Identification of Serum Metabolites Associated With Risk of Type 2 Diabetes Using a Targeted Metabolomic Approach.

0.46 MB·English

by  Floegel, Anna

ORIGINALARTICLE
fi
Identi cation of Serum Metabolites Associated With Risk
of Type 2 Diabetes Using a Targeted
Metabolomic Approach
Anna Floegel,1 Norbert Stefan,2 Zhonghao Yu,3 Kristin Mühlenbruch,4 Dagmar Drogan,1
Hans-Georg Joost,5 Andreas Fritsche,2 Hans-Ulrich Häring,2 Martin Hrabe(cid:1) de Angelis,6
Annette Peters,7 Michael Roden,8,9 Cornelia Prehn,6 Rui Wang-Sattler,3 Thomas Illig,3,10
Matthias B. Schulze,4 Jerzy Adamski,6 Heiner Boeing,1 and Tobias Pischon1,11
Metabolomicdiscoveryofbiomarkersoftype2diabetes(T2D)risk
mayrevealetiologicalpathwaysandhelptoidentifyindividualsat T
risk for disease. We prospectively investigated the association ype 2 diabetes (T2D) is characterized by im-
betweenserummetabolitesmeasuredbytargetedmetabolomics paired insulin sensitivity of several tissues
and risk of T2D in the European Prospective Investigation into and inadequate insulin secretion from b-cells
Cancer and Nutrition (EPIC)-Potsdam (27,548 adults) among all
(1). A detailed understanding of the patho-
incidentcasesofT2D(n=800,meanfollow-up7years)andaran-
physiology of T2D is a prerequisite for the development
domlydrawnsubcohort(n=2,282).Flowinjectionanalysistan-
of preventive strategies. In particular, the identification
dem mass spectrometry was used to quantify 163 metabolites,
includingacylcarnitines,aminoacids,hexose,andphospholipids, of early metabolic alterations is promising in the
in baseline serum samples. Serum hexose; phenylalanine; and study of etiological pathways and may further help to
diacyl-phosphatidylcholines C32:1, C36:1, C38:3, and C40:5 were identify high-risk individuals. A number of biomarkers
independently associated with increased risk of T2D and serum have been proposed as indicators for the estimation of
glycine; sphingomyelin C16:1; acyl-alkyl-phosphatidylcholines T2D risk, such as fasting plasma glucose and glycated
C34:3,C40:6,C42:5,C44:4,andC44:5;andlysophosphatidylcholine hemoglobin A (HbA ) (2), triglycerides (3), HDL
C18:2withdecreasedrisk.Varianceofthemetaboliteswaslargely cholesterol (4)1,cinflamm1catory markers (5), adiponectin
explained by two metabolite factors with opposing risk associa-
tions(factor1relativeriskinextremequintiles0.31[95%CI0.21– (5,6), liver enzymes (7), and fetuin-A (8). However,
0.44], factor 2 3.82 [2.64–5.52]). The metabolites significantly most biomarkers fail to grasp the complexity of T2D
improvedT2Dpredictioncomparedwithestablishedriskfactors. etiology (3). Design and advancement of high-through-
Theywerefurtherlinkedtoinsulinsensitivityandsecretioninthe put analytical techniques determined the emergence
TübingenFamilystudyandwerepartlyreplicatedintheindepen- of metabolomics, which is the simultaneous study of
dentKORA(CooperativeHealthResearchintheRegionofAugs- numerous low-molecular weight compounds, namely
burg) cohort. The data indicate that metabolic alterations, metabolites. Metabolites represent intermediates and
includingsugarmetabolites,aminoacids,andcholine-containing end products of metabolic pathways that reflect more
phospholipids,areassociatedearlyonwithahigherriskofT2D.
Diabetes 62:639–648, 2013 rapidly physiological dysfunctions than current bio-
markers and, thus, may mirror earlier stages of T2D (9).
Cross-sectional studies have linked alterations in meta-
Fromthe1DepartmentofEpidemiology,GermanInstituteofHumanNutrition bolic profiles with obesity (10), glucose tolerance (11),
Potsdam-Rehbruecke,Nuthetal,Germany;the2DepartmentofInternalMed- and prevalent diabetes (12–14). The most prominent
icine IV, Divisions of Endocrinology, Diabetology, Nephrology, Vascular
metabolic shifts involved blood acylcarnitines and
Disease, and Clinical Chemistry, University of Tübingen, Tübingen, Ger-
many;the3ResearchUnitofMolecularEpidemiology,HelmholtzZentrum branched-chain amino acids (BCAAs). Recently, obser-
München,GermanResearchCenterforEnvironmentalHealth,Neuherberg, vations from a prospective study found that a set of five
Germany;the4DepartmentofMolecularEpidemiology,GermanInstituteof
Human Nutrition Potsdam-Rehbruecke, Nuthetal, Germany; the 5Depart- amino acids was predictive for T2D, representing pio-
ment of Pharmacology, German Institute of Human Nutrition Potsdam-- neering work in the emerging field of systems epidemi-
Rehbruecke, Nuthetal, Germany; the 6Institute of Experimental Genetics, ology (15).
HelmholtzZentrumMünchen,GermanResearchCenterforEnvironmental
Health,Neuherberg,Germany;the7InstituteofEpidemiologyII,Helmholtz In the current study, we investigated whether a tar-
Zentrum München, German Research Center for Environmental Health, geted metabolomic approach involving a broader spec-
Neuherberg,Germany;the8InstituteofClinicalDiabetology,GermanDia- trum of metabolites and a larger number of study
betesCenter,LeibnizCenterforDiabetesResearchatHeinrichHeineUni-
versity, Düsseldorf, Germany; the 9Department of Metabolic Diseases, participants may help to identify metabolites associated
UniversityClinics,Düsseldorf,Germany;the10HannoverUnifiedBiobank, with the risk of T2D and the mechanisms involved.
HannoverMedicalSchool,Hannover,Germany;andthe11MolecularEpide- Therefore,weprofiled163serummetabolitesinoriginally
miologyGroup,MaxDelbrückCenterforMolecularMedicine(MDC)Berlin-
healthy individuals who were consecutively followed up
Buch,Berlin,Germany.
Correspondingauthor:AnnaFloegel,anna.fl[email protected]. for incident T2D in two large-scale prospective cohort
Received22April2012andaccepted22July2012. studies in Germany and studied cross-sectional relation-
DOI:10.2337/db12-0495 ships of the identified metabolites with insulin sensitivity
This article contains Supplementary Data online at http://diabetes
.diabetesjournals.org/lookup/suppl/doi:10.2337/db12-0495/-/DC1. and secretion in precisely phenotyped participants. In
(cid:1)2013bytheAmericanDiabetesAssociation.Readersmayusethisarticleas addition, we evaluated the usefulness of the metabolites
longastheworkisproperlycited,theuseiseducationalandnotforprofit,
for T2D risk prediction compared with the German
and the work is not altered. See http://creativecommons.org/licenses/by
-nc-nd/3.0/fordetails. Diabetes Risk Score (DRS) (16) and established bio-
See accompanying commentary, p. 349. markers.
diabetes.diabetesjournals.org DIABETES,VOL.62,FEBRUARY2013 639
SERUMMETABOLITESANDRISKOFTYPE2DIABETES
RESEARCHDESIGNANDMETHODS Serum metabolite concentrations. Serum concentrations of metabolites
European Prospective Investigation into Cancer and Nutrition- were determined with the AbsoluteIDQ p150 and p180 Kits (Biocrates Life
Potsdam study. The European Prospective Investigation into Cancer and Sciences AG, Innsbruck, Austria) using the flow injection analysis tandem
Nutrition(EPIC)-PotsdamispartoftheongoingmulticenterEPICstudyand mass spectrometry (FIA-MS/MS) technique (30). The metabolomic method
comprises27,548participantsfromthegeneralpopulationintheareaofPotsdam simultaneouslyquantified163metabolites,including41acylcarnitines(Cx:y),
ineasternGermanywhoweremainly35–65yearsofageattimeofrecruitment 14 amino acids, 1 hexose (sum of six-carbon monosaccharides without dis-
tinction of isomers), 92 glycerophospholipids (lyso-, diacyl-, and acyl-alkyl-
between1994and1998(17).Atbaseline,participantsunderwentanexamina-
tionbyqualifiedstaff,includingmedicalhistory,bloodpressuremeasurement, phosphatidylcholines),and15sphingomyelins.Toensurevalidmeasurements,
metabolites below the limit of detection (n = 30) and those with very high
andanthropometry(18).Participantsalsocompletedsociodemographicand
analytical variance (n = 6) in our samples were excluded, leaving 127
lifestylequestionnairesandavalidatedfoodfrequencyquestionnaire.Inaddi-
metabolitesforthepresentanalysis.
tion,30mLbloodweredrawn(randomsampling)andimmediatelyprocessed
MetabolomicmeasurementswereperformedintheGenomeAnalysisCenter
(19).Onlyparticipantswithmorningappointmentswereaskedtofastover-
attheHelmholtzZentumMünchen.Samplepreparationwasdoneaccordingto
night.Bloodwasfractionatedintoserum,plasma,buffycoat,anderythrocytes;
themanufacturer’sprotocol(Biocratesuser’smanualUM-P150)andhasbeen
aliquottedintostrawsof0.5mLeach;andstoredintanksofliquidnitrogenat
2196°Cuntilanalysis.Besidesmetabolomicprofiling,otherbiomarkershave described previously (30). In brief, after centrifugation, 10 mL serum were
inserted into a filter on a 96-well sandwich plate, which already contained
beenmeasuredinbaselinebloodsamplesasdescribedpreviously(7,20,21).
Every2–3years,follow-upquestionnairesweresenttoparticipantstoidentify stableisotope-labeledinternalstandards.Aminoacidswerederivatedwith5%
incidentcasesofT2D,withresponseratesof;95%(22,23).Onceaparticipant phenylisothiocyanate reagent. Metabolites and internal standards were
was identified as a potential case, disease status was further verified with extracted with 5 mmol/L ammonium acetate in methanol. The solution was
thencentrifugedthroughafiltermembraneanddilutedwithmassspectrom-
medicalrecords,includingthecorrectdiagnosis(InternationalClassificationof
etry running solvent. Final extracts were analyzed by FIA-MS/MS, and
Diseases,10threvision,E11,non–insulin-dependentdiabetes),thedateofthe metaboliteswerequantifiedinmmol/Lbyappropriateinternalstandards.The
diagnosis, and the means of diagnosis confirmation. This verification was methodhasbeenvalidated,andanalyticalspecificationswereprovidedinthe
achievedbysendingastandardinquiryformtothetreatingphysician.Consent
BiocratesmanualAS-P150.Themanufacturerselectedthemetabolitesbased
wasobtainedfromallstudyparticipantsapriori,andthestudywasapprovedby
on the robustness of their measurements. The uncertainty of the measure-
theethicscommitteeoftheMedicalSocietyoftheStateofBrandenburg. mentswas,10%formostofthemetabolites.Regardingaccuracy,allincluded
We constructed a case-cohort study within EPIC-Potsdam, including all metaboliteswerefoundintherangeof80–115%oftheirtheoreticalvalues.The
incidentcasesofT2Dofthefullcohortidentifiedupto31August2005(n=
mediananalyticalvarianceofEPIC-Potsdamsampleswasa7.3%within-plate
849,meanfollow-up7years),andasubcohort(n=2,500)randomlydrawn coefficient of variation and a 11.3% between-plates coefficient of variation
fromtheEPIC-Potsdamstudypopulation.Becausethesubcohortwasrepre-
(31).Toaccountforrun-ordereffects,serumsampleswererandomlyanalyzed
sentativeofthefullcohort,itincluded2,415noncasesand85casesofincident
together,regardlessofthecasestatus.Wehaveshownpreviouslythatmostof
T2D (i.e., internal cases). The remaining 764 cases did not belong to the the metabolites had moderate to high intraclass correlation coefficients
subcohort(externalcases).Byrandomlyselectingthesubcohortandusingthe
measured in participants over a 4-month period, indicating reasonable re-
appropriatestatisticforthisstudydesign,thebiomarkersonlyneededtobe
liabilityofthemeasurements(31).
measuredinthecase-cohortsample;however,theresultsareexpectedtobe
Statisticalanalysis
generalizable to the full cohort (7). The case-cohort design was previously Step1:IdentificationofmetabolitesassociatedwithT2DriskinEPIC-
chosenbasedonitsadvantages,includingareducedchanceofselectionbias Potsdam.Coxproportionalhazardsregressionwithweightingassuggestedby
forthecontrolgroup(24).
Prentice (32) and robust sandwich covariance estimates to account for the
Forthepresentanalysis,wefurtherexcludedparticipantswithprevalent
case-cohortdesignwereusedtocalculatemultivariable-adjustedhazardratios
T2D at baseline (n = 110), with missing or nonverified data on incident or
asameasureofrelativerisk(RR)and95%CI,withageastheunderlyingtime
prevalentT2D(n=13),withmissingbloodsamplesorbiomarkermeasure-
scale from recruitment to study exit (T2D diagnosis or censoring) of each
ments (n=64), andwithmissing covariate information(n=80).Thus, the participant. We considered z score–standardized metabolite concentrations
analytical sample included 2,282 individuals of the subcohort and 800 indi-
(mean 0 [SD 1]) as the exposure variable and calculated a multivariable-
vidualswithincidentT2D.
adjustedmodeltoselectmetabolitesassociatedwithT2Drisk.Thismodel
Cooperative Health Research in the Region of Augsburg study. The
was adjusted for age, sex, alcohol intake from beverages (nonconsumers;
CooperativeHealthResearchintheRegionofAugsburg(KORA)studyconsists women.0–6,6–12,and.12g/day;andmen.0–12,12–24,and.24g/day),
ofpopulation-basedsurveysandfollow-upperiodsintheareaofAugsburgin smoking(never,former,current#20cigarettes/day,current.20cigarettes/
southernGermany.Atotalof4,261individualsbetween25and74yearsofage day),cyclingandsports(h/week),education(nodegree/vocationaltraining,
participatedintheS4surveybetween1999and2001(25).IntheKORAcohort, trade/technicalschool,universitydegree),coffeeintake(cups/day),redmeat
bloodwasdrawnintoserumgeltubesafterafastingperiodofatleast8h. intake(g/day),whole-grainbreadintake(g/day),prevalenthypertension(yes/
Bloodsampleswererestedforcoagulationfor30minatroomtemperature; no), BMI (kg/m2), and waist circumference (cm). Because the metabolomic
serumwasobtainedbycentrifugationat2,750gat15°Cfor10minandstored approachisexploratory,thePvaluesfromCoxregressionwerecorrectedfor
inafreezerat280°C.Atotalof3,080individualstookpartintheF4follow-up multiple testing (n = 127) using the Bonferroni-Holm procedure (33), and
surveyduringtheyears2006–2008(26).TheidentificationofincidentT2Dwas acorrectedP,0.05(two-sidedtesting)wasconsideredsignificanttoselect
basedonanoralglucosetolerancetest(OGTT)oravalidatedphysiciandi- metabolites.
agnosis(27).AllKORAparticipantsgavewritteninformedconsent,andthe Wenextcalculatedamodelthatincludedthecovariatesandalltheidentified
KORAstudywasapprovedbytheethicscommitteeoftheBavarianmedical metabolitesandusedstepwiseCoxregressiontoselecttheindependentpre-
association. dictors.Toalsoaccountfortheintercorrelationofsomeofthemetabolites,we
Asubcohortof876S4participants55–74yearsofagewithoutT2Datbaseline conductedaprincipalcomponentanalysis(PCA).Inbrief,thePCAaggregatesthe
andwithmetabolomicsdataavailablewasincludedinthecurrentstudy.Of individualmetabolitesbasedontheirdegreeofcorrelationwithoneanotherto
them,91developedincidentT2Dduringthe7-yearfollow-up. asmallernumberofmetabolitefactors(principalcomponents).Thesemetabolite
TübingenFamilystudyforT2D.TheTübingenFamily(TüF)studyisinan factorsareextractedinawaythatexplainsthemajorfractionofthevarianceof
ongoing investigation of the pathophysiology of T2D in southern Germany individualmetabolites.WeincludedallmetabolitesassociatedwithT2Driskin
(28).Individualsmeetingatleastoneofthefollowingcriteriawereincludedin thePCA,basedthePCAonthecorrelationmatrixofmetabolites,andusedan
thestudy:afamilyhistoryofT2D,aBMI.27kg/m2,andpreviousimpaired orthogonalrotationprocedurewiththevarimaxmethod.Weretainedtwome-
glucose tolerance or gestational diabetes mellitus. They were considered tabolitefactorsaccordingtothescreetestandbecausetheyaccountedformost
healthy according to a physical examination and routine laboratory tests. oftheobservedvariance.Thus,theproportionofexplainedvarianceoffactor1
Writteninformedconsentwasobtainedfromallparticipants,andthemedical andfactor2were34.2and16.1%,respectively.Toinvestigatetheassociation
ethicscommitteeoftheUniversityofTübingenapprovedtheprotocol. betweenmetabolitefactorsandT2Drisk,weestimatedRRand95%CIacross
Allindividualsunderwenta75-gOGTT.Venousplasmasamplesweredrawn quintiles of metabolite factors, particularly choosing quintiles of metabolite
at 0, 30, 60, 90, and 120 min for plasma glucose, insulin, C-peptide, and factorstofacilitatetheinterpretation.Wealsoinvestigatedapossibleeffect
metabolomicanalyses(minute0).Insulinsensitivitywascalculatedfromthe modificationofsexorfastingstatusontheassociationbetweenmetabolitefac-
OGTT(29).TheplasmaglucoseandC-peptideareasunderthecurve(AUCs) torsandT2Driskbyincludingmultiplicativeinteractiontermsintothemodels.We
duringtheOGTTwerecalculatedbyapplyingthetrapezoidmethod.Insulin thenrepeatedthePCAtoincludeonlyfastingbloodsamplesinordertoevaluate
secretion was calculated from AUC /AUC . The present analysis whetherthemetabolitefactorsweredifferentfromthoseobtainedfromrandom
C-peptide glucose
included76CaucasiansfromtheTüFstudywhohadmeasurementsofinsulin bloodsamples.Finally,wecalculatedhazardratesofT2Dduringdifferentperiodsof
sensitivityandinsulinsecretionaswellasmetabolomicsdataavailable. follow-upandtestedwhethertheyweredifferentwithatestofheterogeneity(34).
640 DIABETES,VOL.62,FEBRUARY2013 diabetes.diabetesjournals.org
A.FLOEGELANDASSOCIATES
Step2:Additionalanalyses,riskprediction,andreplicationinKORA. the others (Table 2). Specifically, hexose, phenylalanine,
For significant metabolites found in step 1, we calculated multivariable- and diacyl-phosphatidylcholines C32:1, C36:1, C38:3, and
adjusted models with additional adjustment for blood glucose, HbA1c, HDL C40:5 were significantly positively associated with T2D
cholesterol,andtriglycerides.Wealso adjusted for theaminoacids phenyl-
risk, whereas glycine, sphingomyelin C16:1, lysophospha-
alanine,tyrosine,andisoleucine,whichhavebeenfoundtobeassociatedwith
tidylcholine C18:2, and acyl-alkyl-phosphatidylcholines
T2D risk (35). Using data of the EPIC-Potsdam subcohort, we calculated
Spearmanpartialcorrelationcoefficientsbetweenidentifiedmetabolitesand C34:3, C40:6, C42:5, C44:4, and C44:5 were significantly
established T2D biomarkers. Data of the TüF study were used to calculate inverselyrelatedtoT2Drisk.UsingPCA,weidentifiedtwo
Spearmanpartialcorrelationcoefficientsbetweenidentifiedmetabolitesand metabolite factors that included multiple metabolites and
measuresofinsulinsensitivityandsecretion.Wecalculatedmeasuresofdis- explainedmostoftheirvariation(Fig.1).Thesemetabolite
crimination and calibration in different multivariable-adjusted logistic re- factorsshowedsignificant andopposingassociations with
gression models using the DRS (16) as the reference model and adding
establishedT2Dbiomarkersandtheidentifiedmetabolites.Receiveroperating T2D risk. When comparing extreme quintiles, metabolite
characteristic(ROC)AUCswerecomparedusingtheDeLongtest(36). factor 1, which mainly contains acyl-alkyl-phosphatidylcho-
TheresultswerereplicatedintheprospectiveKORAstudy,andmetabolite lines, sphingomyelins, and lysophosphatidylcholines, was
factorswererecalculatedinKORAusingthelinearfactorequationsretrieved associated with a significant 69% reduced risk of T2D (RR
from the PCA in the EPIC-Potsdam sample. The risk estimates from EPIC- 0.31 [95% CI 0.21–0.44]) (Table 3), whereas metabolite
Potsdam and KORA were combined using a meta-analytical approach (37).
factor 2, consistent of diacyl-phosphatidylcholines, BCAA
PowercalculationsuggestedadetectableRRperSDof1.26(38).Additionally,
andaromaticaminoacids,propionylcarnitine,andhexose,
multivariable-adjustedRRsofT2Dwerecalculatedfortheaminoacidsthat
wererecentlyidentifiedbyWangetal.(35)andthatwerealsomeasuredinthe wasassociatedwithasignificant3.82-foldincreasedriskof
EPIC-Potsdamstudy.ThestatisticalanalyseswereconductedwithSASver- T2D (3.82 [2.64–5.52]). When we restricted the PCA to
sion 9.2 (SAS Institute, Inc, Cary, NC) and in the R statistical environment fasting samples (n = 429), very similar metabolite factors
(www.r-project.org). could be generated with only minor differences in factor
loadings (Supplementary Table 2). We also estimated the
RESULTS joint effects of both metabolite factors by summing them
Baseline characteristics of the EPIC-Potsdam sample are (factor1receivedanegativesignbecauseitwasinversely
presentedinTable1.Ofallthemetabolitesmeasuredusing associated with T2D risk) and calculating RR of T2D
a targeted metabolomic approach, 34 were significantly across quintiles of combined factors. The RR (95% CI) of
associated with T2D risk in the EPIC-Potsdam study after T2D from quintile 1 to quintile 5 of summed metabolite
correction for multiple testing (Supplementary Table 1). factors was as follows: 1.0, 1.47 (0.96–2.24), 2.29 (1.54–
These relations were also independent of relevant dietary 3.38), 2.67 (1.79–4.0), and 6.69 (4.50–9.96) (P for trend
and lifestyle factors as well as anthropometry and hyper- ,0.0001).
tension.Amongthese34metabolites,14wereidentifiedto Adjustment for established T2D biomarkers only mar-
be significantly associated with T2D risk independent of ginallyaffectedthemagnitudeofriskassociationformost
TABLE1
Baselinecharacteristics oftheEPIC-Potsdamcase-cohort sample(1994–1998)
Subcohort IncidentT2D
(n=2,282) cases(n=800) Pfordifference†
Age(years)‡ 49.5(8.9) 54.7(7.3) ,0.0001
Women(%)‡ 62.0 42.2 ,0.0001
BMI(kg/m2) 26.1(0.09) 30.1(0.15) ,0.0001
Waist circumference,men(cm)§ 93.7(0.34) 103.6(0.46) ,0.0001
Waist circumference,women(cm)§ 80.6(0.30) 93.4(0.62) ,0.0001
Prevalenthypertension 49.5 70.8 ,0.0001
Education
Nodegree/vocational training 37.1 45.6 ,0.0001
Trade/technical school 24.0 25.4 0.412
Universitydegree 39.0 29.0 ,0.0001
Smoking status
Never 46.9 36.2 ,0.0001
Former 33.0 42.3 ,0.0001
Current 20.1 21.5 0.472
Among smokers,cigarettes/day 12.6(0.43) 16.0(0.74) ,0.0001
Physicalactivity(h/week)¶ 2.8(0.07) 2.2(0.13) ,0.0001
Alcoholintakefrombeverages(g/day) 14.8(0.41) 14.5(0.71) 0.761
Coffeeconsumption (cups/day) 2.8(0.04) 2.7(0.08) 0.148
Whole-grainbreadintake(g/day) 45.9(1.11) 38.2(1.91) 0.0003
Redmeatintake(g/day) 43.3(0.61) 48.9(1.06) ,0.0001
Biomarkers
Glucose(mg/dL) 88.1(0.53) 107.0(0.92) ,0.0001
HbA (%)‖ 5.42(0.01) 6.30(0.03) ,0.0001
1c
HDLcholesterol(mg/dL) 47.5(0.25) 40.9(0.43) ,0.0001
Triglycerides(mg/dL) 114.8(2.12) 177.2(3.65) ,0.0001
Data are age- and sex-adjusted mean (SE) for continuous variables or % for categorical variables. †P for difference comparing incident
cases of T2D with noncases of the subcohort. ‡Unadjusted mean (SD) or %. §Age-adjusted mean (SE). ¶Average of cycling and sports
duringsummerandwinterseasons.‖Datawereonlyavailableforn=2,900.
diabetes.diabetesjournals.org DIABETES,VOL.62,FEBRUARY2013 641
SERUMMETABOLITESANDRISKOFTYPE2DIABETES
EPIC-Potsdam RRperSD(95%CI) +Glucose+HbA1c+HDLchol‡‡‡‡+HbA+Triglycerides+HDLchol+Triglycerides1c ––––2.11(1.822.44)2.43(2.112.79)2.34(2.042.69)1.51(1.281.80)––––1.37(1.231.52)1.34(1.211.49)1.30(1.161.44)1.25(1.121.39)––––0.81(0.710.93)0.74(0.650.84)0.74(0.650.84)0.89(0.781.01)––––0.74(0.670.83)0.87(0.780.97)0.79(0.710.88)0.91(0.801.03)––––0.67(0.580.77)0.72(0.630.82)0.66(0.580.74)0.92(0.781.07)––––0.83(0.740.94)0.82(0.730.92)0.76(0.670.85)0.94(0.821.08)––––0.82(0.720.92)0.76(0.680.86)0.73(0.640.82)0.96(0.831.12)––––0.84(0.750.96)0.82(0.720.93)0.77(0.680.87)0.97(0.841.12)––––0.81(0.710.91)0.77(0.680.87)0.74(0.650.83)0.95(0.821.09)––––1.12(1.021.23)1.23(1.111.36)1.12(1.021.22)1.15(1.031.29)––––1.09(0.981.21)1.26(1.141.39)1.07(0.961.19)1.25(1.101.41)––––1.31(1.181.45)1.38(1.251.52)1.24(1.111.38)1.38(1.221.55)––––1.12(1.021.24)1.21(1.101.32)1.07(0.961.18)1.19(1.061.32)––––0.77(0.680.88)0.79(0.700.90)0.72(0.630.81)0.84(0.730.96) †phosphatidylcholine;SM,sphingomyelin.MetaboliteswereselectedbystepwiseCoxntrationstoameanof0(SD1)andincludingallmetabolitesassociatedwithT2Drisk.––ustedforage,sex,alcoholintakefrombeverages(nonconsumers;women06,612,.s/day,current20cigarettes/day),physicalactivity(cyclingandsportsinh/week),2),andwaistcircumferencentake(g/day),prevalenthypertension(yes/no),BMI(kg/ms.
TABLE2fiMeanserumconcentrationsofidentiedmetabolitesandtheirassociationwithriskofT2Din CasesSubcohortnn(=800)(=2,282)mm†‡(mol/L)(mol/L)Basicmodel+Glucose ––Hexose4,698(984)5,783(2,167)2.36(2.062.71)1.62(1.371.91)––Phenylalanine56.4(12.0)61.0(12.9)1.35(1.221.49)1.24(1.121.37)––Glycine256(77.2)227(59.6)0.73(0.640.83)0.81(0.710.92)––SMC16:117.6(3.86)17.4(4.02)0.80(0.720.88)0.82(0.730.92)––PCaeC34:38.51(2.48)7.01(2.02)0.64(0.560.72)0.70(0.610.80)––PCaeC40:65.52(1.43)5.03(1.34)0.77(0.690.86)0.81(0.720.92)––PCaeC42:52.35(0.54)2.06(0.46)0.71(0.630.79)0.76(0.670.86)––PCaeC44:40.38(0.10)0.33(0.09)0.76(0.670.85)0.77(0.680.87)––PCaeC44:51.77(0.48)1.53(0.43)0.71(0.630.80)0.74(0.650.84)––PCaaC32:116.9(9.76)20.4(11.3)1.18(1.081.29)1.12(1.021.24)––PCaaC36:157.3(15.6)62.1(17.2)1.18(1.081.30)1.18(1.071.30)––PCaaC38:355.4(14.8)65.3(18.2)1.37(1.241.51)1.35(1.221.50)––PCaaC40:511.4(3.49)12.8(4.17)1.18(1.081.29)1.21(1.101.33)––LysoPCaC18:235.2(13.7)29.6(11.1)0.74(0.660.84)0.72(0.640.81) Dataaremean(SD)metaboliteconcentrations.a,acyl;aa,diacyl;ae,acyl-alkyl;chol,cholesterol;PC,fiproportionalhazardregressionmodiedbythePrenticemethodafterstandardizingmetaboliteconceaswellasrelevantcovariates.RRsforthebasicmodelwerethencalculatedinacontinuousmodeladj..––.#and12g/day;men012,1224,and24g/day),smoking(never,former,current20cigaretteeducation(low,medium,high),coffeeintake(cups/day),redmeatintake(g/day),whole-grainbreadi‡fi(cm).Adjustmentaccordingtothebasicmodelwithadditionaladjustmentforspeciedbiomarker
642 DIABETES,VOL.62,FEBRUARY2013 diabetes.diabetesjournals.org
A.FLOEGELANDASSOCIATES
FIG.1.TwometabolitefactorsassociatedwithriskofT2D.Presentedisatwo-dimensionalfactorloadingplotobtainedfromPCA.Forsimple
interpretation,metabolitesthatclustertogetherintheplotarerelatedtooneanother.Metabolitespresentedinblueareassociatedwithde-
creasedriskofT2D,whereasmetabolitespresentedinredareassociatedwithincreasedriskofT2D.Morespecifically,thefactorloadingsrep-
resent the correlation coefficients of individual metabolites with corresponding metabolite factors and may range from 21 to 1. They were
identified byPCA basedon thecorrelation matrixof allmetabolitessignificantly associated with riskof T2Din theEPIC-Potsdamstudy. An
orthogonalvarimaxrotationwasused,andtwofactorswereretainedbecausetheyaccountedfor>50%oftheobservedvariance.a,acyl;aa,diacyl;
ae,acyl-alkyl;C3,propionylcarnitine;Gly,glycine;H1,hexose;PC,phosphatidylcholine;Phe,phenylalanine;SM,sphingomyelin;Trp,tryptophan;
Tyr,tyrosine;Val;valine;xLeu,isoleucine.
of the metabolites (Table 2). An exception was that the potentialofidentifiedmetabolitestodiscriminatebetween
inverse association between acyl-alkyl-phosphatidylcho- T2DcasesandnoncaseswascomparabletothatoftheDRS
lines and sphingomyelin C16:1 and T2D risk was attenu- (16) (ROC AUC 0.849 and 0.847, respectively, P for differ-
ated and no longer significant after adjustment for blood ence=0.838)(Fig.2).Whenthemetaboliteswereaddedto
glucose, HbA HDL cholesterol, and triglycerides. The established risk prediction models of T2D, discrimination
1c,
positive association between hexose and T2D risk was wasslightlybutsignificantlyimproveduptoaROCAUCof
considerably weakened but remained significant after ad- 0.912, and these models were well calibrated (Fig. 2 and
justment for plasma glucose. When we also adjusted for Supplementary Table 4). Replication in KORA revealed
phenylalanine, tyrosine, and isoleucine, the associations significantassociationswithT2Driskformetabolitefactor2
for metabolite factor 1 were unchanged. The associations andhexose(Table3andSupplementaryTable5).InKORA,
for metabolite factor 2, which included these three amino similar trends as in EPIC-Potsdam were seen for metab-
acids, were weakened but remained significant (data not olite factor 1, acyl-alkyl-phosphatidylcholines, glycine,
shown). lysophosphatidylcholine C18:2, and sphingomyelin C16:1,
We further observed that metabolites were linked to withborderlinesignificance.Althoughtheriskestimatesfor
established T2D biomarkers (Supplementary Table 3). diacyl-phosphatidylcholines were considerably lower in
Specifically,metabolitefactor1,whichwasinverselyasso- KORAthaninEPIC-Potsdam,therewasnosignificanthet-
ciated with T2D risk, was negatively correlated to plasma erogeneity between studies. We also calculated the RR of
glucose, HbA , and triglycerides and positively related to T2D for the BCAA and aromatic amino acids, which have
1c
HDL cholesterol and adiponectin. Metabolite factor 2, recentlybeenreportedtobeassociatedwithT2Driskinthe
which was positively associated with T2D risk, was posi- Framingham Offspring cohort and the Malmö Diet and
tivelycorrelatedwithtriglyceridesandliverenzymes.Data Cancer study, to facilitate the comparison (35). In EPIC-
from the TüF study revealed that acyl-alkyl-phosphati- Potsdam, isoleucine, valine, tyrosine, and phenylalanine
dylcholines,lysophosphatidylcholineC18:2,andglycinewere were positively associated with T2D risk (RR per SD 1.30
positivelyassociatedwithinsulinsensitivity,whereashex- [95% CI 1.17–1.43], 1.27 [1.16–1.40], 1.31 [1.18–1.45], 1.35
oseanddiacyl-phosphatidylcholineswereinverselyrelated [1.22–1.49], respectively); leucine was not measured in
toinsulinsensitivity(Table4).Furthermore,phenylalanine EPIC-Potsdam. When combining isoleucine, tyrosine, and
was positively associated with insulin secretion, whereas phenylalanine,theRRofT2Dfromlowesttohighestquartile
hexose, sphingomyelin C16:1, and acyl-alkyl-phosphati- was 1.0, 1.13 (0.82–1.54), 1.45 (1.07–1.98), and 2.18 (1.62–
dylcholineswereinverselyrelatedtoinsulinsecretion.The 2.95),respectively(Pfortrend,0.0001).
diabetes.diabetesjournals.org DIABETES,VOL.62,FEBRUARY2013 643
SERUMMETABOLITESANDRISKOFTYPE2DIABETES
TABLE3 TABLE4
RRofT2Dbyquintilesofmetabolitefactors Correlation between metabolites associated with T2D risk and
measuresofinsulinsensitivityandsecretionintheTüFstudy
EPIC-Potsdam Replication inKORA
Insulinsensitivity Insulinsecretion
Cases Cases
(N) RR(95%CI)* (N) RR(95%CI)* Hexose 20.52 20.39
Factor1† Phenylalanine 20.08 0.24
Glycine 0.34 0.08
Quintile1 296 1.00 37 1.00
Quintile2 194 0.61(0.46–0.80) 15 0.59(0.31–1.33) SMC16:1 0.01 20.13
Quintile3 148 0.50(0.37–0.67) 20 0.60(0.31–1.14) PCaeC34:3 0.15 20.24
Quintile4 95 0.37(0.27–0.51) 12 0.56(0.27–1.21) PCaeC40:6 0.25 20.08
Quintile5 67 0.31(0.21–0.44) 7 0.49(0.17–1.25) PCaeC42:5 0.11 20.24
Pfortrend 2.95310213 1.0031021 PCaeC44:4 0.08 20.21
Factor2† PCaeC44:5 0.12 20.30
PCaaC32:1 20.15 20.02
Quintile1 51 1.00 6 1.00
Quintile2 102 1.23(0.82–1.85) 9 1.38(0.45–4.25) PCaaC36:1 20.05 20.09
Quintile3 149 1.72(1.17–2.51) 15 1.50(0.52–4.32) PCaaC38:3 20.24 0.01
Quintile4 191 1.80(1.24–2.63) 18 1.81(0.62–5.23) PCaaC40:5 0.07 0.12
Quintile5 307 3.82(2.64–5.52) 43 4.95(1.96–12.48) LysoPCaC18:2 0.36 20.14
Pfortrend 6.64310218 1.1031025 Factor1† 0.23 20.27
Factor2‡ 20.15 0.00
a,acyl;aa,diacyl;ae,acyl-alkyl;PC,phosphatidylcholine;SM,sphin-
gomyelin. *Relative risks were calculated with multivariate Cox re- DataarepartialSpearmancorrelationcoefficients.Insulinsensitivity
gression (using the Prentice method to account for the case-cohort wasadjustedforageandsex;insulinsecretionwasadjustedforage,
designinEPIC-Potsdam)acrossquintilesofmetabolitefactorsafter sex, and insulin sensitivity. Insulin sensitivity and insulin secretion
standardizing metaboliteconcentrationstoameanof0(SD1).The were estimated from OGTTs in the TüF study (n = 76). a, acyl; aa,
modelwasadjustedforage,sex,alcoholintakefrombeverages(non- diacyl; ae, acyl-alkyl; PC, phosphatidylcholine; SM, sphingomyelin.
consumers; women .0–6, 6–12, and .12 g/day; men .0–12, 12–24, †Factor1=(0.823PCaaC42:0)+(0.793PCaaC42:1)+(0.803
and.24g/day),smoking(never,former,current#20cigarettes/day, PC ae C32:1) + (0.78 3 PC ae C32:2) + (0.70 3 PC ae C34:2) +
current.20cigarettes/day),physicalactivity(cyclingandsportsinh/ (0.72 3 PC ae C34:3) + (0.71 3 PC ae C36:2) + (0.71 3 PC ae
week),education(low,medium,high),coffeeintake(cups/day),red C36:3) + (0.85 3 PC ae C40:5) + (0.76 3 PC ae C40:6) + (0.82 3
meatintake(g/day),whole-grainbreadintake(g/day),prevalenthy- PC ae C42:3) + (0.85 3 PC ae C42:4) + (0.87 3 PC ae C42:5) +
pertension(yes/no),BMI(kg/m2),andwaistcircumference(cm).†By (0.76 3 PC ae C44:4) + (0.78 3 PC ae C44:5) + (0.83 3 PC ae
conductingaPCAwithorthogonalvarimaxrotationofallmetabolites C44:6)+(0.543SMC16:1)+(0.573SMOHC22:2)+(0.413lysoPC
associated with T2D risk in EPIC-Potsdam, two metabolite factors aC17:0).‡Factor2=(0.553propionylcarnitine) +(0.663phenyl-
couldbeidentifiedthatexplained.50%ofthevariationofthemetab- alanine)+(0.613tryptophan)+(0.663tyrosine)+(0.683valine)+
olites. The corresponding linear factor equations (which equal (0.663isoleucine)+(0.593PCaaC32:1)+(0.703PCaaC36:1)+
the summed products of each metabolite’s standardized concentra- (0.653PCaaC36:3)+(0.763PCaaC38:3)+(0.723PCaaC40:4)+
tion and corresponding factor loading) were as follows: Factor 1 = (0.713PCaaC40:5)+(0.443hexose).
(0.803PCaeC32:1)+(0.783PCaeC32:2)+(0.703PCaeC34:2)+
(0.723PCaeC34:3)+(0.713PCaeC36:2)+(0.713PCaeC36:3)+
(0.853PCaeC40:5)+(0.763PCaeC40:6)+(0.823PCaeC42:3)+ during the first 2 years of follow-up (n = 208). The risk
(0.853PCaeC42:4)+(0.873PCaeC42:5)+(0.763PCaeC44:4)+ associations were slightly lower, but not markedly differ-
(0.783PCaeC44:5)+(0.833PCaeC44:6)+(0.823PCaaC42:0)+ ent.
(0.793PCaaC42:1)+(0.543SMC16:1)+(0.573SMOHC22:2)+
(0.41 3 lysoPC a C17:0). Factor 2 = (0.55 3 propionylcarnitine) +
(0.663phenylalanine)+(0.613tryptophan)+(0.663tyrosine)+ DISCUSSION
(0.683valine)+(0.663isoleucine)+(0.593PCaaC32:1)+(0.703
PCaaC36:1)+(0.653PCaaC36:3)+(0.763PCaaC38:3)+(0.723 In this prospective investigation using a targeted meta-
PCaaC40:4)+(0.713PCaaC40:5)+(0.443hexose). bolomicapproachatpopulationlevel,wefoundincreased
concentrations of hexose; phenylalanine; and diacyl-phos-
phatidylcholinesC32:1,C36:1,C38:3,andC40:5andreduced
We conducted several sensitivity analyses. In EPIC- concentrationsofglycine;sphingomyelinC16:1;acyl-alkyl-
Potsdam, a small proportion of the participants (14.3%) phosphatidylcholinesC34:3,C40:6,C42:5,C44:4,andC44:5;
had fasted. The proportion of incident T2D cases was and lysophosphatidylcholine C18:2 to be independently
equally distributed among fasting and nonfasting partic- predictiveofT2DinEPIC-Potsdam.Theresultsagreewith
ipants (26.8% and 26.7%, respectively). Additional adjust- datafromcross-sectionalstudiesshowingthatpatientswith
mentforfastingstatusdidnotchangetheresults.Further, T2D had increased concentrations of sugar metabolites
wedidnotobserveaneffectmodification offastingstatus (13),acylcarnitines(14),andBCAA(13)andreducedcon-
ontheassociationbetweenmetabolitefactors1and2and centrationsofglycine(12).Wewereabletofurtherreplicate
T2Drisk(Pforinteraction=0.115and0.688,respectively). the results of Wang et al. (35), who recently reported that
Weobservednointeractionwithsex(P=0.407and0.441, BCAAandaromaticaminoacidspredictedT2Dinthepro-
respectively), and in both men and women, the risk asso- spectiveFraminghamOffspringcohortandtheMalmöDiet
ciations were similar. Hazard rates of T2D in different and Cancer study. In agreement with Wang et al. (35), we
periodsoffollow-upwerenotdifferent(factor1P=0.126, found higher concentrations of phenylalanine, isoleucine,
factor 2 P = 0.994), indicating that follow-up time did not tyrosine,andvalinetobeassociatedwithincreasedriskof
affect the association between metabolite factors and risk T2DandglycinetobeassociatedwithreducedriskofT2D.
of T2D. To ensure that the metabolite changes preceded However,in thepresentstudy, BCAA and aromaticamino
the onset of T2D and were not attributed to prediabetic acidswerelinkedtoeachother,andonlyphenylalaninewas
conditions,werepeatedtheanalysis,excludingallcasesof independently associated with T2D risk when accounting
T2D that occurred shortly after the baseline examination for the other metabolites. BCAAs may serve as substrates
644 DIABETES,VOL.62,FEBRUARY2013 diabetes.diabetesjournals.org
A.FLOEGELANDASSOCIATES
FIG.2.RelativecontributionofmetabolitestopredictT2DinEPIC-Potsdam.PresentedareROCcurvescomparingdifferentmultivariable-ad-
justedmodelstopredictT2D,includingtheDRS,theidentifiedmetabolites,glucose(Glc),andHbA .TheDRS(16)combinesinformationon
1c
severaldiabetesriskfactors,suchasdiet,lifestyle,andanthropometry,toestimateriskofdevelopingT2D.TheDRSiscomputedaccordingtothe
followingformula:DRS=(7.43waistcircumference[cm])2(2.43height[cm])+(4.33age[years])+(463hypertension[self-report])+(493
redmeat[each150g/day])–(93whole-grainbread[each50g/day])–(43coffee[each150g/day])–(203moderatealcohol[between10and40g/
day])–(23physicalactivity[h/week])+(243formersmoker)+(643currentheavysmoker[‡20cigarettes/day]).Metabolitesarehexose;
phenylalanine;glycine;sphingomyelinC16:1;diacyl-phosphatidylcholinesC32:1,C36:1,C38:3,andC40:5;acyl-alkyl-phosphatidylcholinesC34:3,
C40:6,C42:5,C44:4,andC44:5;andlysophosphatidylcholineC18:2.
for the glucose-alanine cycle in skeletal muscle. Through presentstudy,maybeinvolvedinpathwaystocompensate
alanine aminotransferase–catalyzed transamination reac- early stages of insulin resistance through stimulation of
tions,thismayresultinincreasedsubstrateavailabilityfor insulin secretion. In contrast, increased hexose concen-
hepatic gluconeogenesis, thereby increasing hepatic glu- trations may indicate manifest insulin resistance and de-
cose production (39). Conversely, glycine is a gluconeo- fect of b-cells.
genic amino acid; therefore, reduced serum glycine may Itisnoteworthythatweobservedsignificantassociations
alsoreflectincreasedgluconeogenesis.Alternativetheories between choline-containing phospholipids (i.e., diacyl-,
suggestthatglycinedepletionmayreflectglutathionecon- acyl-alkyl-, and lysophosphatidylcholines and sphingo-
sumption driven by oxidative stress (40) or abundance of myelins)and T2Drisk. Diacyl-phosphatidylcholines consist
incompletely oxidized fuels that are excreted as urinary of glycerol linked to phosphocholine and two fatty acid
acylglycine conjugates (41–43). The frequently observed residues, and removal of one fatty acid produces lyso-
increaseofBCAAinsubjectswithinsulinresistanceisalso phosphatidylcholines. The corresponding acyl-alkyl-phos-
believedtobetheresultofreducedactivitiesofkeyBCAA phatidylcholines comprise an ether linkage to one alkyl
catabolic enzymes in liver and adipose tissue (44). Fur- chain and one polyunsaturated fatty acid (47). Sphingo-
thermore,aminoacidsmaydirectlycausemuscularinsulin myelins are built of a ceramide core linked to one fatty
resistancebydisruptinginsulinsignaling(45). acid and a phosphocholine or phosphoethanolamine (Fig.
The positive association between hexose and T2D risk 3). Together, these phospholipids make up the main con-
remained significant after adjustment for glucose. This stituent of cellular membranes and may be involved in
observation could be an artifact from the different meth- cellular signal transduction (48). In addition, they repre-
ods used to measure hexose and glucose. However, it has sent a major fraction of the human plasma lipidome be-
to be noted that hexose represented not only glucose but cause they are most abundant in all lipoproteins (49).
also the sum of all six-carbon monosaccharides. Previous Diacyl-phosphatidylcholines are particularly essential for
studies have shown that in addition to glucose, fructose hepatic secretion of triglyceride-rich VLDL particles and
levels were elevated in individuals with T2D (12) and that HDL (48), whereas acyl-alkyl-phosphatidylcholines may
intake of fructose was positively associated with risk of actasserumantioxidantstopreventlipoproteinoxidation
insulin resistance and T2D (46). In insulin-resistant con- (50). Their hepatic synthesis requires dietary choline (48).
ditions, the body aims to compensate for decreased glu- It was previously shown that choline-deficient mice on
cose uptake of peripheral tissues through increased a high-fat diet showed reduced phosphatidylcholine bio-
pancreatic insulin secretion (1). However, at the stage of synthesis and accumulated hepatic fat, but at the same
overt insulin resistance, this system will eventually be time, they had reduced fasting insulin and improved glu-
exhausted, as caused by b-cell dysfunction and, sub- cose tolerance (51). In addition, impaired hepatic phos-
sequently, insulin secretion decreases (1). Phenylalanine, phatidylcholine biosynthesis led to reduced levels of
which was positively correlated to insulin secretion in the plasma triglycerides and HDL cholesterol in vivo (52).
diabetes.diabetesjournals.org DIABETES,VOL.62,FEBRUARY2013 645
SERUMMETABOLITESANDRISKOFTYPE2DIABETES
FIG.3.ExamplesofmetabolitesassociatedwithriskofT2D.▲Metaboliteswithanincreasedrisk(hexose,phenylalanine,anddiacyl-phospha-
tidylcholines[PCs]C32:1,C36:1,C38:3,andC40:5).*Metaboliteswithadecreasedrisk(glycine;sphingomyelinC16:1;acyl-alkyl-PCsC34:3,C40:6,
C42:5,C44:4,andC44:5;andlysophosphatidylcholineC18:2).Notethatthemassspectrometricassayuseddoesnotdistinguishmolecularlipids
andsugartypes amonghexoses.Therefore,formulasare givenfor amolecule corresponding tomolecularmassandcomposition. Positionsof
doublebondsandchainlengthmayvaryifmorethanoneacidresidueispresent.Arrowsrepresentmanyreactions,andkeyintermediatesare
giveninthebrackets.aa,diacyl;ae,acyl-alkyl.(Ahigh-qualitycolorrepresentationofthisfigureisavailableintheonlineissue.)
Accordingly, phosphatidylcholines and sphingomyelins In summary, the present data suggest that the identified
were positively related to plasma HDL cholesterol in metabolitescouldbepartofdifferentpathwaysinvolvedin
the present study. Furthermore, acyl-alkyl-phosphatidyl- theearlygenesisofT2D.Therefore,thesenovelcandidates
cholines were inversely correlated to plasma triglyc- could be useful in clinical practice to identify high-risk
erides,oppositetodiacyl-phosphatidylcholines,andhigher individuals earlier in order to delay or prevent disease
levels of acyl-alkyl-phosphatidylcholines but not diacyl- onset. Furthermore, serum metabolites in this study pre-
phosphatidylcholines were linked to improved insulin dictedriskofT2Dinasimilarmannertoacombinationof
sensitivity and reduced insulin secretion. Previous studies classic risk factors; thus, their measurement may be
reported that acyl-alkyl-phosphatidylcholine levels were ausefulapproachtopredictT2Driskforindividualsinthe
lower in obese subjects and subjects with insulin re- future. The metabolites could also serve as markers for
sistance (50,53). Thesemechanisms maycontribute tothe specific metabolic pathways that are deranged and,
antithetical association between two phosphatidylcholine thereby, allow the implementation of individualized pre-
subclasses and T2D risk found in the present study and ventive and therapeutic strategies. However, future
may indicate a key role of the type of linkage between investigations are warranted to calculate in detail the in-
phospholipid core and fatty acid residue. Furthermore, dividual risks and to better understand the metabolic
those phosphatidylcholines containing fatty acids with effects of these biomarkers and their biological mecha-
a lower number of carbons and double bonds were posi- nisms.
tively associated with T2D risk, contrary to those with The primary strength of this study is that, to our
a higher number of carbons and double bonds. Similar knowledge, we were among the first to adopt a targeted
observations have recently been reported for fatty acid metabolomic approach at population level and included
compositionsoferythrocytemembranephospholipids(54) a large sample from three independent, well-described
andtriglycerides(55),suggestingthatlipidswithashorter studypopulations.Furthermore,ourtargetedmetabolomic
chain length and saturated fatty acid residues may trigger platform covered a wide variety of metabolites with
development of T2D, whereas those containing longer known identity and quantitative measurements. Because
chains and unsaturated fatty acids may offer protection. we used a prospective design with consecutive follow-up,
646 DIABETES,VOL.62,FEBRUARY2013 diabetes.diabetesjournals.org
A.FLOEGELANDASSOCIATES
we were able to investigate time-dependent exposure- M.H.A.,A.P.,M.R.,C.P.,R.W.-S.,T.I.,M.B.S.,J.A.,H.B.,and
disease associations. T.P. provided their expertise and contributed to the
The study, however, had several limitations. First, be- writing of the manuscript. A.Fr., A.P., T.I., and H.B.
cause we used independent study populations, the con- collectedthedata.C.P.andJ.A.conductedandsupervised
ditions of biosample collection, storage, and preparation the metabolomic measurements. All authors take full
were not necessarily the same, which may be a source of responsibility for the contents of the manuscript. A.Fl. is
variation. In the KORA and TüF studies, fasting blood theguarantor ofthis work and,assuch,hadfullaccess to
samples were collected from all participants, whereas in all the data in the study and takes responsibility for the
EPIC-Potsdam only a small proportion of participants integrityofthedataandtheaccuracy ofthedataanalysis.
provided fasting blood samples. Nevertheless, we did not The authors thank all the study participants of EPIC-
observe an effect modification of fasting status on the as- Potsdam, KORA, and TüF. Special thanks go to Sven
sociationbetweenmetabolitefactorsandT2Drisk,andwe Knüppel and Wolfgang Bernigau (Department of Epidemi-
couldreproduceverysimilarmetabolitefactorscomparing ology, German Institute of Human Nutrition Potsdam-
fasting to nonfasting samples. Furthermore, the metab- Rehbruecke) for statistical advice and to Martin Floegel
olomic analyses were based on serum samples in the (Max Born Institute for Non-Linear Optics, Berlin) for his
EPIC-Potsdam and KORA studies but on plasma samples support in figure formatting.
in the TüF study. As previously reported (56), the corre-
lation between these serum and plasma metabolites was
high; however, the absolute metabolite concentrations REFERENCES
were higher in serum, which could lead to systematic 1. DeFronzoRA.BantingLecture.Fromthetriumviratetotheominousoctet:
changes. Second, the analytical method detected most a new paradigm for the treatment of type 2 diabetes mellitus. Diabetes
of the metabolites with high specificity; however, it may 2009;58:773–795
2. PetersAL,DavidsonMB,SchrigerDL,HasselbladV;Meta-analysisRe-
not have detected all possible interferences among me-
searchGroupontheDiagnosisofDiabetesUsingGlycatedHemoglobin
tabolites. Of the metabolites that we identified, diacyl- Levels. A clinical approach for the diagnosis of diabetes mellitus: an
phosphatidylcholine C38:3 and sphingomyelin C16:1 may analysis using glycosylated hemoglobin levels. JAMA 1996;276:1246–
be interfering compounds for sphingomyelin C24:1 and 1252
diacyl-phosphatidylcholine C30:2, respectively. Third, we 3. HerderC,BaumertJ,ZiererA,etal.Immunologicalandcardiometabolic
risk factors in the prediction of type 2 diabetes and coronary events:
only had a limited number of incident T2D cases available
from KORA. We may not have had sufficient statistical MONICA/KORAAugsburgcase-cohortstudy.PLoSONE2011;6:e19852
4. Abu-Qamar M, Wilson A. Evidence-based decision-making: the case for
power, and the replication results have to be interpreted diabetescare.IntJEvidBasedHealthc2007;5:254–260
with caution. Fourth, there is a chance that reverse causa- 5. Swellam M, Sayed Mahmoud And M, Abdel-Fatah Ali A. Clinical im-
tion may explain the results, implying that overt diabetic plicationsofadiponectinandinflammatorybiomarkersintype2diabetes
conditions that were undiagnosed may have caused these mellitus.DisMarkers2009;27:269–278
6. LiS,ShinHJ,DingEL,vanDamRM.Adiponectinlevelsandriskoftype2
metabolite changes. When we accounted for this issue, the
diabetes:asystematicreviewandmeta-analysis.JAMA2009;302:179–188
results remained robust. Last, because this was an obser-
7. FordES,SchulzeMB,BergmannMM,ThamerC,JoostHG,BoeingH.Liver
vational study, we cannot prove causality but only show enzymes and incident diabetes: findings from the European Prospective
associations.However,theidentifiedmetaboliteswerealso Investigation intoCancerandNutrition(EPIC)-PotsdamStudy.Diabetes
correlated to established T2D biomarkers as well as to Care2008;31:1138–1143
measures of insulin sensitivity and secretion in a different 8. StefanN,FritscheA,WeikertC,etal.Plasmafetuin-Alevelsandtheriskof
type2diabetes.Diabetes2008;57:2762–2767
population, which underlines the biological plausibility of
9. BainJR,StevensRD,Wenner BR,IlkayevaO,MuoioDM,NewgardCB.
the results.
Metabolomics applied todiabetes research: moving from information to
Inconclusion,thisprospectiveinvestigationusingmeta- knowledge.Diabetes2009;58:2429–2443
bolomicsdataofindependent studypopulations identified 10. Newgard CB, An J, Bain JR, et al. A branched-chain amino acid-related
sugar metabolites, amino acids, and choline-containing metabolic signature that differentiates obese and lean humans and con-
phospholipids to be independently associated with risk of tributestoinsulinresistance.CellMetab2009;9:311–326
11. WopereisS,RubinghCM,vanErkMJ,etal.Metabolicprofilingofthere-
T2D. Beyond the classic pathways, these candidates point
sponsetoanoralglucosetolerancetestdetectssubtlemetabolicchanges.
toward a novel role of phospholipid and lipoprotein me-
PLoSONE2009;4:e4525
tabolism in T2D pathophysiology. Future studies should 12. FiehnO,GarveyWT,NewmanJW,LokKH,HoppelCL,AdamsSH.Plasma
furtherelucidatethebiologicalmechanisms. metabolomicprofilesreflectiveofglucosehomeostasisinnon-diabeticand
type2diabeticobeseAfrican-Americanwomen.PLoSONE2010;5:e15234
13. Suhre K, Meisinger C, Döring A, et al. Metabolic footprint of diabetes:
ACKNOWLEDGMENTS a multiplatform metabolomics study inan epidemiological setting. PLoS
ONE2010;5:e13953
The current study was supported by a grant from the 14. AdamsSH,HoppelCL,LokKH,etal.Plasmaacylcarnitineprofilessuggest
Federal Ministry of Education and Research, Germany incompletelong-chainfattyacidbeta-oxidationandalteredtricarboxylic
(Bundesministerium für Bildung und Forschung), to the acidcycleactivityintype2diabeticAfrican-Americanwomen.JNutr2009;
GermanCenter for DiabetesResearch (Förderkennzeichen 139:1073–1081
01GI0922toD.Z.D.).N.S.issupportedbyaHeisenbergpro- 15. Hu FB. Metabolic profiling of diabetes: from black-box epidemiology to
systemsepidemiology.ClinChem2011;57:1224–1226
fessorship from the Deutsche Forschungsgemeinschaft.
16. SchulzeMB,HoffmannK,BoeingH,etal.Anaccurateriskscorebasedon
No potential conflicts of interest relevant to this article
anthropometric,dietary,andlifestylefactorstopredictthedevelopmentof
were reported. type2diabetes.DiabetesCare2007;30:510–515
A.Fl.,H.B., andT.P.designedthestudy.A.Fl.,N.S., Z.Y., 17. BoeingH,KorfmannA,BergmannMM.RecruitmentproceduresofEPIC-
and K.M. analyzed the data. A.Fl., N.S., Z.Y., K.M., D.D., Germany. European Investigation into Cancer and Nutrition. Ann Nutr
H.-G.J.,A.Fr.,H.-U.H.,M.H.A.,A.P.,M.R.,C.P.,R.W.-S.,T.I., Metab1999;43:205–215
18. KrokeA,BergmannMM,LotzeG,JeckelA,Klipstein-GrobuschK,Boeing
M.B.S., J.A., H.B., and T.P. discussed the data and
H. Measures of quality control in the German component of the EPIC
interpreted the results. A.Fl. wrote the first draft of the
study.EuropeanProspectiveInvestigationintoCancerandNutrition.Ann
manuscript. N.S., Z.Y., K.M., D.D., H.-G.J., A.Fr., H.-U.H., NutrMetab1999;43:216–224
diabetes.diabetesjournals.org DIABETES,VOL.62,FEBRUARY2013 647
SERUMMETABOLITESANDRISKOFTYPE2DIABETES
19. BoeingH,WahrendorfJ,BeckerN.EPIC-Germany—asourceforstudies 38. Hsieh FY, Lavori PW. Sample-size calculations for the Cox proportional
intodietandriskofchronicdiseases.EuropeanInvestigationintoCancer hazardsregressionmodelwithnonbinarycovariates.ControlClin Trials
andNutrition.AnnNutrMetab1999;43:195–204 2000;21:552–560
20. WeikertC,StefanN,SchulzeMB,etal.Plasmafetuin-Alevelsandtheriskof 39. RudermanNB.Muscleaminoacidmetabolismandgluconeogenesis.Annu
myocardialinfarctionandischemicstroke.Circulation2008;118:2555–2562 RevMed1975;26:245–258
21. MontonenJ,DroganD,JoostHG,etal.Estimationofthecontributionof 40. SekharRV,McKaySV,PatelSG,etal.Glutathionesynthesisisdiminished
biomarkersofdifferentmetabolicpathwaystoriskoftype2diabetes.Eur in patients with uncontrolled diabetes and restored by dietary supple-
JEpidemiol2011;26:29–38 mentationwithcysteineandglycine.DiabetesCare2011;34:162–167
22. Bergmann MM, Bussas U, Boeing H. Follow-up procedures in EPIC- 41. Koves TR, Ussher JR, Noland RC, et al. Mitochondrial overload and in-
Germany—dataqualityaspects.EuropeanProspectiveInvestigationinto complete fatty acid oxidation contribute to skeletal muscle insulin re-
CancerandNutrition.AnnNutrMetab1999;43:225–234 sistance.CellMetab2008;7:45–56
23. SchienkiewitzA,SchulzeMB,HoffmannK,KrokeA,BoeingH.Bodymass 42. Muoio DM, Neufer PD. Lipid-induced mitochondrial stress and insulin
indexhistoryandriskoftype2diabetes:resultsfromtheEuropeanPro- actioninmuscle.CellMetab2012;15:595–605
spective Investigation into Cancer and Nutrition (EPIC)-Potsdam Study. 43. Vianey-Liaud C, Divry P, Gregersen N, Mathieu M. The inborn errors of
AmJClinNutr2006;84:427–433 mitochondrialfattyacidoxidation.JInheritMetabDis1987;10(Suppl1):
24. RothmanKJ,GreenlandS,LashTL.Case-controlstudies.InModernepi- 159–200
demiology.3rded.RothmanKJ,GreenlandS,Eds.Philadelphia,Lippin- 44. SheP,VanHornC,ReidT,HutsonSM,CooneyRN,LynchCJ.Obesity-
cott-Williams&Wilkins,2008,p.111–127 related elevations in plasma leucine are associated with alterations in
25. Meisinger C, Strassburger K, Heier M, et al. Prevalence of undiagnosed enzymesinvolvedinbranched-chainaminoacidmetabolism.AmJPhysiol
diabetesandimpairedglucoseregulationin35-59-year-oldindividualsin EndocrinolMetab2007;293:E1552–E1563
SouthernGermany:theKORAF4Study.DiabetMed2010;27:360–362 45. TremblayF,BrûléS,HeeUmS,etal.IdentificationofIRS-1Ser-1101as
26. RathmannW,StrassburgerK,HeierM,etal.Incidenceoftype2diabetesin atargetofS6K1innutrient-andobesity-inducedinsulinresistance.Proc
theelderlyGermanpopulationandtheeffectofclinicalandlifestylerisk NatlAcadSciUSA2007;104:14056–14061
factors:KORAS4/F4cohortstudy.DiabetMed2009;26:1212–1219 46. MontonenJ,JärvinenR,KnektP,HeliövaaraM,ReunanenA.Consump-
27. KowallB,RathmannW,StrassburgerK,MeisingerC,HolleR,MielckA. tionofsweetenedbeveragesandintakesoffructoseandglucosepredict
Socioeconomicstatusisnotassociatedwithtype2diabetesincidencein type2diabetesoccurrence.JNutr2007;137:1447–1454
anelderlypopulationinGermany:KORAS4/F4cohortstudy.JEpidemiol 47. MagnussonCD,HaraldssonGG.Etherlipids.ChemPhysLipids2011;164:
CommunityHealth2011;65:606–612 315–340
28. StefanN,KantartzisK,MachannJ,etal.Identificationandcharacterization 48. Cole LK, Vance JE, Vance DE. Phosphatidylcholine biosynthesis and li-
of metabolically benign obesity in humans. Arch Intern Med 2008;168: poproteinmetabolism.BiochimBiophysActa2012;182:754–761
1609–1616 49. QuehenbergerO,DennisEA.Thehumanplasmalipidome.NEnglJMed
29. Matsuda M, DeFronzo RA. Insulin sensitivity indices obtained from oral 2011;365:1812–1823
glucosetolerancetesting:comparisonwiththeeuglycemicinsulinclamp. 50. Wallner S, Schmitz G. Plasmalogens the neglected regulatory and scav-
DiabetesCare1999;22:1462–1470 enginglipidspecies.ChemPhysLipids2011;164:573–589
30. Römisch-MarglW,PrehnC,BogumilR,RöhringC,SuhreK,AdamskiJ. 51. Raubenheimer PJ, Nyirenda MJ, Walker BR. A choline-deficient diet ex-
Procedure for tissue sample preparation and metabolite extraction for acerbates fatty liver but attenuates insulin resistance and glucose in-
high throughput targeted metabolomics. Metabolomics 11 March 2011 toleranceinmicefedahigh-fatdiet.Diabetes2006;55:2015–2020
[Epubaheadofprint] 52. JacobsRL,DevlinC,TabasI,VanceDE.TargeteddeletionofhepaticCTP:
31. FloegelA,DroganD,Wang-SattlerR,etal.Reliabilityofserummetabolite phosphocholinecytidylyltransferasealphainmicedecreasesplasmahigh
concentrations over a 4-month period using a targeted metabolomic ap- density and very low density lipoproteins. J Biol Chem 2004;279:47402–
proach.PLoSONE2011;6:e21103 47410
32. PrenticeRL.Designissuesincohortstudies.StatMethodsMedRes1995;4: 53. PietiläinenKH,Sysi-AhoM,RissanenA,etal.Acquiredobesityisassoci-
273–292 atedwithchangesintheserumlipidomicprofileindependentofgenetic
33. HolmS.Asimplesequentiallyrejectivemultipletestprocedure.ScandJ effects—amonozygotictwinstudy.PLoSONE2007;2:e218
Stat1979;6:65–70 54. KrögerJ,ZietemannV,EnzenbachC,etal.Erythrocytemembranephos-
34. HardyRJ,ThompsonSG.Detectinganddescribingheterogeneityinmeta- pholipidfattyacids,desaturaseactivity,anddietaryfattyacidsinrelation
analysis.StatMed1998;17:841–856 toriskoftype2diabetesintheEuropeanProspectiveInvestigationinto
35. WangTJ,LarsonMG,VasanRS,etal.Metaboliteprofilesandtheriskof CancerandNutrition(EPIC)-PotsdamStudy.AmJClinNutr2011;93:127–
developingdiabetes.NatMed2011;17:448–453 142
36. DeLongER,DeLongDM,Clarke-PearsonDL.Comparingtheareasunder 55. Rhee EP, Cheng S, Larson MG, et al. Lipid profiling identifies a tri-
two or more correlated receiver operating characteristic curves: a non- acylglycerol signature of insulin resistance and improves diabetes pre-
parametricapproach.Biometrics1988;44:837–845 dictioninhumans.JClinInvest2011;121:1402–1411
37. HigginsJP,ThompsonSG,DeeksJJ,AltmanDG.Measuringinconsistency 56. YuZ,KastenmüllerG,HeY,etal.Differencesbetweenhumanplasmaand
inmeta-analyses.BMJ2003;327:557–560 serummetaboliteprofiles.PLoSONE2011;6:e21230
648 DIABETES,VOL.62,FEBRUARY2013 diabetes.diabetesjournals.org