreviously determined. However, it should be noted that thequantitative nature of the targeted studies offers the distinctadvantage of easier translation into a clinical setting. Even witha "hypothesis-free" approach there is not a single analyticalmethodology that can, even closely, cover the whole metabo-lome, which is estimated to lie within the range of 104 to 105

metabolites (54). Thus, untargeted approaches also carry withthem some bias based on their chosen methods such as thechromatography and detection techniques used to measuremetabolites.

The number of metabolites identified in the selected studiesvaried by orders of magnitude and was dependent on themethods and technologies employed. Some studies reportedthe number of peaks, spectral regions, or metabolite features,which may not necessarily correspond to individual metabo-lites. Furthermore, while some studies included all metabolitesin their analyses, others restricted analysis to those metaboliteswhich could be annotated. This may complicate the interpre-tation and translation of the findings. In those studies includingall metabolites, the unknowns cannot be biologically inter-preted and potentially important information linking the meta-bolites with the disease process may be missed; furthermore,any pathway analysis will be inherently biased towards theknown metabolites. Conversely, studies restricting to knownmetabolites may miss important components in metabolomicsignatures reducing their discriminatory ability. When compar-ing between the two types of study studies, it is difficult to knowwhether replication was not achieved because it did not exist, orbecause of differences in the metabolites compared. To date,the definitive annotation of identified metabolites remains abottleneck in untargeted metabolomics studies; absolute iden-tification and quantification can only be made in the presenceof an authentic standard. Only seven of the included studieswere quantitative or semiquantitative in nature (16, 32, 35,39, 47, 48, 50).

Statistical analysesRegardless of the technologic methods used, complex analyt-

ical methods are required to deal with the multivariate, highlycollinear noisy datasets produced (56, 57). Metabolomics studiesoften employ pattern recognition or clustering techniques as a

means of reducing dimensionality including unsupervised meth-ods, such as principal components analysis (PCA) which convertsa set of observations into a set of linearly uncorrelated variablescalled principal components, and supervised methodologies,such as orthogonal partial least squares, discriminant analysis(OPLS-DA) a related method that is able to separate the compo-nents into predictive and uncorrelated information (9). Thesewere employed by 15 of the selected studies, with the others usingmore classical statistical methods, including t tests and regressionmodeling. One study used a novel feature selection algorithmtermed associative voting (49), which integrates class associationrule data mining and classification, and one utilized an entirelypathway-based approach: subpathway-GM (33) which mapsgenes and metabolites of interest to metabolic pathways toidentify biologically relevant subpathway regions [readers arereferred to (49) and (33) for full methodological details]. In fact,the subpathway-GMmethod revealed novel associations beyondthe original analyses of the same dataset using a more classicalstatistical approach (52). Thus pathway-based methods for ana-lyzing metabolomics data may help to provide an improvedmechanistic interpretation, but are limited by the fact that theyare entirely reliant on the underlying databases utilized (58), andcurrently comprise only a small percentage of the metabolomicsliterature.

ResultsThe selected studies indicated that distinct clusters based on

metabolite profiles could be identified in a variety of biologicmedia including blood (40, 41, 51), urine (36, 38, 42), tissue (16,28, 31, 32, 45, 47, 52), and expressed prostatic secretions (39).These clusterswere able to distinguish cancer frombenign (28, 31,32, 36, 38–42, 47), tumors by their degree of aggressiveness (16,32, 47), caseswho recurred (45), and thosewho respondedwell totherapy (51). Thedistinctionbetween the groups of interest on thebasis of theirmetabolic profiles suggested that thedevelopment ofmetabolomics-based biomarkers may be possible. Although allincluded studies were ultimately interested in the identification ofdiscriminatory metabolites or profiles, 17 studies specificallyaimed at the development and assessments of biomarkers eitherfor diagnostic or prognostic purposes, and evaluated the utility ofthese biomarkers using ROC curve analyses and indices of spec-ificity and sensitivity. In the first part of the results section, thefindings of these 17 studies (Table 2) will be explored. In thesecond part, these findings will be placed in a wider biologiccontext using the results of the remaining 16 papers (Table 3),which did not explicitly search for, or assess biomarkers. The finalpart will detail the attempts at replicating and validating thesefindings.

Part 1: Assessing BiomarkersBiomarkers of prostate cancer

Nine of the studies comparing malignant and nonmalignantprostate cancer samples reported AUCs for diagnostic accuracyranging from 0.67 for urinary sarcosine levels in a U.S.-basedcase–control study (52) to 0.982 for a biomarker profile includingphosphocholine and choline,whichwasdevelopedby comparingtumor to benign prostate tissue from the same patients (28).The second highest reported AUC was 0.973 from Zhou andcolleagues' case–control study. This blood-based profile included15 phosophocholine-containing species, and although an AUC

Metabolomic Biomarkers of Prostate Cancer

www.aacrjournals.org Cancer Epidemiol Biomarkers Prev; 25(6) June 2016 891

on June 16, 2020. © 2016 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

Published OnlineFirst April 6, 2016; DOI: 10.1158/1055-9965.EPI-15-1223

Table

2.Assessm

entoftheutility

ofproposedmetab

olomicsbiomarkers

Autho

rsNo.ofmetab

olites

Biomarke

rROCAUC

Sensitivity

Spec

ificity

Other

results

Conc

lusions

Validation

Aim

:maligna

ntvs.b

enign

Che

nget

al.(20

05;

ref.28

)36

metab

olitegroup

sThe

13th

principal

compone

nt,

pho

spho

cholinean

dcholine

0.982

The

classificationaccuracy

of

theprofile

was

92.3%

.The

profilewas

highlyco

rrelated

withPSAleve

ls,w

asab

leto

iden

tify

less

aggressive

tumors

andto

predict

perineu

ralinva

sion

Metab

oliteprofilescan

differentiate

maligna

ntfrom

ben

ignsamples,

iden

tify

aggressive

tumors

andpredicttumor

perineu

ralinva

sion

Comparisonto

histopatho

logy

find

ings

Oslet

al.(20

08;

ref.49)

112metab

olites

Biomarkerprofile

includ

ing

lysopho

spha

tidylcholin

es(LPC)withsaturatedfatty

acid

chains,serotonin,

amono

amine,

asparticacid

(Asp)an

dornithine

0.716–0

.864

Metab

olic

profilingofblood

samplescandisting

uish

healthycasesan

dco

ntrols

Cross

valid

ation

Serko

vaet

al.

(2008;ref.3

9)

10metab

olites

Citrate

0.89

Citrate,m

yo-ino

sitolan

dspermineco

ncen

trations

areinve

rselyassociated

withprostatecancer

risk,

andrepresent

potentially

importan

tag

e-dep

enden

tmarkers

ofprostate

cancer

inhu

man

EPS.

Prospective

valid

ationis

ong

oing

Myo

-ino

sitol

0.87

Spermine

0.79

Sreekum

aret

al.

(2009;ref.52

)Plasm

a:478

metab

olites,U

rine

:58

3metab

olites,

Plasm

a:626

metab

olites

Sarco

sine

inurinesedim

ent

0.71

Sarco

sine

leve

lswere

increa

sedin

thetissue

of

prostatecancer

patients

compared

withhe

althy

controls,b

utno

differences

weredetectedin

plasm

a.Sarco

sine

leve

lsin

tissue

andtheleve

lsoffive

other

metab

olites;uracil,

kynu

renine

,glycerol-3-

pho

spha

te,leu

cine

and

prolinewerealso

elev

ated

intheprogressionfromben

ign

tometastaticprostate

cancer

Sarco

sine

may

have

potentialas

adiagno

stic

biomarker

Sarco

sine

find

ingswere

valid

ated

inan

indep

enden

tpopulation,

andthroug

hcell

linestud

ies

Sarco

sine

inurine

superna

tant

0.67

Lokh

ovet

al.(20

10;

ref.34

)25

ofthemost

intense

pea

ksof1900

detectedions

acylcarnitine

arachidono

ylam

ine

0.97

0.86

Four

other

cancer

associated

metab

olites

werealso

iden

tified

Isolitho

cholic

acid,T

estosterone

sulfate/

Deh

ydroep

iand

rosterone

sulfate,

And

rosterone

sulfate/5a

-hy

drotestosterone

sulfate

Six

potentiald

iagno

stic

markers

forprostate

cancer,twoofw

hich

were

determined

toha

vehighe

rAUCsthan

thePSA

test

inthispopulation

Cross

valid

ationan

dco

mparison

totw

oother

statistical

metho

ds

(Continue

donthefollowingpag

e)

Kelly et al.

Cancer Epidemiol Biomarkers Prev; 25(6) June 2016 Cancer Epidemiology, Biomarkers & Prevention892

on June 16, 2020. © 2016 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

Published OnlineFirst April 6, 2016; DOI: 10.1158/1055-9965.EPI-15-1223

Table

2.Assessm

entoftheutility

ofproposedmetab

olomicsbiomarkers

(Cont'd)

Autho

rsNo.ofmetab

olites

Biomarke

rROCAUC

Sensitivity

Spec

ificity

Other

results

Conc

lusions

Validation

Aim

:maligna

ntvs.b

enign

Miyag

ietal.(20

11;

ref.50

)19

aminoacidsan

drelatedmolecules

Ala,Ile,O

rn,L

ys(downreg

ulated

),Gln,

Val,T

rpan

dArg

(upregulated

)

Line

ardiscrim

ination

analysismodel:

0.786

Metab

olic

profilescanbe

used

todisting

uish

prostatecancer

cases

from

controls

Cross

valid

ation

Cond

itiona

llogistic

regressionmodel:

0.783

Wuet

al.(20

11;

ref.36

)81(ofwhich

59co

uldbe

iden

tified

)First

threeco

mpone

ntsof

theprostatecancer

model

based

onnine

discrim

inatory

metab

olites:Pyrim

idine,

Creatinine,

Purine,

Gluco

pyran

oside,

Xylopyran

ose

and

Ribofurano

side

(downreg

ulated

),Propen

oic

acid,

Dihyroxybutan

oic

acid

andXylonicacid

(upregulated

)(discrim

inatory

metab

olites

0.943

Sarco

sine

leve

lsdid

not

significantly

differbetwee

nprostatecancer

casesan

dhe

althyco

ntrols

Urina

rymetab

olomic

profilesmay

have

potentialdiagno

stic

ability

No

Zho

uet

al.(20

12;

ref.40)

390lip

idspecies

Top15

species,allo

fwhich

containe

dpho

spho

choline

0.973

93.60%

90.1%

LPC(22:6)dem

onstrated

the

most

significant

difference

inplasm

aco

ncen

trations

betwee

ncasesan

dco

ntrol.

Three

lipid

classes,

pho

spha

tidyle-

than

olamine(PE),ethe

r-linkedpho

spha

tidyle-

than

olamine(ePE)an

dethe

r-linked

pho

spha

tidylcholin

e(ePC)co

uldbe

considered

asbiomarkers

indiagno

sisofprostate

cancer

No

Zha

nget

al.(20

13;

ref.38

)52

00features

ureidoisobutyric

acid,

indolylacryloyg

lycine

,acetylvanilalininean

d2-

oxo

glutarate

0.896

83.3%

83.3%

Nosignificant

differencein

sarcosine

leve

lsbetwee

ncasesan

dco

ntrols

The

combined

biomarker

hasdiagno

sticpotentialin

prostatecancer

that

isco

mparab

leto

PSA

yes-find

ingswereva

lidated

ina

geo

graphically

indep

enden

tco

hort

Giskeødeg

a� rdet

al.

(2013;ref.4

7)23

metab

olites

Leve

lsofcitrate,tau

rine

and

crea

tine

(downreg

ulated

)GPC,P

Cho

,Cho

,and

glycine

(upregulated

)

86.9%

85.2%

Metab

olic

profileswere

significantly

correlated

with

Gleasonscore

Metab

olic

profileswereab

leto

disting

uish

norm

alan

dtumortissue

,and

indolent

from

aggressiveprostate

cancer

No

Zan

get

al.(20

14;

ref.41)

Biomarkerpan

elco

ntaining

top40

features

Biomarkerpan

elco

ntaining

top40features

92.1%

94.4%

Biomarkerpan

eldisplaye

d93%

accuracy

for

discrim

inatingprostate

cancer

casesfrom

healthy

controls

Leve

lsoffattyacids,am

ino

acids,lysopho

spho

lipids,

andbile

acidsae

dysregulated

inprostate

cancer

patients,proving

furthe

rinsight

into

the

metab

olic

alterations

associated

with

carcinogen

esis.

Internal

cross-validation

(Continue

donthefollowingpag

e)

Metabolomic Biomarkers of Prostate Cancer

www.aacrjournals.org Cancer Epidemiol Biomarkers Prev; 25(6) June 2016 893

on June 16, 2020. © 2016 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

Published OnlineFirst April 6, 2016; DOI: 10.1158/1055-9965.EPI-15-1223

Table

2.Assessm

entoftheutility

ofproposedmetab

olomicsbiomarkers

(Cont'd)

Autho

rsNo.ofmetab

olites

Biomarke

rROCAUC

Sensitivity

Spec

ificity

Other

results

Conc

lusions

Validation

Aim

:maligna

ntvs.b

enign

Struck-Le

wicka

etal.(20

15;ref.42)

1132

features

235features

selected

from

LC-TOF/M

San

alyses

inpositive

ionizationmode

R2¼

0.756

,Q2¼

0.579

Metab

olites

invo

lved

inam

inoacid,o

rgan

icacid,

sphing

olip

id,fatty

acid

andcarbohy

drate

pathw

aysshowed

significant

differences

intheurineofprostate

cancer

patientsco

mpared

tohe

althyco

ntrols.

Results

suggestprostate

carcinogen

esisis

associated

witha

metab

olic

phe

notype

promoting

abno

rmal

cell

growth

andintensivecell

proliferation.

Internal

cross-validation

248features

selected

from

LC-TOF/M

San

alyses

inne

gativeionizationmode

R2¼

0.763,

Q2¼

0.508

28features

selected

from

GC-M

San

alyses

R2¼

0.789,Q

2¼

0.711

AIM:N

orm

alvs.B

HP

Hah

net

al.(1997;

ref.26

)450

pointspectral

region

6discrim

inatory

spectral

regions

includ

ingthose

correspond

ingto

citrate,

taurinean

dglutamate

100.0%

95.5%

Ove

rallclassificationaccuracy

of96.6%

(Trainingset:

100%,T

estset:92.7%

)

1HHMRScanbeused

toreliably

disting

uish

betwee

nben

ignan

dmaligna

ntprostatictissue

Datawas

partitione

dinto

atraining

andtest

set

Fan

etal.(20

11;

ref.53

)9metab

olites

Acetoacetate,

cystine,

form

ate,

glutamate,

lysine

,tyrosine

,lipids

0.876

Metab

olomicswas

out

perform

edbyproteomics

atdisting

uishingBPH

from

prostatecancer

patients

No

Wuet

al.(20

11;

ref.36

)81(ofwhich

59co

uldbe

iden

tified

)First

threeco

mpone

ntsofa

prostatecancer

model

based

onfive

discrim

inatory

metab

olites:

dihyroxybutan

oic

acid

andxylonicacid

(upregulated

),pyrim

idine,

xylopyran

ose,

andribofurano

side

(downreg

ulated

)

0.825

Sarco

sine

leve

lsdid

not

significantly

differbetwee

nprostatecancer

casesan

dBPHpatients

Urina

rymetab

olomic

profilesmay

have

potentialdiagno

stic

ability,b

utsarcosine

had

nodiagno

stic

potentialin

thispopulation.

No

Aim

:Tum

orag

gressiven

ess

Oslet

al.(20

08;

ref.49)

112metab

olites

Profile

includ

ed:

Phsopha

tidyl

choline,

serotonin,

aspartate,

ornithine

,threo

nine

,arachidonicacid,fatty

acids,sphing

omye

lins,

tyrosine

andleucine

0.522

-0.673

(dep

enden

ton

featureselection

metho

d)

Metab

oliteprofiles

displaye

dlittleab

ility

todiscrim

inatebyGleason

grade

Cross

valid

ation

Fan

etal.(20

11;

ref.53

)9metab

olites

Acetoacetate,

cystine,

form

ate,

glutamate,

lysine

,tyrosine

,lipids

Gleason5

vs.

Gleason7:

AUC:

0.532

,

Metab

olomicsdoes

not

perform

aswella

sproteomicsin

classifying

tumortypes

No

Organ

confi

nedvs.

non-organ

confi

ned:A

UC

0.311

(Continue

donthefollowingpag

e)

Kelly et al.

Cancer Epidemiol Biomarkers Prev; 25(6) June 2016 Cancer Epidemiology, Biomarkers & Prevention894

on June 16, 2020. © 2016 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

Published OnlineFirst April 6, 2016; DOI: 10.1158/1055-9965.EPI-15-1223

Table

2.Assessm

entoftheutility

ofproposedmetab

olomicsbiomarkers

(Cont'd)

Autho

rsNo.ofmetab

olites

Biomarke

rROCAUC

Sensitivity

Spec

ificity

Other

results

Conc

lusions

Validation

Miyag

ietal.(20

11;

ref.50

)19

aminoacidsan

drelatedmolecules

Alanine

,isoleucine,

ornithine

,lysine

(downreg

ulated

),glycine

,valine,

tryp

topha

n,an

darginine(upregulated

)

Controlsvs.S

tageII

(B)patients:

0.764

Controlsvs.StageIII

(C)patients:

0.777

Metab

olic

profilescanbe

used

toclassify

prostate

cancer

stag

e

Cross

valid

ation

Controlsvs.StageIV

(D)patients:

0.873

Aim

:Norm

alvs.B

HP

McD

unnet

al.(20

13;

ref.16)

326co

mpoun

ds

iden

tified

inboth

coho

rts

5,6-dihyd

rouracil,choline

pho

spha

te,g

lycerol,an

dmethy

lpalmitate

combined

withGleason

score,serum

PSA,a

ndclinical

stag

e

0.62

Sarco

sine

was

significantly

elev

ated

onlyinthose

tissue

sections

withGleason

pattern

8orworseprostate

cancer

The

inclusionofmetab

olic

markers

improve

dthe

predictionoftumor

aggressiven

essco

mpared

toclinical

indices

alone

No,b

utrean

alysisof

Sreekum

ar'sdata

Aim

:Recurrence

Men

ardet

al.(20

01;

ref.44)

450

pointspectral

region

Biomarkerprofile

includ

ing

choline,

crea

tine

,glutamine,

andlip

ids

88.9%

92%

Ove

rallclassificationaccuracy

of91.4

%The

choline,

crea

tine

,glutamine,

andlip

idspectral

regions

dem

onstratediagno

stic

ability

Cross

valid

ation

Maxiene

ret

al.

(2010;ref.4

5)27

most

commonan

dintensespectral

metab

olic

regions

First

nine

principal

compone

nts:major

contributors

totheprofile

wereglutamine,

glutamate,

myo

-ino

sitol,

myo

-ino

sitol,

scylloinositol,pho

spho

ryl

choline,

spermine/

polyam

ines

0.78

Recurrencewas

predicted

withan

accuracy

of78

%.

The

iden

tified

tissue

metab

olomic

profiles

represent

potential

biomarkers

forthe

predictionofrecurren

ce

No

Stableret

al.(20

11;

ref.48)

8metab

olites

Homocysteine

0.74

Additionofserum

homocysteine

provided

the

greatestim

prove

men

tof

thelogisticregression

modelsco

mpared

tothe

basemodel

withPSAan

dbiopsy

Gleason(P

¼0.0007),follo

wed

by

cysteine

(P¼

0.0017),an

dcystathionine

(P¼

0.0037

)

Methionine

metab

olites

combined

withserumPSA

couldacta

sbiomarkersto

increa

setheab

ility

topredictag

gressive

prostatecancer

features

andea

rlybiochem

ical

recurren

ceove

ran

dab

ove

existent

clinical

variab

les

No

Homocysteine

þPSA

0.78

Cystathionine

0.7

Cystathionine

þPSA

0.79

Cysteine

0.8

Cysteineþ

PSA

0.82

Homocysteine

þcystathionine

þcysteine

þPSAþ

Gleasongrade

0.86

McD

unnet

al.(20

13;

ref.16)

326co

mpoun

ds

iden

tified

inboth

coho

rts

7-a-hy

droxy-3-oxo

-4-

cholesten

oate,

pregne

n-diold

isulfate,a

ndman

nosyltryptopha

nco

mbined

withGleason

score,serum

PSA,a

ndclinical

stag

e

0.64

Sarco

sine

was

significantly

elev

ated

onlyinthose

tissue

sections

withGleason

pattern

8orworseprostate

cancer

The

inclusionofmetab

olic

markers

improve

dthe

predictionofbiological

recurren

ceco

mpared

toclinical

indices

alone

No,b

utrean

alysisof

Sreekum

ar'sdata

Metabolomic Biomarkers of Prostate Cancer

www.aacrjournals.org Cancer Epidemiol Biomarkers Prev; 25(6) June 2016 895

on June 16, 2020. © 2016 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

Published OnlineFirst April 6, 2016; DOI: 10.1158/1055-9965.EPI-15-1223

Table

3.Results

from

themetab

olomicsstud

iesthat

did

notassess

biomarkerutility

Autho

rsNo.ofmetab

olites

Results

Validated

?Conc

lusions

Hallid

ayet

al.(1988;ref.2

4)

49metab

olites

Lactateleve

lsan

dlip

idsigna

lsarehighe

rintumor

tissue

,citrate

leve

lsaredecreased

.No

13CNMRspectroscopycanbeused

todifferentiate

neoplastic

andno

n-ne

oplastic

prostatetissue

Schiebleret

al.(1993)

13metab

olitepea

ksCitrate,A

cetate,ino

sitola

ndlactatediffered

significantly

betwee

nad

enocarcinomaan

dno

rmal

periphe

ralzo

netissue

,howev

erthere

wereanu

mber

ofsimilarities

betwee

nthe

spectraofBPHan

dad

enocarcinoma

No

The

citratestan

dardized

pea

karea

alone

cann

otbe

used

todiagno

seprostatead

enocarcinoma

Swan

sonet

al.(20

03)

8metab

olites

Gland

ular

tissue

:hea

lthy

tissue

hadsignificantly

highe

rleve

lsofcitrate,

Tau

rine

,myo

-ino

sitol,

andscyllo-ino

sitola

ndpolyam

ines,a

ndlower

leve

lsofcholine,

pho

spho

choline(PC),an

dglyceropho

spho

choline(G

PC)relative

totumortissue

.Stromaltissue

:hea

lthy

tissue

hadlower

leve

lsof

cholineco

mpoun

dsan

dhighe

rleve

lsof

Tau

rine

,myo

-ino

sitol,an

dscyllo-ino

sitol.

Cross

valid

ation

Distinctive

metab

olic

patternswereiden

tified

for

tumoran

dhe

althytissue

san

dforcancersof

varyinggrade;ho

wev

er,tissuetypemay

affect

the

find

ings.

Cho

etal.(20

09)

21co

rticosteroids

Urina

ryco

rtisollev

elsweresignificantlyhighe

rin

prostatecancer

patientsthan

inhe

althy

controlsan

dBHPpatients

Currentlyong

oingin

alarger

population

The

resultsindicatedysregulated

cortisol

metab

olism

inprostatecancer

patientsan

dsuggestthat

metab

olic

profilingmay

bethe

optimal

way

tomea

sure

this.

Thy

sellet

al.(20

10;ref.4

3)Bone

123metab

olites

of

which

49co

uldbe

iden

tified

.Tissue:

157

metab

olites

ofwhich

59co

uldbeiden

tified

.Plasm

a:179

metab

olites

ofwhich

50co

uldbeiden

tified

Bone

:58%

ofmetab

olites

differedbetwee

nno

rmal

andmetastaticbone

.Aminoacid

synthe

sisan

dmetab

olism

wereup

regulated

inmetastaticbone

,and

highleve

lsofcho

lesterol,

myo

-ino

sitol-1-pho

sphate,

citric

acid,

fumarate,

glycerol-3-pho

spha

te,a

ndfatty

acidsdetected.T

issue:

8%

ofmetab

olites

differedbetwee

ntissue

from

patientswithan

dwitho

utmetastasesinclud

ingfour

ofthose

significantly

increa

sedin

metastaticbone

:asparag

ine,

threonine

,fum

aric

acid,a

ndlinoleic

acid.P

lasm

a:15%

ofmetab

olites

differedin

theplasm

aofpatientswithan

dwitho

utbone

metastasesinclud

ingfour

iden

tified

inmetastaticbone

;glutamic

acid,

taurine,

andphe

nylalanine

andstea

ricacid.

Sarco

sine

was

foun

dto

beincrea

sedin

the

bone

ofm

enwithmetastaticdisea

sebut

notin

theirtumortissue

orplasm

a

No

Cho

lesterolisapossible

therap

eutictarget

for

advanced

prostatecancer.

Komoroskie

tal.(20

11)

4pho

spho

lipid

metab

olites

Pho

spho

etha

nolaminean

dglyceropho

spho

ethan

olamineleve

lsdiffered

significantly

betwee

nthecancer

andBPH

group

s.None

ofthefour

metab

olites

were

associated

withGleasonscore

No

The

semetab

olites

may

beuseful

inthediagno

sisof

prostatecancer

andmay

help

toexplain

thehigh

cholineresona

nceiden

tified

inother

stud

ies.

(Continue

donthefollowingpag

e)

Kelly et al.

Cancer Epidemiol Biomarkers Prev; 25(6) June 2016 Cancer Epidemiology, Biomarkers & Prevention896

on June 16, 2020. © 2016 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

Published OnlineFirst April 6, 2016; DOI: 10.1158/1055-9965.EPI-15-1223

Table

3.Results

from

themetab

olomicsstud

iesthat

did

notassess

biomarkerutility

(Cont'd)

Autho

rsNo.ofmetab

olites

Results

Validated

?Conc

lusions

Shu

ster

etal.(20

11;ref.3

0)

260metab

olites

83(32%

)ofmetab

olites

weresignificantly

different

betwee

ncancer

andben

igntissue

,with82at

highe

rleve

lsin

cancer

tissue

.Commonam

inoacids,long

chainfattyacids

andpho

spho

lipidswereincrea

sed.H

ighe

rleve

lsofuracil,kynu

renine

,glycerol-3-

pho

spha

te,leu

cine

andprolinewerereported

inag

reem

entwithSreekum

ar'sstud

ybut

sarcosine

was

below

thelim

itofdetection.

No

Molecularbiomarkersco

uldau

gmen

thistologyinthe

characterizationofdisea

se.

Brownet

al.(20

12;ref.3

1)26

0metab

olites

Ala,Ile,O

rnan

dLy

sweredownreg

ulated

inprostatecancer

patientsco

mpared

tohe

althy

controlsan

dGln,V

al,T

rpan

dArg

were

upregulated

No

Itispossibleto

determineametab

olomicsigna

ture

of

prostatecancer

Gam

aged

araet

al.(20

12;ref.37

)4metab

olites

Nodifferencein

themea

nleve

lsofproline,

kynu

renine

,uracilorGlycerol-3-pho

spha

tebetwee

nprostatecancer

casesan

dhe

althy

controls

Validating

Sreekum

ar's

find

ings

The

exploredbiomarkers

hadno

diagno

stic

or

progno

sticpotentialinprostatecancer,and

could

notdisting

uish

prostatecancer

from

other

maligna

ncies

Say

loret

al.(20

12)

292iden

tified

metab

olites

56metab

olites

chan

ged

significantly

from

baselineto

threemonths:M

ultiple

steroids,

Markers

oflip

idbeta-oxidation,

markers

of

omeg

a-oxidationan

dmarkers

ofinsulin

resistan

ce(2-hyd

roxybutyrate,

branchchain

keto-aciddeh

ydrogen

aseco

mplexproducts)

werelower.M

ost

bile

acidsan

dtheir

metab

olites

werehighe

r.

Cross

valid

ation

Iden

tified

nove

land

clinicallyim

portan

tADT-ind

uced

metab

olic

chan

ges

Kam

ietal.(20

13;ref.3

2)86(ofw

hich

39co

uldbe

absolutelyqua

ntified

)TCAcycleinterm

ediates,Succina

te,fum

arate,

malate,

pyruv

atean

dlactateleve

lswere

highe

r,an

dADPan

dpho

spho

enolypyruv

ate

lower

intumorthan

inno

rmal

tissue

.

Cross

valid

ation

Tum

ormetab

olic

profilescanhe

lpto

disting

uish

norm

alfrom

tumortissue

,and

tumorstag

e

Liet

al.(20

13)

626

metab

olites

53metab

olites

associated

withmetastatic

prostatecancer,16significant

pathw

ays

includ

ingam

inoacid

metab

olism,tryptopha

nmetab

olism,cysteinean

dmethionine

metab

olism,arachidonicacid

metab

olism

and

histidinemetab

olism

No

Iden

tified

nove

ldisea

sereleva

ntpathw

aysusingan

alternativestatisticalm

etho

donSreekum

aret

al.'s

data

Hua

nget

al.(20

14;ref.5

1)8,232

signa

lsIn

patientswho

did

notdev

elopcastration-

resistan

tprostatecancer

(CRPC)forat

least2

years,serum

deo

xycholic

acid

(DCA),

glyco

chen

odeo

xycholate

(GCDC),L

-tryptopha

n,doco

sapen

taen

oic

acid

(DPA),

arachidonicacid,d

eoxycytidinetripho

spha

te,

andpyridinolineleve

lsreve

rted

tone

arhe

althy

controllev

elsduringen

docrinetherap

y.In

contrast,the

metab

oliteleve

lsremaine

dab

norm

alin

patientswho

dev

eloped

CRPC

within1ye

ar

Validationcurren

tly

ong

oing

DCA,G

CDC,L

-tryptopha

n,DPA,a

rachidonicacid,

deo

xycytidinetri-pho

spha

te,a

ndpyridinoline

represent

potentialb

iomarkers

forev

alua

ting

patient

responseto

endocrinetherap

y.The

seresultssuggestarole

forcholesterolin

prostate

cancer

progression

(Continue

donthefollowingpag

e)

Metabolomic Biomarkers of Prostate Cancer

www.aacrjournals.org Cancer Epidemiol Biomarkers Prev; 25(6) June 2016 897

on June 16, 2020. © 2016 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

Published OnlineFirst April 6, 2016; DOI: 10.1158/1055-9965.EPI-15-1223

Table

3.Results

from

themetab

olomicsstud

iesthat

did

notassess

biomarkerutility

(Cont'd)

Autho

rsNo.ofmetab

olites

Results

Validated

?Conc

lusions

Mond

ulet

al.(20

14;ref.2

2)420

metab

olites

Circulating

1-stea

roylglycerolwas

inve

rsely

associated

withtherisk

ofd

evelopingprostate

cancer

upto

23ye

arsafterbloodco

llection.

The

mag

nitudeofthisassociationdidno

tdiffer

bydisea

seag

gressiven

ess.The

rewerealso

suggestive

inve

rseassociations

forglycerol

andalpha

-ketoglutarate.

Onlytheassociation

betwee

nalpha

-ketoglutarate

and

aggressive

prostatecancer

was

replicated

inasubseque

ntstud

yinclud

ing

different

participan

tsfrom

thesame

population.

The

resultssupportarole

fordysregulationoflip

idmetab

olism

inthedev

elopmen

tofp

rostatecancer

Mond

ulet

al.(20

15;ref.2

3)626

metab

olites

Strong

inve

rseassociations

betwee

nen

ergyan

dlip

idmetab

olites

particularly

glyceropho

spho

lipidsan

dfattyacidsan

dag

gressivecancer

wereobserved

with

aggressivedisea

serisk.T

hyroxine

and

trim

ethy

lamineoxidewereassociated

with

aggressivedisea

serisk

while

Alpha

-ketoglutarate

andcitratewereinve

rsely

associated

.Metab

olites

associated

with

nona

ggressivecancersinclud

ed20-

deo

xyuridine,

aden

osine

50-m

ono

pho

spha

te(A

MP),11-deh

ydroco

rticosterone

,21-

hydroxypregne

nolonemono

sulfate,

cotinine

,an

dhy

droxyco

tinine

.

Meta-an

alyses

with

thefind

ingsofa

previous

stud

yco

nfirm

edarole

forglyceropho

s-pho

lipidsan

dlong

chainfatty

acids

Prospective

stud

ydataindicatethat

seve

ral

circulatingglyceropho

spho

lipid,fatty

acid,ene

rgy

andrelatedmetab

olites

areinve

rselyassociated

withag

gressiveprostatecancer

upto

20ye

ars

priorto

diagno

sis.Metab

oliteassociations

vary

by

cancer

aggressiven

ess.

Kelly et al.

Cancer Epidemiol Biomarkers Prev; 25(6) June 2016 Cancer Epidemiology, Biomarkers & Prevention898

on June 16, 2020. © 2016 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

Published OnlineFirst April 6, 2016; DOI: 10.1158/1055-9965.EPI-15-1223

was not computed, Giskeødega�rd and colleagues also included

phosphocholine in their tissue-based biomarker profile, whichhad a sensitivity and specificity >85% (47). Three of the articles(34, 39, 52) reported the AUC for individual metabolites, withacylcarnitine ranking the highest (AUC: 0.97 in plasma; ref. 34).Carnitines were also identified in Struck-Lewicka and colleagues'diagnostic signature in urine. Similarly, another metabolite witha high individual AUC, citrate, which had an AUC of 0.89 inSerekova's study (39), was a component of Giskeødega

�rd and

colleagues' signature (47). The remaining studies reported onprofiles comprising multiple markers. Interestingly, there wassome crossover between the profiles developed in the case–control studies of Osl and colleagues (49), Miyagi and colleagues(50), Zang and colleagues (41), and Struck-Lewicka and collea-gues (42), with all including multiple amino acids such as lysine,glutamine, and ornithine in their profiles. Osl and colleagues (49)also included arachidonic acid in their signature, a compoundrelated to arachidonoyl amine, which was reported to have anAUC of 0.86 in Lokhov and colleagues' study (34). Furthermore,Osl and colleagues (49) included a number of phosphorylcho-lines, in agreement with the findings of Zhou and colleagues'targeted lipidomics analysis (40), while Zang and colleagues (41)identified multiple lysophospholipids. The sensitivity and spec-ificity of Zang and colleagues' diagnostic profile, which alsocontained metabolites of the steroid hormone biosynthesis path-way and bile acids, was 92%, and 94% respectively. Struck-Lewicka and colleagues also did not report AUC's but rather, theR2 and Q2 values for their partial least squares-discriminantanalysis model which were 0.789, 0.711 respectively, under thebest performing GC-MS–derived model, indicating a robust sig-nature with good predictive ability (42).

The studies by Osl (49), Miyagi (50), Zhou (40), Zang (41),and Lokhov (34) and colleagues developed their signatures inblood samples, and it is of note that there was little crossover interms of the specific constituent metabolites with the urine-based diagnostic signatures proposed by Zhang and colleagues(38), Struck-Lewicka and colleagues (42), or Wu and colleagues(36). All but two of the studies (28, 47), compared prostatecancer cases with healthy controls. However, the results ofGiskeødega

�rd and colleagues' and Cheng and colleagues' inter-

individual comparisons among prostate cases were markedlyconcordant with these healthy versus diseased comparisons,both in terms of the constituents of the signatures they devel-oped including phosphocholines, and citrate, and in the pre-dictive ability of these signatures. Similarly, study populationsize which ranged from 40 (20 prostate cancer cases and 20healthy controls) in Wu and colleagues' study (36) to 800 (134cases and 666 controls) in Miyagi and colleagues' study (50),did not appear to confer any advantages in terms of thepredictive ability of the developed signature.

In summary, the most promising candidate biomarkers fordistinguishing prostate cancer cases fromhealthy controls includesarcosine, choline, phosphocholines, phosphorylcholines, carni-tines, citrate, amino acids, arachidonoyl amine, and lysopho-spholipids (Table 2).

Biomarkers of benign prostatic hyperplasia and tumoraggressiveness

Wuand colleagues also reported on a biomarker profile with anAUC of 0.825 to distinguish benign prostatic hyperplasia (BPH)andprostate cancer patients constituting fivemetabolites (36). All

five (dihyroxybutanoic acid, xylonic acid, pyrimdine, xylopyra-nose, and ribofuranoside) were also constituents of the healthyversus diseased profile. Hahn and colleagues reported that MRSspectra could be used to classify benign and malignant prostatetissue with a sensitivity of 100%, a specificity of 96%, and anoverall classification accuracy of 97%, with citrate, glutamate, andtaurine playing important discriminatory roles (26). Fan andcolleagues' 9-feature profile had an AUC of 0.876 for distinguish-ing prostate cancer and BPH patients. Similar to their malignantversus benign analyses, this serum-based signature included glu-tamate, lysine, and lipid species, suggesting a possible dose-dependent relationship with the progression fromnormal to BPHto cancer. However Fan and colleagues' found that the signaturedid not perform well at distinguishing Gleason 5 from Gleason 7(AUC, 0.532) or organ-confined from non-organ–confined can-cers (AUC, 0.311; ref. 53).

Three other articles also developed biomarker profiles of tumoraggressiveness. Osl and colleagues compared Gleason 6 withGleason 8–10 cancers, and reported that the discriminatorymetabolites, which included a number of sphingomyelin lipids,were distinct from those that differed between normal and tumortissue (49). However, the AUCs were much lower in the tumoraggressiveness analyses suggesting discriminatory power is poor.Conversely, Miyagi and colleagues reported increasing AUCswhen comparing healthy controls to stage II patients (AUC,0.764), stage III patients (AUC, 0.777), and stage IV patients(AUC: 0.873) using a profile comprising 8 amino acids: alanine,isoleucine, ornithine, lysine, glutamine, valine, tryptophan, andarginine (50). McDunn and colleagues' study the AUC for acombination of four different biomarkers: 5,6-dihydrouracil,glycerol, methylpalmitate, and choline phosphate, was 0.62 fordiscriminating organ-confined from non-organ–confined pros-tate cancer (16).

In summary, the most promising candidate biomarkers foridentifying BPH and tumor aggressiveness include dihyroxybu-tanoic acid, xylonic acid, pyrimdine, xylopyranose, ribofurano-side citrate, glutamate, sphingomyelin lipids, amino acids, 5,6-dihydrouracil, glycerol, methylpalmitate, and choline phosphate(Table 2). Again, there was little difference in the results betweenthose studies that comparedhealthy individuals to prostate cancercases and those investigating intertumor differences within con-trols, or any differences due to study population size.

Biomarkers of disease recurrenceInterestingly, choline phosphate (phosphorylcholine) was also

identified as a major contributor to the metabolic profile predict-ing recurrence by Maxeiner and colleagues (45), which had anAUC of 0.78. This profile was based on the first nine principalcomponents of all the measured metabolites and the loadingsplots determined that myoinositol and spermine, both of whichwere identified in Serkova and colleagues' (39) diagnostic signa-ture, and glutamate, which was a component of Fan and collea-gues' (53) aggressiveness signature,weremajor contributors to therecurrence profile.

Similarly cysteine, a further component of Fan and collea-gues's aggressiveness signature, was investigated as a marker ofrecurrence by Stabler and colleagues (48). This metabolite wasfound to have an AUC of 0.82 when combined with PSA,outperforming two other methionine metabolites: homocyste-ine (AUC, 0.78) and cystathionine (AUC, 0.79). When thesethree metabolites were combined, the AUC was 0.86, providing

www.aacrjournals.org Cancer Epidemiol Biomarkers Prev; 25(6) June 2016 899

Metabolomic Biomarkers of Prostate Cancer

on June 16, 2020. © 2016 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

Published OnlineFirst April 6, 2016; DOI: 10.1158/1055-9965.EPI-15-1223

an increased ability to detect recurrence over clinical indicesalone.

Although Menard and colleagues (44) did not perform ROCanalysis their profile, including choline, creatine, glutamine, andlipids, but theywere able to identify amalignant biopsy followingradiotherapywith a sensitivity of 89%, a specificity of 92%, and anoverall classification accuracy of 91%.

Despite the differences in study designs,Menard and colleagues(44) and McDunn and colleagues (16) considered intertumordifferences, while Maxiener and colleagues (45) and Stabler andcolleagues (48) compared the recurrent and nonrecurrent cases intissue and in blood and urine, respectively; the results were largelyconcordant between the studies. It is also of note that the smallestpredictive ability was reported for the largest study (16). Insummary, the most promising candidate biomarkers for predict-ing disease recurrence included phosphorylcholine, myoinositol,spermine, glutamate, cysteine, choline, creatine, glutamine, andlipids (Table 2).

PSA testing and current gold standardsProstate cancer is a relative rarity amongmalignancies in that it

has been shown to be amenable to population-wide screeningprograms utilizing prostate-specific antigen (PSA), which is alsomonitored as a biomarker of biochemical recurrence (6). There-fore, novel biomarkers must perform better than this current goldstandard to be useful. Seven of the studies discussed aboveexplicitly compared their biomarkers to the use of PSA (34,39–41, 48, 52, 53). An often cited study reports an AUC for PSAtesting of 0.682 [95% confidence interval (CI) 0.67–0.69] forthe diagnosis of prostate cancer, and that a PSA cut-off value of4.1 ng/mL has a specificity of 93.8% but a sensitivity of only20.5% (59). Zhou and colleagues' multi-marker plasma metab-olite profiles outperformed these metrics in diagnosis (40),although parsimony must also be taken into account whenconsidering clinical translation. Similarly, Serkova and colleaguesreported AUCs for three metabolites that would outperform PSA,and added that unlike PSA, they are not associated with age,suggesting improved specificity (39). Although it should be notedthat Serkova was comparing the discriminatory ability of thesemetabolites in expressedprostatic secretions to ablood-basedPSAtest. Lokhov (34), Zang (41), and Stabler and colleagues (48)computed the utility of PSA in their respective study populationsfor the diagnosis of prostate cancer [Lokhov (34) and Zang andcolleagues (41)] and the prediction of recurrence (Stabler andcolleagues; ref. 48), and all reported that their blood-basedbiomarkers outperformed PSA. Although the serum biomarkerprofile developed by Fan and colleagues outperformed PSA atdistinguishing BPH from prostate cancer in their population, itwas comparable or inferior to PSA testing in discriminatingtumors by their degree of aggressiveness (53). Finally, Sreekumarand colleagues reported that the measurement of sarcosine inurine was superior to PSA at predicting a positive prostate cancerbiopsy within the clinical PSA gray zone of 2–10 ng/mL (52).

For the tissue-based diagnostic studies (26, 28, 47), thecurrent gold diagnostic standard is histopathology, but ashistopathologic analysis is used to determine the presence ofprostate cancer, it cannot be compared with metabolomicsprofiling. Nevertheless in all the studies, good correlationbetween the metabolic findings and the histopathologic find-ings was demonstrated. Furthermore for disease recurrence,McDunn and colleagues reported that the inclusion of a meta-

bolomics profile afforded increased prediction compared withclinical indices alone (16).

Part 2: Hypothesis-Generating StudiesMetabolites and pathways implicated in prostate cancertumorigenesis, progression, and recurrence

In the remaining metabolomics studies, the primary aim wasnot the identification of biomarkers or indices of biomarker utilitywere not reported upon. However, among the results, there wassubstantial crossover with the metabolites and pathways dis-cussed in Part 1, in particular with those thought to be involvedin pathogenesis. In studies comparing prostate cancer patientswith healthy controls, differences in metabolites and pathwaysrelating to energy metabolism were reported including TCA cycleintermediates (24, 27, 32), lactate (24, 32), citrate (23), phos-phoenolypyruvate, and adenosine diphosphate (32).Metabolitesvital to cell growth and proliferation were also identified; theseincluded common amino acids (15, 30) bile acids (60), poly-amines (27), glycerol-3-phosphate (30), and a number of con-stituents of cells membranes including long chain fatty acids(22, 23, 51), phospholipids (30), phosphocholines (61), andcholine (27, 55). Steroid hormones (23) which help regulatethe growth and function of the prostate were also implicated (62),as were inositol and its isomers (27) which are involved inosmoregulation, and have been shown to be dysregulated inseveral other cancers (55) and cortisol (35) which is thoughtto be related to cancer development via the mechanism of chro-nic stress (63). A number of these metabolites, citrate, inositol,lactate (25), and cortisol (35, 63), were additionally observed todiffer between BPH and prostate cancer patients, along withphosphoethanolamine, and glycerophosphoethanolamine (29)which are also components ofmembranes and acetate (25)whichis thought to support cell proliferation through de novo lipidbiosynthesis (64).

Levels of metabolites including sarcosine, uracil, kynurenine,leucine, and proline (43) were shown to be increased duringthe progression to metastatic disease (32, 33, 43), as werepathways involved in nitrogen breakdown (43). Nitrogenmetabolism is known to be altered in tumors to accommodatetheir enhanced glutamine requirements and the increase innucleotide and protein synthesis (65). Arachidonic acid metab-olism was also altered which is in line with the suspectedassociation between dietary fat and prostate cancer (33, 66).Metabolites from the pathways of energy and lipid metabolismwere again demonstrated to be of importance (23, 43) in thedegree of disease aggressiveness. Taken together with the resultsof the biomarkers studies, these findings suggest a particularlyimportant role for pathways involved in abnormal cell growthand intensive cell proliferation in prostate carcinogenesis andprogression. Dysregulation of lipid and fatty acid metabolismmay be particularly crucial to the disease process.

Multiple steroids, markers of lipid beta-oxidation, markers ofomega-oxidation, and markers of insulin resistance (67) wereobserved to decrease following therapy in the study by Saylor andcolleagues (46), while bile acids (60), steroids, and their meta-bolites increased. This was in line with the findings of Huang andcolleagues (51), who reported that the metabolite profiles ofpatients successfully treated with endocrine therapy closelyresembled those of healthy controls. Among all the includedstudies, one did not report any significant findings (37). This

Kelly et al.

Cancer Epidemiol Biomarkers Prev; 25(6) June 2016 Cancer Epidemiology, Biomarkers & Prevention900

on June 16, 2020. © 2016 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

Published OnlineFirst April 6, 2016; DOI: 10.1158/1055-9965.EPI-15-1223

study, by Gamagedara and colleagues was targeted to only fourmetabolites that had previously been reported as significant intissue (37). Discussed further in the following section, it isperhaps not surprising that they were unable to replicate thefindings in a different biologic media. Furthermore, they wereunable to reliably measure the strongest candidate, sarcosine.Nevertheless, it must be taken into consideration that the verysmall number of null findings may reflect more on the biastowards publishing studies with positive findings, than on theapplication of metabolomics profiling.

Part 3: Replication and ValidationTo validate their findings, 11 studies used internal cross-vali-

dation methods (16, 26, 40–44, 47, 49, 50, 53), while onecompared the tissue metabolomics findings to those obtainedby histopathology (27). Two further studies reported that vali-dation in independent cohorts was ongoing at the time of pub-lication (39, 51) but these data are not yet publically available.Two studies (22, 23) were based within the same parent cohort,but there was no overlap between the populations included. Thesame metabolite classes were identified as potentially predictivebiomarkers in the both studies, and a meta-analyses of thefindings provided robust results, particularly for aggressivedisease.

Only two studies attempted replication of their findings in anentirely independent cohort (38, 52). Zhang and colleagues (38)compared an additional 30 prostate cancer patients from adifferent geographic region to their original control population,reporting that 14 of 33 (42%) putative diagnostic biomarkersretained statistical significance. These included four novel meta-bolites; ureido isobutyric acid, indolylacryloyglycine, acetylani-lalinine, and 2-oxoglutarate. After isolating sarcosine as a differ-ential metabolite between benign and prostate cancer tissuesamples, Sreekumar and colleagues replicated the experiment in89 independent samples and reported that not only were sarco-sine levels in tissue significantly increased in the cancer specimensthere was a further increase among patients with metastaticdisease, thereby both confirming and extending upon their orig-inal findings (52).

Sreekumar and colleagues (52) further explored the potentialof urinary sarcosine, observing significantly increased levels inbiopsy positive patients. Two other studies (43, 48) also usedmultiple biologic media. Stabler and colleagues (48) onlyreplicated one of their markers of recurrence, cysteine, betweenurine and serum blood samples. Thysell and colleagues lookedat bone, tissue, and plasma, and again only replicated a smallnumber of metabolites between media, and none were com-mon to all three. Finally, Brown (31) and Shuster and collea-gues. (30) analyzed the same eight prostate samples usingdifferent statistical techniques. They reported similar resultsand both concluded that a metabolic approach could reliablydistinguish between benign and cancerous samples and indi-cate tumor aggressiveness. In the remaining studies, no attemptat replication was made.

The sarcosine debateFollowing the publication of Sreekumar and colleagues' find-

ings on sarcosine, an intermediate and byproduct in glycinesynthesis and degradation, a number of studies attempted toreplicate these promising results (52). Li and colleagues' (33)

pathway-based analysis of Sreekumar and colleagues' datasetreported an association between metastatic prostate cancer andmethionine metabolism pathways, which can be involved in theformation of sarcosine. McDunn and colleagues (16) observedsignificantly elevated sarcosine levels in Gleason grade 8 tumorsor higher compared with benign tissue, while Thysell and collea-gues (43) found a significant increase in sarcosine in the bone ofmen withmetastatic disease. Although recurrence was not a focusof Sreekumar and colleagues' original study (52), Stabler andcolleagues (48) observed that urinary Sarcosine levels at the timeof surgery were significantly higher among those men whosecancer subsequently recurred.

Conversely, Wu (36) and Zhang and colleagues (38), reportedno significant differences in urinary sarcosine levels betweenprostate cancer cases, BPH cases and healthy controls, and sarco-sine was not significant in Mondul and colleagues' (2014) blood-based study (22). Contrary to their findings in bone, Thysell andcolleagues found no association between prostate cancer andsarcosine levels in blood or tissue (43). Sarcosine could not bereliably measured in two further studies (30, 37); however,Shuster and colleagues (30) replicated Sreekumar and colleagues'(52) positive findings for uracil, kynurenine, glycerol-3-phos-phate, leucine, and proline levels in tissue (30), while Gamega-dara and colleagues (37) observed no diagnostic potential forthese metabolites in urine.

It has been suggested that these differences in findings may bethe result of technical differences between the studies. It is itnotable that is it analytically challenging to precisely and accu-rately measure sarcosine, particularly at low concentrations (68),as is evidenced by the two studies that were unable to do so (30,37). It has been suggested that liquid chromatography for sampleseparation prior to MS, may be preferable to gas chromatographyfor the measurement of sarcosine and that this may explain theconflicting findings (68). However, within the studies reportedhere, a combination of liquid- and gas-based chromatographywas used by both the studies reporting positive findings (16, 43,48, 52) and those reporting null findings (22, 36, 43, 69). Thestudies were also coherent with regards to the statistical analysesperformed (comparisons of means, OPLS-DA, and ROC curves).With the exception of McDunn and colleagues, all compared toprostate cancer cases to controls and there was no striking differ-ences in population sizes between the positive and null studies(16). It has also been suggested that population differences insarcosine levels may lead to false-negative or false-positive find-ings. Among the studies reporting positive findings, three werebased in the United States (16, 48, 52) and one in Sweden (43),while among the null studies, twowere based in Asia (36, 38) andtwo in Europe (22, 43). However, the explanation for the differ-ences in findings most likely relates to the multitude of technicalchallenges and that too, in particular, in the urine-based studies,the difficulty in accurately determining the sarcosine/creatineratio is likely to be playing a role (36, 68).

Methodological ConsiderationsComparisons with other cancers

One of the most common applications of metabolomics is inthe study of cancer (54), and a number of metabolomics bio-markers with discriminatory abilities comparable with or evenbetter than those for prostate cancer have been proposed, partic-ularly in cancers of the gastrointestinal system (70–72) and

Metabolomic Biomarkers of Prostate Cancer

www.aacrjournals.org Cancer Epidemiol Biomarkers Prev; 25(6) June 2016 901

on June 16, 2020. © 2016 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

Published OnlineFirst April 6, 2016; DOI: 10.1158/1055-9965.EPI-15-1223

pancreas (73, 74).Malignant cells are known to possessmetabolicphenotypes that differ from many normal tissues, characterizedby a shift toward aerobic glycolysis and pathway alterations thatsupport biomass accumulation for cell proliferation (14, 15, 75,76). As such, these reported findings, particularly in the non-tissue–based studies, may identify signatures that reflect malig-nancy in general and which are not specific to prostate cancer.With this possibility in mind, seven studies investigated addi-tional malignancies (24, 31–33, 37, 43, 50).

Halliday and colleagues (24) compared the metabolic profilesof prostate cancer, lung cancer, and colon adenocarcinoma, andreported tumor-specific differences. Similarly, Kami and collea-gues (32) detected a clear distinction between lung and prostatecancer samples based on their metabolic profiles, and reportedthat lung versus prostate differences were greater than normalversus tumor differences within the same organ. However, theydid note that compared with their normal counterparts bothtumor types shared a number of features such as higher levels ofamino acids, lactate, succinate, fumarate, and malate. Intriguing-ly, these TCA intermediates have also been shown to be increasedin both colon and stomach cancer (77).

Conversely, Gamegedara and colleagues (37), who investi-gated the biomarkers identified by Sreekumar and colleagues(52) in urine, reported that in their population these bio-markers could not distinguish between prostate cancer casesand controls, nor could they distinguish prostate from breastcancer. While Thysell and colleagues (43), who consideredbreast, esophageal, lung, and kidney cancer, observed that thesignificant differences in sarcosine levels between normal boneand the bone metastases of prostate cancer patients were alsoevident in these other cancers, indicating such differences maynot be prostate cancer specific.

Three other studies investigated additional cancers, includingkidney (31), colorectal adenoma (33), lung, colorectal, breast,and gastric cancer (50), although they did not specifically com-pare the metabolome of these malignancies to prostate cancer. Acommon theme between all cancers was a significant alteration inamino acid levels. A large number of differentialmetabolites wereidentified for all the investigated cancers, including prostate,particularly among the amino acids. Nevertheless, Miyagi andcolleagues (50) reported that the greatest proportion of differen-tial metabolites were tumor site–specific.

In summary, in these studies, a prostate cancer–specific meta-bolome was demonstrated both in tissue (24), and in plasma(50), although these metabolomes are characterized more on thebasis of the patterns and behaviors of grouped metabolitesincluding lipids and TCA cycle intermediates, rather than onindividual metabolites. A prostate cancer–specific metabolomewas not demonstrated in urine (37).Within the wider cancermetabolomic literature, a number of the putative "prostate bio-markers" discussed in this review have also been proposed asbiomarkers of other malignancies. Aspartic acid, 2-hydroxybuty-rate, and kynurenine have been suggested as metabolomics bio-markers of colorectal cancer; the malignancy in which the field ofmetabolomics is perhapsmost advanced (78). Lactate, threonine,acetate, uracil, succinate, lysine and tyrosine,myoinositol, taurineand creatine have been shown to be associated with the presenceof rectal cancer, and correlated with its progression (79). Taurineis also increased in squamous cell carcinoma (80), while lactatehas additionally been shown to be associated with esophagogas-tric cancer (81), along with fumarate, valine, glutamine, gluta-

mate (81), xylonic acid (81, 82), tyrosine, phenylalanine, andtryptophan (83). Together, these metabolites indicate a generaldysregulation in the metabolism of cellular respiration, energy,amino acids, ketone body, and choline metabolism which, asdiscussed, could be applicable to all cancers. Similarly, choline,phosphocholine, phosphatidylcholine, lysophosphocholine,and glycerophosphocholine, which are necessary for cell mem-brane synthesis and intercellular signaling (55), have been iden-tified inmetabolomic profiling studies ofmultiple cancers includ-ing brain, breast, lung, and liver (84). One of the prostatediagnostic biomarkers with the highest reported AUCs, acylcar-nitine in blood, has also been shown to distinguish kidney cancerpatients from controls, and by their degree of severity whenmeasured in urine (85); this is hypothesized to reflect alterationsin immune surveillance and again may point to the overallphenotype required to support the growth and proliferation ofmalignant cells. Even citrate, which couldhave beenhypothesizedto be prostate cancer–specific given its importance in the prostatehas been shown to be increased in bile samples of patients withbiliary tract cancers (86).

In fact, the vast majority of the metabolites reported here havealso been identified in other studies supporting the concept of acarcinogenesismetabolome.However, tumors still retainmuchoftheir unique organ-specific metabolism (32, 75), and, to date,neither spermine nor sarcosine have been proposed as metabo-lomic biomarkers of any other malignancies. However whetherthis denotes these metabolites as prostate specific, or it merelyreflects the nature of metabolomics and the possibility thesemetabolites have not been measured in the profiling of othercancers, remains to be seen.

Technological and analytical issuesIn addition to the "epidemiologic validation" of findings

through replication, is it also vital to show that a proposedbiomarker displays `technical validation' in terms of intrinsicmeasurement error and analytic sensitivity (87). Much of thebetween-study heterogeneity and inconsistency can be attributedto the use of differing technological, experimental, and analyticalmethods, which can affect themeasurements of metabolites in anas yet to be determined way (88). Only seven studies reported ontheir quality control (QC) procedures, to allow the considerationof system stability, with varying degrees of detail (22, 23, 35,38, 42, 46, 51), while only eight (22, 23, 30, 35, 38, 42, 46, 51)reported on the relative SD (RSD%) threshold for the includedfeatures (38).

Bias and confoundingTo fully understand the impact of disease, the composition of

the "normal" metabolome in healthy individuals must also beestablished. The metabolome will be dependent on the originat-ing tissue or biologic media, but will additionally vary by age,BMI, diet, and other lifestyle factors (89). This is further compli-cated by the fact that a genetic component to themetabolome hasbeen demonstrated (90). There is also a wide range of stabilityamong metabolites and those metabolites that are more stable,including amino acids, are more likely to show differences if theyexist between cases and controls. The temporal fluctuation of themetabolome also introduces novel challenges particularly inthose studies assessing recurrence and treatment response, andit is of note that none of the diagnostic or aggressiveness studiesutilized repeat measurements to address this (87). Similarly none

Kelly et al.

Cancer Epidemiol Biomarkers Prev; 25(6) June 2016 Cancer Epidemiology, Biomarkers & Prevention902

on June 16, 2020. © 2016 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

Published OnlineFirst April 6, 2016; DOI: 10.1158/1055-9965.EPI-15-1223

of the studies addressed the temporal fluctuation in the metabo-lome through the use of repeat samples.

Consequently, as well as technologically induced variation,confounding represents an important issue in any metabolomicsstudy (91). In epidemiologic studies of prostate cancer, a numberof potential confounders are commonly included in statisticalmodels including age, race, BMI, and often Gleason grade, due totheir reported associations with prostate cancer independence,progression, and lethality (92, 93). Importantly, these variablesmay also affect the metabolome (94, 95). However they wererarely taken into account in the selected studies.

The impact of interindividual variation represents a strongargument for cases acting as their own controls (96), and in 10of the studies intratumor differences within the same patientswere compared (16, 24, 27, 28, 30–33, 44, 47). Similarly, Saylorand colleagues compared blood samples pre- and post- therapy(46). The remainder utilized external controls. Although the useof external controls ensures that the comparison group is"healthy" and not subject to underlying or latent disease pathol-ogy, it also introduces the potential for false positives to arisethrough batch effects or confounding. Among the studies in thisreview, the majority of blood- and urine-based studies utilizedexternal controls, so it is difficult to assess whether this resulted inan excess of positive findings. Eight studies matched on age (22,23, 41, 42, 45, 50, 51, 97), 4 on sampling time (22, 23, 43, 45),one on Gleason grade (45) and one on BMI (42). However,neither age (39, 40), smoking status (22, 50), clinical variables(48), BMI, serum cholesterol, educational level, nor variousdietary factors (22, 23) were found to act as confounders in thosestudies where they were considered. Confounding was notaddressed in the remainder of the studies, and none consideredother potentially important factors such as subtype heterogeneity(98). Treatment was shown to be a further modifier of themetabolome in the studies investigating its effect (46, 51), andalthough six studies stated that the biologic samples were col-lectedprior to any radiationor hormonal therapy (32, 36, 47–50),one study included metabolic profiles ascertained after therapy(43), and the remaining case–control studies did not report onthis covariate.

The majority of the studies were conducted in predominantlyCaucasian populations with the exception of six Asia-basedstudies (32, 35, 36, 38, 50, 51), and only two studies (40, 41)reported on the ratio of races within their study. This is ofparticular importance given the largely unexplained disparitiesin prostate cancer incidence between ethnicities (99); however, inZhou and colleagues' (40) study, race was not found to act as aconfounder. Of the remaining studies, 19 were based in NorthAmerica and eight were based in Europe, therefore given thesuspected impact of environmental and dietary factors on themetabolome, the wider generalizability of the reported findingsmust be considered.

Multiple testingDespite the high-dimensional nature of metabolomics, only

five of the selected studies controlled for multiple testing (16, 22,23, 47, 52). Although the significant findings reported in thesethree studies were robust to such correction, many applied anominal significance level of 0.05 regardless of the number ofmetabolites under investigation. There was additional possibilityof false positives arising in those studies comparing multiplebiologic media. Only four studies (16, 40, 50, 53) reported on

the statistical power of their approach in the search for discrim-inatory metabolites.

Biologic samplesThe media in which a biomarker can be reliably measured is

of importance in the consideration of its forward translation.The tissue-based biomarkers would have no utility in predictingincident disease, and even for diagnostic or prognostic pur-poses, blood and urine can be obtained in a less invasive andmore cost-effective fashion. Furthermore, tissue is a limitedresource, and it may be prudent to preserve it for other usesincluding subsequent histopathologic analysis. Disease-associ-ated metabolites tend to be more concentrated the closer inproximity they are to the organ of interest (30); however, onlyone study considered prostatic secretions (39), possibly due tochallenges related to its collection. Interestingly, the use oftissue did not appear to confer any advantages over the otherbiologic specimens, with some of the strongest results reportedin blood samples.

DiscussionBecause widely used PSA testing remains somewhat controver-

sial (36, 100, 101), additional biomarkers that could help refinepractices would be a welcome addition to the management ofprostate cancer. With the advancement of high-throughput tech-nologies, metabolomics is emerging as a promising tool inbiomarker development (9, 12, 69, 91, 102–106). Its downstreamnature provides a holistic picture of the malignant state andconsequently insight into dysregulated metabolic pathways andinherent disease development. The selected literature providesencouraging results in the field of prostate cancer; however, it alsodemonstrates the novel challenges faced bymetabolomics, whichare only just beginning to be addressed.

Themetabolomics of prostate cancer remains a small field withthemajority of studies focusedon the identificationof biomarkersto distinguish malignant and benign prostate tissue, with fewstudies investigating disease progression and treatment response.Only two studies to date have employed a prospective design tolook for predictive biomarkers (22, 23). Therefore, importantissues of cause and effect must be considered when evaluating theutility of the diagnostic biomarkers and the role of BPH as adisease continuum-intermediatemay be particularly important inthis respect.

Themajority of the included studies reported distinct clusteringby metabolome profiles, with differentiation status playing animportant role in the determination of the profiles (32). In thosestudieswhichdevelopedbiomarkerswith thepotential for clinicaltranslation, the reported AUCs associated with the biomarkerswere high. In many cases these AUCs were higher than for PSA,suggesting these new biomarkers would outperform the currentgold standard in many cases. This is in line with the generallyaccepted consensus that the metabolome represents a rich sourcefor biomarker identification. However, replication, particularlybetween biologicmedia, and independent validationwas lacking,multiple testing was rarely accounted for, and the extent to whichthe reported findings may represent false positives is difficultto assess.

This was true even of one of the most commonly cited "meta-bolomics successes" sarcosine, and the importance of technicalissues in metabolomics studies is perhaps best exemplified by the

Metabolomic Biomarkers of Prostate Cancer

www.aacrjournals.org Cancer Epidemiol Biomarkers Prev; 25(6) June 2016 903

on June 16, 2020. © 2016 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

Published OnlineFirst April 6, 2016; DOI: 10.1158/1055-9965.EPI-15-1223

debate over the potential use of this biomarker. Nevertheless anumber of metabolites and pathways were repeatedly implicatedby the studies with amino acid and lipidmetabolism appearing toplay a predominant role in carcinogenesis, progression, andrecurrence. Caution must be taken to ensure such findings donot merely reflect the most abundant, easy to measure or stablemetabolites, although encouragingly, a number of the studiesindicate that the "metabolome" of prostate cancer is distinct fromthat of other malignancies.

These challenges inherent to metabolomics extend evenbeyond those facing researchers when they first tried to charac-terize the human genome, due to the additional temporal com-ponent as well as issues regarding the stability, variation, andplasticity of the metabolites (107). Therefore, collaborationbetween groups conducting metabolomics-based studies is vital,both in terms of standardizing the optimal methods and analyt-ical strategies, to maximize reproducibility, reliability, and sen-sitivity and also for replication or the accrual of sufficient samplesizes for these highly dimensional studies (108).

Clinical translation remains the end goal, but a number ofimportant factors remain to be considered before this isfeasible for the current studies, including issues of bias, con-founding, and generalizability. Beyond the efficacy of a bio-marker, the feasibility of clinical translocation must also beconsidered. More than two decades since PSA testing wasintroduced no such biomarkers have been clinically approved(109). In fact it may be that the utility of the metabolites andmetabolite profiles identified here lies not in their clinicalusage but in the insights they provide into the mechanisms of

carcinogenesis. For example McDunn and colleagues (16)postulated that there may be a variety of pathways that leadto the development and progression of prostate cancer, andtherefore multiple metabolite models that are able to predictthe outcome of interest in a certain subpopulations. Thediffering results and lack of replication in the included studiesmay support this theory.

In conclusion, the study of the metabolome of prostate cancerremains in the early phases, but could yet represent an importanttool both in the understanding of prostate cancer developmentand progression, and in the development of biomarkers to aid itsmanagement.

Disclosure of Potential Conflicts of InterestNo potential conflicts of interest were disclosed.

Grant SupportThis work is supported by the DF/HCC Specialized Programs in Research

Excellence (SPORE) in Prostate Cancer funded by the NCI/NIH (grant number:2P50CA090381-11A1; Principal Investigator: Philip Kantoff), P01 CA055075,CA133891, CA141298, CA136578, and UM1 CA167552 (to M.G. VanderHeiden, E. Giovannucci, and L.A Mucci). L.A. Mucci is a Young Investigatorof the Prostate Cancer Foundation.

The costs of publication of this article were defrayed in part by thepayment of page charges. This article must therefore be hereby markedadvertisement in accordance with 18 U.S.C. Section 1734 solely to indicatethis fact.

ReceivedDecember 2, 2015; revisedMarch 8, 2016; acceptedMarch 23, 2016;published OnlineFirst April 6, 2016.

References1. Siegel R,NaishadhamD, Jemal A. Cancer statistics, 2012. CACancer J Clin

2012;62:10–29.2. Crawford ED, Ventii K, Shore ND. New biomarkers in prostate cancer.

Oncology 2014;28:135–42.3. Nicholson A, Mahon J, Boland A, Beale S, Dwan K, Fleeman N, et al. The

clinical effectiveness and cost-effectiveness of the PROGENSA� prostatecancer antigen 3 assay and the prostate health index in the diagnosis ofprostate cancer: a systematic review and economic evaluation. HealthTechnol Assess 2015;19:1–13.

4. Sooriakumaran P, Nyberg T, Akre O, Haendler L, Heus I, Olsson M, et al.Comparative effectiveness of radical prostatectomy and radiotherapy inprostate cancer: observational study of mortality outcomes. BMJ 2014;348.

5. Xia S, Kohli M, Du M, Dittmar RL, Lee A, Nandy D, et al. Plasmagenetic and genomic abnormalities predict treatment response andclinical outcome in advanced prostate cancer. Oncotarget 2015;6:16411–21.

6. Kohli M, Tindall DJ. New developments in the medical management ofprostate cancer. Mayo Clin Proc 2010;85:77–86.

7. Olmos D, Brewer D, Clark J, Danila DC, Parker C, Attard G, et al.Prognostic value of blood mRNA expression signatures in castration-resistant prostate cancer: a prospective, two-stage study. Lancet Oncol2012;13:1114–24.

8. Danila DC, Fleisher M, Scher HI. Circulating tumor cells as biomarkers inprostate cancer. Clin Cancer Res 2011;17:3903–12.

9. Spratlin JL, Serkova NJ, Eckhardt SG. Clinical applications of metabolo-mics in oncology: a review. Clin Cancer Res 2009;15:431–40.

10. Goodacre R, Vaidyanathan S, Dunn WB, Harrigan GG, Kell DB. Metabo-lomics by numbers: acquiring and understanding global metabolite data.Trends Biotechnol 2004;22:245–52.

11. Holmes E, Loo RL, Stamler J, Bictash M, Yap IKS, Chan Q, et al. Humanmetabolic phenotype diversity and its association with diet and bloodpressure. Nature 2008;453:396–400.

12. Denkert C, Bucher E, Hilvo M, Salek R, Oresic M, Griffin J, et al. Meta-bolomics of human breast cancer: new approaches for tumor typing andbiomarker discovery. Genome Med 2012;4:37.

13. Cantor JR, Sabatini DM. Cancer cell metabolism: one hallmark, manyfaces. Cancer Discov 2012;2:881–98.

14. Warburg O, Wind F, Negelein E. The metabolism of tumors in the body. JGen Physiol 1927;8:519–30.

15. Vander Heiden MG, Cantley LC, Thompson CB. Understanding thewarburg effect: the metabolic requirements of cell proliferation. Science2009;324:1029–33.

16. McDunn JE, Li Z, Adam K-P, Neri BP, Wolfert RL, Milburn MV, et al.Metabolomic signatures of aggressive prostate cancer. Prostate 2013;73:1547–60.

17. McGrowder DA, Jackson LA, Crawford TV. Prostate cancer andmetabolic syndrome: is there a link? Asian Pac J Cancer Prev 2012;13:1–13.

18. Hsing AW, Sakoda LC, Chua SC. Obesity, metabolic syndrome, andprostate cancer. Am J Clin Nutr 2007;86:843S–57S.

19. Trock BJ. Application of metabolomics to prostate cancer. Urol Oncol2011;29:572–81.

20. Costello LC, Franklin RB. Concepts of citrate production andsecretion by prostate. 1. Metabolic relationships. Prostate 1991;18:25–46.

21. Fiehn O.Metabolomics–the link between genotypes and phenotypes.Plant Mol Biol 2002;48:155–71.

22. Mondul A, Moore S, Weinstein S, M€annist€o S, Sampson J, Albanes D.1-Stearoylglycerol is associated with risk of prostate cancer: resultsfrom a serum metabolomic profiling analysis. Metabolomics 2014;10:1036–41.

23. Mondul A, Moore SC, Weinstein SJ, Karoly ED, Sampson JN, Albanes D.Metabolomic analysis of prostate cancer risk in a prospective cohort:the alpha-tocolpherol, beta-carotene cancer prevention (ATBC) study.Int J Cancer 2015;137:2124–32.

Kelly et al.

Cancer Epidemiol Biomarkers Prev; 25(6) June 2016 Cancer Epidemiology, Biomarkers & Prevention904

on June 16, 2020. © 2016 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

Published OnlineFirst April 6, 2016; DOI: 10.1158/1055-9965.EPI-15-1223

24. Halliday KR, Fenoglio-Preiser C, Sillerud LO. Differentiation of humantumors from nonmalignant tissue by natural-abundance 13C NMRspectroscopy. Magn Reson Med 1988;7:384–411.

25. Schiebler ML, Miyamoto KK, White M, Maygarden SJ, Mohler JL. In vitrohigh resolution 1H-spectroscopy of the human prostate: benign prostatichyperplasia, normal peripheral zone and adenocarcinoma. Magn ResonMed 1993;29:285–91.

26. Hahn P, Smith ICP, Leboldus L, Littman C, Somorjai RL, Bezabeh T. Theclassification of benign and malignant human prostate tissue by multi-variate analysis of 1H magnetic resonance spectra. Cancer Res 1997;57:3398–401.

27. Swanson MG, Vigneron DB, Tabatabai ZL, Males RG, Schmitt L, CarrollPR, et al. Proton HR-MAS spectroscopy and quantitative pathologicanalysis of MRI/3D-MRSI-targeted postsurgical prostate tissues. MagnReson Med 2003;50:944–54.

28. Cheng LL, Burns MA, Taylor JL, He W, Halpern EF, McDougal WS, et al.Metabolic characterization of humanprostate cancer with tissuemagneticresonance spectroscopy. Cancer Res 2005;65:3030–4.

29. Komoroski RA, Holder JC, Pappas AA, Finkbeiner AE. 31P NMR ofphospholipid metabolites in prostate cancer and benign prostatic hyper-plasia. Magn Reson Med 2011;65:911–3.

30. Shuster JR, Lance RS, Troyer DA.Molecular preservation by extraction andfixation, mPREF: a method for small molecule biomarker analysis andhistology on exactly the same tissue. BMC Clin Pathol 2011;11:14-.

31. BrownM,McDunn J, Gunst P, Smith E,MilburnM, Troyer D, et al. Cancerdetection and biopsy classification using concurrent histopathologicaland metabolomic analysis of core biopsies. Genome Med 2012;4:33.

32. Kami K, Fujimori T, Sato H, Sato M, Yamamoto H, Ohashi Y, et al.Metabolomic profiling of lung and prostate tumor tissues by capillaryelectrophoresis time-of-flight mass spectrometry. Metabolomics 2013;9:444–53.

33. Li C, Han J, Yao Q, Zou C, Xu Y, Zhang C, et al. Subpathway-GM:identification of metabolic subpathways via joint power of interestinggenes and metabolites and their topologies within pathways. NucleicAcids Res 2013;41:e101–e.

34. LokhovP,DashtievM,Moshkovskii S, ArchakovA.Metabolite profilingofblood plasma of patients with prostate cancer. Metabolomics 2010;6:156–63.

35. Cho H-J, Kim JD, Lee W-Y, Chung BC, Choi MH. Quantitative metabolicprofiling of 21 endogenous corticosteroids in urine by liquid chroma-tography–triple quadrupole-mass spectrometry. Analytica Chimica Acta2009;632:101–8.

36. Wu H, Liu T, Ma C, Xue R, Deng C, Zeng H, et al. GC/MS-basedmetabolomic approach to validate the role of urinary sarcosine and targetbiomarkers for human prostate cancer by microwave-assisted derivatiza-tion. Anal Bioanal Chem 2011;401:635–46.

37. Gamagedara S, Kaczmarek AT, Jiang Y, Cheng X, Rupasinghe M, Ma Y.Validation study of urinary metabolites as potential biomarkers forprostate cancer detection. Bioanalysis 2012;4:1175–83.

38. Zhang T, Watson DG, Wang L, Abbas M, Murdoch L, Bashford L, et al.Application of holistic liquid chromatography-high resolution massspectrometry based urinary metabolomics for prostate cancer detectionand biomarker discovery. PLoS One 2013;8:e65880.

39. Serkova NJ, Gamito EJ, Jones RH, O'Donnell C, Brown JL, Green S, et al.The metabolites citrate, myo-inositol, and spermine are potential age-independent markers of prostate cancer in human expressed prostaticsecretions. Prostate 2008;68:620–8.

40. Zhou X, Mao J, Ai J, Deng Y, Roth MR, Pound C, et al. Identification ofplasma lipid biomarkers for prostate cancer by lipidomics and bioinfor-matics. PLoS One 2012;7:e48889.

41. Zang X, Jones CM, Long TQ, Monge ME, Zhou M, Walker LD, et al.Feasibility of detecting prostate cancer by ultraperformance liquid chro-matography–mass spectrometry serum metabolomics. J Proteome Res2014;13:3444–54.

42. Struck-Lewicka W, Kordalewska M, Bujak R, Yumba Mpanga A, Markus-zewskiM, Jacyna J, et al. Urinemetabolic fingerprinting using LC–MS andGC–MS reveals metabolite changes in prostate cancer: a pilot study.J Pharm Biomed Anal 2015;111:351–61.

43. Thysell E, Surowiec I, H€ornberg E, Crnalic S, Widmark A, Johansson AI,et al. Metabolomic characterization of human prostate cancer bone

metastases reveals increased levels of cholesterol. PLoS One 2010;5:e14175.

44. Menard C, Smith ICP, Somorjai RL, Leboldus L, Patel R, Littman C, et al.Magnetic resonance spectroscopy of the malignant prostate gland afterradiotherapy: a histopathologic study of diagnostic validity. Int J RadiatOncol Biol Phys 2001;50:317–23.

45. Maxeiner A, Adkins CB, Zhang Y, Taupitz M, Halpern EF, McDougal WS,et al. Retrospective analysis of prostate cancer recurrence potential withtissue metabolomic profiles. Prostate 2010;70:710–7.

46. Saylor PJ, Karoly ED, Smith MR. Prospective study of changes in themetabolomic profiles of men during their first three months ofandrogen deprivation therapy for prostate cancer. Clin Cancer Res2012;18:3677–85.

47. GiskeodegardGF, BertilssonH, Selnaes KM,Wright AJ, Bathen TF, Viset T,et al. Spermine and citrate as metabolic biomarkers for assessing prostatecancer aggressiveness. PLoS One 2013;8:e62375.

48. Stabler S, Koyama T, Zhao Z, Martinez-Ferrer M, Allen RH, Luka Z, et al.Serum Methionine Metabolites Are Risk Factors for Metastatic ProstateCancer Progression. PLoS One 2011;6:e22486.

49. Osl M, Dreiseitl S, Pfeifer B, Weinberger K, Klocker H, Bartsch G, et al.A new rule-based algorithm for identifying metabolic markersin prostate cancer using tandem mass spectrometry. Bioinformatics2008;24:2908–14.

50. Miyagi Y, Higashiyama M, Gochi A, Akaike M, Ishikawa T, Miura T, et al.Plasma free amino acid profiling of five types of cancer patients and itsapplication for early detection. PLoS One 2011;6:e24143.

51. Huang G, Liu X, Jiao L, Xu C, Zhang Z, Wang L, et al. Metabolomicevaluation of the response to endocrine therapy in patients with prostatecancer. Eur J Pharmacol 2014;729:132–7.

52. Sreekumar A, Poisson LM, Rajendiran TM, Khan AP, Cao Q, Yu J, et al.Metabolomic profiles delineate potential role for sarcosine in prostatecancer progression. Nature 2009;457:910–4.

53. Fan Y, Murphy TB, Byrne JC, Brennan L, Fitzpatrick JM, Watson RWG.Applying random forests to identify biomarker panels in serum 2D-DIGEdata for the detection and staging of prostate cancer. J Proteome Res2011;10:1361–73.

54. LiesenfeldDB,HabermannN,OwenRW,Scalbert A,UlrichCM.Reviewofmass spectrometry–based metabolomics in cancer research. Cancer Epi-demiol Biomarkers Prev 2013;22:2182–201.

55. Griffin JL, Shockcor JP. Metabolic profiles of cancer cells. Nat Rev Cancer2004;4:551–61.

56. Chadeau-HyamM, Athersuch TJ, KeunHC, De Iorio M, Ebbels TM, JenabM, et al. Meeting-in-the-middle using metabolic profiling - a strategy forthe identification of intermediate biomarkers in cohort studies. Biomar-kers 2011;16:83–8.

57. Nicholson G, RantalainenM,Maher AD, Li JV, Malmodin D, Ahmadi KR,et al. Human metabolic profiles are stably controlled by genetic andenvironmental variation. Mol Syst Biol 2011;7:525.

58. Xia J, Broadhurst D, Wilson M, Wishart D. Translational biomarkerdiscovery in clinical metabolomics: an introductory tutorial. Metabolo-mics 2013;9:280–99.

59. Thompson IM, Ankerst DP. Prostate-specific antigen in the early detectionof prostate cancer. Canadian Med Assoc J 2007;176:1853–8.

60. Li T, Apte U. Bile acid metabolism and signaling in cholestasis, inflam-mation, and cancer. In: James PH, editor. Advances in pharmacology :Academic Press; 2015. p. 263–302.

61. Currie E, Schulze A, Zechner R, Walther Tobias C, Farese Jr Robert V.Cellular fatty acid metabolism and cancer. Cell Metab 2013;18:153–61.

62. Soronen P, Laiti M, T€orn S, H€ark€onen P, Patrikainen L, Li Y, et al. Sexsteroid hormonemetabolism and prostate cancer. J Steroid BiochemMolBiol 2004;92:281–6.

63. Moreno-Smith M, Lutgendorf SK, Sood AK. Impact of stress on cancermetastasis. Fut Oncol 2010;6:1863–81.

64. Hosios AM, Vander Heiden MG. Acetate metabolism in cancer cells.Cancer Metab 2014;2:27.

65. Medina M�A. Glutamine and cancer. TJ Nutr 2001;131:S2539–S42.66. Yang P, Cartwright CA, Li JIN, Wen S, Prokhorova IN, Shureiqi I, et al.

Arachidonic acid metabolism in human prostate cancer. Int J Oncol2012;41:1495–503.

Metabolomic Biomarkers of Prostate Cancer

www.aacrjournals.org Cancer Epidemiol Biomarkers Prev; 25(6) June 2016 905

on June 16, 2020. © 2016 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

Published OnlineFirst April 6, 2016; DOI: 10.1158/1055-9965.EPI-15-1223

67. Djiogue S, Nwabo Kamdje AH, Vecchio L, Kipanyula MJ, Farahna M,Aldebasi Y, et al. Insulin resistance and cancer: the role of insulin and IGFs.Endocr Rel Cancer 2013;20:R1–R17.

68. Cernei N, Heger Z, Gumulec J, Zitka O, Masarik M, Babula P, et al.Sarcosine as a potential prostate cancer biomarker—a review. Int JMol Sci2013;14:13893–908.

69. Zhang A, Sun H, Yan G, Wang P, Han Y, Wang X. Metabolomics indiagnosis and biomarker discovery of colorectal cancer. Cancer Lett2014;345:17–20.

70. Ni Y, Xie G, Jia W. Metabonomics of human colorectal cancer: newapproaches for early diagnosis and biomarker discovery. J Proteome Res2014;13:3857–70.

71. Jayavelu ND, Bar NS. Metabolomic studies of human gastric cancer:review. World J Gastroenterol 2014;20:8092–101.

72. Antonowicz S, Kumar S, Wiggins T, Markar SR, Hanna GB. Diagnosticmetabolomic blood tests for endoluminal gastrointestinal cancer—asystematic review and assessment of quality. Cancer Epidemiol Biomar-kers Prev 2016;25:6–15.

73. Xie G, Lu L, Qiu Y, Ni Q, Zhang W, Gao Y-T, et al. Plasma metabolitebiomarkers for the detection of pancreatic cancer. J Proteome Res2015;14:1195–202.

74. Mayers JR, Wu C, Clish CB, Kraft P, Torrence ME, Fiske BP, et al.Elevated circulating branched chain amino acids are an early eventin pancreatic adenocarcinoma development. Nat Med 2014;20:1193–8.

75. Hu J, Locasale JW, Bielas JH, O'Sullivan J, Sheahan K, Cantley LC, et al.Heterogeneity of tumor-induced gene expression changes in the humanmetabolic network. Nat Biotechnol 2013;31:522–9.

76. Vander Heiden MG.Targeting cancer metabolism: a therapeutic windowopens. Nat Rev Drug Discov 2011;10:671–84.

77. HirayamaA, Kami K, SugimotoM, SugawaraM, Toki N,OnozukaH, et al.Quantitative metabolome profiling of colon and stomach cancer micro-environment by capillary electrophoresis time-of-flight mass spectrom-etry. Cancer Res 2009;69:4918–25.

78. Nishiumi S, Kobayashi T, Ikeda A, Yoshie T, KibiM, Izumi Y, et al. A novelserum metabolomics-based diagnostic approach for colorectal cancer.PLoS ONE 2012;7:e40459.

79. Wang H, Wang L, Zhang H, Deng P, Chen J, Zhou B, et al. 1H NMR-basedmetabolic profiling of human rectal cancer tissue. Mol Cancer 2013;12:121-.

80. RochaCM, Barros AS,GoodfellowBJ, Carreira IM,GomesA, Sousa V, et al.NMR metabolomics of human lung tumours reveals distinct metabolicsignatures for adenocarcinoma and squamous cell carcinoma. Carcino-genesis 2015;36:68–75.

81. Abbassi-Ghadi N, Kumar S, Huang J, Goldin R, Takats Z, Hanna GB.Metabolomic profiling of oesophago-gastric cancer: a systematic review.Eur J Cancer 2013;49:3625–37.

82. Song H, Wang L, Liu HL, Wu XB, Wang HS, Liu ZH, et al. Tissuemetabolomic fingerprinting reveals metabolic disorders associated withhuman gastric cancer morbidity. Oncol Rep 2011;26:431–8.

83. Wiggins T, Kumar S, Markar SR, Antonowicz S, Hanna GB. Tyro-sine, phenylalanine, and tryptophan in gastroesophageal malig-nancy: a systematic review. Cancer Epidemiol Biomarkers Prev2015;24:32–8.

84. Beger R. A review of applications of metabolomics in cancer. Metabolites2013;3:552.

85. Ganti S, Taylor SL, Kim K, Hoppel CL, Guo L, Yang J, et al. Urinaryacylcarnitines are altered in human kidney cancer. Int J Cancer 2012;130:2791–800.

86. WenH, Yoo SS, Kang J, KimHG, Park J-S, Jeong S, et al. A newNMR-basedmetabolomics approach for the diagnosis of biliary tract cancer. J Hepatol2010;52:228–33.

87. Vineis P, Perera F. Molecular epidemiology and biomarkers in etiologiccancer research: the new in light of the old. Cancer Epidemiol BiomarkersPrev 2007;16:1954–65.

88. Garcia-Segura JM, Sanchez-Chapado M, Ibarburen C, Viano J, Angulo JC,Gonzalez J, et al. Invivo proton magnetic resonance spectroscopy ofdiseased prostate: spectroscopic features of malignant versus benignpathology. Magn Reson Imaging 1999;17:755–65.

89. Krug S, Kastenmuller G, Stuckler F, Rist MJ, Skurk T, Sailer M, et al. Thedynamic range of the humanmetabolome revealed by challenges. FASEB J2012;26:2607–19.

90. Gieger C, Geistlinger L, Altmaier E, Hrab�e de Angelis M, Kronenberg F,Meitinger T, et al. Genetics meets metabolomics: a genome-wide associ-ation study of metabolite profiles in human serum. PLoS Genet 2008;4:e1000282.

91. Nordstr€om A, Lewensohn R. Metabolomics: moving to the clinic.J Neuroimmune Pharmacol 2010;5:4–17.

92. Ma J, Li H,Giovannucci E,Mucci L,QiuW,Nguyen PL, et al. Prediagnosticbody-mass index, plasma C-peptide concentration, and prostate cancer-specific mortality in men with prostate cancer: a long-term survivalanalysis. Lancet Oncol 2008;9:1039–47.

93. Wilson KM, Giovannucci EL, Mucci LA. Lifestyle and dietary factors in theprevention of lethal prostate cancer. Asian J Androl 2012;14:365–74.

94. Morris C,O'GradaC, RyanM,RocheHM,GibneyMJ,Gibney ER, et al. Therelationship between BMI and metabolomic profiles: a focus on aminoacids. Proc Nutr Soc 2012;71:634–8.

95. Menni C, Kastenm€uller G, Petersen AK, Bell JT, Psatha MTsai P-C, ,et al. et al.Metabolomic markers reveal novel pathways of ageing andearly development in human populations. Int J Epidemiol 2013;42:111–9.

96. Jentzmik F, Stephan C, Miller K, Schrader M, Erbersdobler A, KristiansenG, et al. Sarcosine in urine after digital rectal examination fails as amarkerin prostate cancer detection and identification of aggressive tumours. EurUrol 2010;58:12–8.

97. Touvier M, Fezeu L, Ahluwalia N, Julia C, Charnaux N, Sutton A, et al.Association between prediagnostic biomarkers of inflammation andendothelial function and cancer risk: a nested case-control study. Am JEpidemiol 2013;177:3–13.

98. Wyatt AW, Mo F, Wang Y, Collins CC. The diverse heterogeneity ofmolecular alterations in prostate cancer identified through next-genera-tion sequencing. Asian J Androl 2013;15:301–8.

99. Centers for Disease Control and Prevention. Prostate cancer rates by raceand ethnicity ; 2014[cited 2015 Feb 27. Available from: http://www.cdc.gov/cancer/prostate/statistics/race.htm.

100. Lehrer S, Diamond EJ, Mamkine B, Droller MJ, Stone NN, Stock RG. C-reactive protein is significantly associated with prostate-specific antigenand metastatic disease in prostate cancer. BJU Int 2005;95:961–2.

101. ChouR, Croswell JM,Dana T, Bougatsos C, Blazina I, FuR, et al. Screeningfor prostate cancer: a reviewof the evidence for theU.S. preventive servicestask force. Ann Intern Med 2011;155:762–71.

102. Bictash M, Ebbels TM, Chan Q, Loo RL, Yap IKS, Brown IJ, et al.Opening up the "Black Box": metabolic phenotyping and metabo-lome-wide association studies in epidemiology. J Clin Epidemiol2010;63:970–9.

103. Ritchie S, Ahiahonu P, Jayasinghe D, Heath D, Liu J, Lu Y, et al. Reducedlevels of hydroxylated, polyunsaturated ultra long-chain fatty acids in theserum of colorectal cancer patients: implications for early screening anddetection. BMC Med 2010;8:13.

104. Hilvo M, Denkert C, Lehtinen L, M€uller B, Brockm€oller S, Sepp€anen-Laakso T, et al. Novel theranostic opportunities offered by characteriza-tion of altered membrane lipid metabolism in breast cancer progression.Cancer Res 2011;71:3236–45.

105. YeG, ZhuB, YaoZ, YinP, LuX,KongH, et al. Analysis of urinarymetabolicsignatures of early hepatocellular carcinoma recurrence after surgicalremoval using gas chromatography–mass spectrometry. J Proteome Res2012;11:4361–72.

106. AsiagoVM,Alvarado LZ, ShanaiahN,GowdaGAN,Owusu-SarfoK, BallasRA, et al. Early detection of recurrent breast cancer using metaboliteprofiling. Cancer Res 2010;70:8309–18.

107. Kyrtopoulos SA.Making sense of OMICS data in population-based envi-ronmental health studies. Environ Mol Mutagen 2013;54:468–79.

108. Schalken JA.Is urinary sarcosine useful to identify patientswith significantprostate cancer? The trials and tribulations of biomarker development.Eur Urol 2010;58:19–20.

109. Wu C-L, Jordan KW, Ratai EM, Sheng J, Adkins CB, DeFeo EM, et al.Metabolomic imaging for human prostate cancer detection. Sci TranslatMed 2010;2:16ra8–ra8.

Cancer Epidemiol Biomarkers Prev; 25(6) June 2016 Cancer Epidemiology, Biomarkers & Prevention906

Kelly et al.

on June 16, 2020. © 2016 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

Published OnlineFirst April 6, 2016; DOI: 10.1158/1055-9965.EPI-15-1223

2016;25:887-906. Published OnlineFirst April 6, 2016.Cancer Epidemiol Biomarkers Prev   Rachel S. Kelly, Matthew G. Vander Heiden, Edward Giovannucci, et al.   Diagnosis, Progression, Prognosis, and RecurrenceMetabolomic Biomarkers of Prostate Cancer: Prediction,

  Updated version

  10.1158/1055-9965.EPI-15-1223doi:

Access the most recent version of this article at:

   

   

  Cited articles

  http://cebp.aacrjournals.org/content/25/6/887.full#ref-list-1

This article cites 106 articles, 17 of which you can access for free at:

  Citing articles

  http://cebp.aacrjournals.org/content/25/6/887.full#related-urls

This article has been cited by 4 HighWire-hosted articles. Access the articles at:

   

  E-mail alerts related to this article or journal.Sign up to receive free email-alerts

  Subscriptions

Reprints and

  .pubs@aacr.orgat

To order reprints of this article or to subscribe to the journal, contact the AACR Publications Department

  Permissions

  Rightslink site. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC)

.http://cebp.aacrjournals.org/content/25/6/887To request permission to re-use all or part of this article, use this link

on June 16, 2020. © 2016 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

Published OnlineFirst April 6, 2016; DOI: 10.1158/1055-9965.EPI-15-1223