Effects ofMetformin onMetaboliteProfiles and LDL Cholesterol inPatients With Type 2 DiabetesDiabetes Care 2015;38:1858–1867 | DOI: 10.2337/dc15-0658

OBJECTIVE

Metformin is used as a first-line oral treatment for type 2 diabetes (T2D). How-ever, the underlying mechanism is not fully understood. Here, we aimed to com-prehensively investigate the pleiotropic effects of metformin.

RESEARCH DESIGN AND METHODS

We analyzed both metabolomic and genomic data of the population-based KORAcohort. To evaluate the effect of metformin treatment on metabolite concentra-tions, we quantified 131metabolites in fasting serum samples and usedmultivari-able linear regressionmodels in three independent cross-sectional studies (n = 151patients with T2D treated with metformin [mt-T2D]). Additionally, we used linearmixed-effect models to study the longitudinal KORA samples (n = 912) and per-formed mediation analyses to investigate the effects of metformin intake onblood lipid profiles. We combined genotyping data with the identified metformin-associated metabolites in KORA individuals (n = 1,809) and explored the underlyingpathways.

RESULTS

We found significantly lower (P < 5.0E-06) concentrations of three metabolites(acyl-alkyl phosphatidylcholines [PCs]) when comparing mt-T2D with four controlgroups whowere not using glucose-lowering oral medication. These findings werecontrolled for conventional risk factors of T2D and replicated in two independentstudies. Furthermore, we observed that the levels of thesemetabolites decreasedsignificantly in patients after they started metformin treatment during 7 years’follow-up. The reduction of these metabolites was also associated with a loweredblood level of LDL cholesterol (LDL-C). Variations of these three metabolites weresignificantly associated with 17 genes (including FADS1 and FADS2) and controlledby AMPK, a metformin target.

CONCLUSIONS

Our results indicate that metformin intake activates AMPK and consequentlysuppresses FADS, which leads to reduced levels of the three acyl-alkyl PCs andLDL-C. Our findings suggest potential beneficial effects of metformin in the pre-vention of cardiovascular disease.

Type 2 diabetes (T2D) is a chronic disease with diminished response to insulin andrelative insulin deficiency (1). Patients with T2D mostly take metformin as first-lineoral treatment to lower their glucose levels and to improve insulin sensitivity (2).Despite metformin’s use as an antihyperglycemic agent for more than 50 years, its

1Research Unit of Molecular Epidemiology,Helmholtz Zentrum München, Neuherberg,Germany2Institute of Epidemiology II, Helmholtz ZentrumMünchen, Neuherberg, Germany3Institute of Structural Biology, Helmholtz Zen-trum München, Neuherberg, Germany4Institute for Clinical Diabetology, German Dia-betes Center, Leibniz Center for Diabetes Re-search at Heinrich Heine University, Düsseldorf,Germany5German Center for Diabetes Research,Düsseldorf, Germany6Department of Biological Psychology, Faculty ofPsychology and Education, VU University Am-sterdam, Amsterdam, the Netherlands7EMGO Institute for Health and Care Research,Amsterdam, the Netherlands8Neuroscience Campus Amsterdam, Amsterdam,the Netherlands9Department of Epidemiology, Erasmus MedicalCenter, Rotterdam, the Netherlands10Department of Human Genetics, LeidenUniversity Medical Center, Leiden, theNetherlands11Institute of Genetic Epidemiology, HelmholtzZentrum München, Neuherberg, Germany

Tao Xu,1,2 Stefan Brandmaier,1,2

Ana C. Messias,3 Christian Herder,4,5

Harmen H.M. Draisma,6,7,8

Ayse Demirkan,9,10 Zhonghao Yu,1,2

Janina S. Ried,11 Toomas Haller,12

Margit Heier,2 Monica Campillos,13

Gisela Fobo,13 Renee Stark,14

Christina Holzapfel,15 Jonathan Adam,1,2

Shen Chi,1,2 Markus Rotter,1,2

Tommaso Panni,2 Anne S. Quante,11,16

Ying He,17,18 Cornelia Prehn,19

Werner Roemisch-Margl,13

Gabi Kastenmüller,13

Gonneke Willemsen,6,7 René Pool,6,7

Katarina Kasa,9 Ko Willems van Dijk,10

Thomas Hankemeier,20

Christa Meisinger,2 Barbara Thorand,2

Andreas Ruepp,13

Martin Hrabé de Angelis,21,22,23

Yixue Li,17,18 H.-Erich Wichmann,2,16,24

Bernd Stratmann,25

Konstantin Strauch,11,16

Andres Metspalu,12 Christian Gieger,1,2

Karsten Suhre,13,26,27

Jerzy Adamski,19,22,23 Thomas Illig,1,28,29

Wolfgang Rathmann,30

Michael Roden,4,5,31 Annette Peters,1,2,23,32

Cornelia M. van Duijn,8,33

Dorret I. Boomsma,6,7

Thomas Meitinger,34,35 and

Rui Wang-Sattler1,2,23

1858 Diabetes Care Volume 38, October 2015

EPIDEM

IOLO

GY/HEA

LTHSERVICES

RESEA

RCH

http://crossmark.crossref.org/dialog/?doi=10.2337/dc15-0658&domain=pdf&date_stamp=2015-09-11

primary mode of action is not yet com-pletely understood (3). Inside a cell,metformin apparently inhibits complexI of themitochondrial electron transportchain and thereby reduces the cellularenergy status and upregulates the cyto-plasmic 59-AMPK pathway (3). ActivatedAMPK stimulates catabolic processes(glycolysis and fatty acid oxidation) andinhibits anabolic pathways (gluconeo-genesis and fatty acid synthesis). Sofar, six metformin targets are docu-mented in the DrugBank (4) database,including the AMPK complex and fivemetformin transporters. Furthermore,metformin was reported to have severalpossible pleiotropic effects, resulting inreduced risks for both cancer (5) andcardiovascular disease (CVD) (6), aswell as reduced levels of LDL cholesterol(LDL-C) (7,8).Metabolomic studies have detected

metabolite profile changes during thedevelopment of T2D (9–12) and identi-fied concentration differences causedby various physiological and environ-mental factors such as age (13), sex(14), smoking status (15), and alcoholconsumption (16). Several metabolomicstudies attempted to unravel the physi-ological effects of metformin (17–21).However, they either used technologiescovering only small sets of metabolitesor examined relatively few participants(e.g., 20 healthy volunteers [18], 15 pa-tients [17,19], 31 patients [20], and 24

patients treated with glipizide and 23patients with metformin [21]). As inter-individual genetic variations contributeto diversemetabolite profiles and differentdrug responses, combining metabolomicsand genomics may help to understand themechanisms underlying the action of med-ications (22–25).

In this study, we discovered metfor-min treatment–associated metabolitesin the Cooperative Health Research inthe Region of Augsburg (KORA) cohort(26,27). We confirmed our finding inlongitudinal KORA data and replicatedthem in two independent studies: theErasmus Rucphen Family study (ERF)(28) and the Netherlands Twin Register(NTR) (29). The biologically relevantpathways for the identified metabolitesand their associated genes were fur-ther analyzed in organ-specific protein-metabolite interaction networks (30,31).Additionally, we assessed the effects ofmetformin treatment on LDL-C levels.

RESEARCH DESIGN AND METHODS

An overview of the analysis work flow isshown in Fig. 1.

Ethics StatementAll participants gave written informedconsent. The KORA study was approvedby the ethics committee of the BavarianMedical Association, Germany; the ERFstudy by the medical ethics board of theErasmusMCRotterdam, theNetherlands;

and the NTR study by the Central EthicsCommittee on Research Involving HumanSubjects of the VU University MedicalCenter, Amsterdam, the Netherlands.

KORA CohortKORA is a population-based cohortstudy conducted in Southern Germany(26). The baseline survey 4 (KORA S4)consists of 4,261 individuals (aged 25–74 years) examined between 1999 and2001. During the years 2006–2008,3,080 participants took part in thefollow-up survey 4 (KORA F4). Clinicaldata for each participant were retrievedfrommedical records. Based onphysician-validated and self-reported diagnosis(9,26), fasting glucose and2-h postglucoseload, and information onmedications (Ta-ble 1), we excluded 1) patients suffer-ing from type 1 and steroid-induceddiabetes (n = 9), 2) patients with T2Dtreated with both metformin and in-sulin (n = 15), 3) patients taking glucose-lowering oral medication other thanmetformin (n = 25), and 4) patients lack-ing clear informationon treatment (n = 1).Furthermore, participants with over-night nonfasting blood samples (n = 16)or isolated impaired fasting glucose(n = 112) were excluded. We previouslyshowed that impaired fasting glucose andimpaired glucose tolerance (IGT) should beconsidered two different phenotypes (9).In KORA F4, we focused on five groups:1) patients with metformin-treated T2D

12Estonian Genome Center, University of Tartu,Tartu, Estonia13Institute of Bioinformatics and Systems Biology,Helmholtz ZentrumMünchen,Neuherberg,Germany14Institute of Health Economics and Health CareManagement, Helmholtz Zentrum München,Neuherberg, Germany15Else Kroener-Fresenius-Center for NutritionalMedicine, Faculty of Medicine, Technische Uni-versität München, Munich, Germany16Institute of Medical Informatics, Biometry andEpidemiology, Chair of Genetic Epidemiology,Ludwig-Maximilians-Universität,Munich, Germany17Shanghai Center for Bioinformation Technol-ogy, Shanghai, China18Bioinformatics Center, Key Laboratory of SystemsBiology, Shanghai Institutes for Biological Sciences,Chinese Academy of Sciences, Shanghai, China19Institute of Experimental Genetics, GenomeAnalysis Center, Helmholtz Zentrum München,Neuherberg, Germany20Faculty of Science, Leiden Academic Centrefor Drug Research, Analytical BioSciences, theNetherlands21Institute of Experimental Genetics, HelmholtzZentrum München, Neuherberg, Germany

22Chair of Experimental Genetics, Center of Lifeand Food Sciences Weihenstephan, TechnischeUniversität München, Freising, Germany23German Center for Diabetes Research, Neuher-berg, Germany24Institute of Medical Statistics and Epidemiology,TechnischeUniversitätMünchen,Munich,Germany25Heart and Diabetes Center NRW, Diabetes Cen-ter, Ruhr-University Bochum, Bad Oeynhausen,Germany26Faculty of Biology, Ludwig-Maximilians-Universität, Planegg-Martinsried, Germany27Department of Physiology and Biophysics,Weill CornellMedical College in Qatar, EducationCity, Qatar Foundation, Doha, Qatar28Hannover Unified Biobank, Hannover MedicalSchool, Hannover, Germany29Institute for Human Genetics, Hannover Med-ical School, Hanover, Germany30Institute for Biometrics and Epidemiology, Ger-manDiabetes Center, Leibniz Center for DiabetesResearch at Heinrich Heine University,Düsseldorf, Germany31Department of Endocrinology and Diabetol-ogy, Medical Faculty, Düsseldorf, Germany

32Department of Environmental Health, HarvardSchool of Public Health, Boston, MA33Center for Medical Systems Biology, Leiden,the Netherlands34Institute of Human Genetics, Helmholtz Zen-trum München, Neuherberg, Germany35Institute of Human Genetics, Technische Uni-versität München, Munich, Germany

Corresponding author: Rui Wang-Sattler, rui.wang-sattler@helmholtz-muenchen.de.

Received 30 March 2015 and accepted 24 June2015.

This article contains Supplementary Data onlineat http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1.

T.X., S.B., A.C.M., C.He., H.H.M.D., and A.D. con-tributed equally. C.M.v.D. and D.I.B. contributedequally.

© 2015 by the American Diabetes Association.Readers may use this article as long as the workis properly cited, the use is educational and notfor profit, and the work is not altered.

care.diabetesjournals.org Xu and Associates 1859

mailto:rui.wang-sattler@helmholtz-muenchen.demailto:rui.wang-sattler@helmholtz-muenchen.dehttp://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org

(mt-T2D), 2) patients with T2D with in-sulin treatment (it-T2D), 3) patientswith T2D without glucose-lowering treat-ment (non–antidiabetes drug treated[ndt-T2D]), 4) participants with predia-betes with IGT, and 5) healthy individualswith normal glucose tolerance (NGT)(Table 1).

Replication StudiesThe ERF includes 3,000 living descen-dants of 22 couples who had at leastsix children baptized in the communitychurch around 1850–1900. The partici-pants are not selected based on any dis-ease or other outcome. Details about

the genealogy of the population havepreviously been provided (28).

The NTR recruits twins and their fam-ily members to study the causes of indi-vidual differences in health, behavior,and lifestyle. Participants are followedlongitudinally; details about the cohorthave previously been published (29). Asubsample of unselected twins and theirfamily members has taken part in theNTR-Biobank (32) in which biologicalsamples, including DNA and RNA, werecollected in a standardizedmanner afterovernight fasting.

Duration of diabetes and 2-h postglu-cose levels were not available in either

the ERF or NTR study. The diagnosis ofpatients with diabetes in both ERF andNTR studies was based on self-report.Owing to the limited number of it-T2Dpatients in these two replication studies(n = 3 and n = 9, respectively), this groupis not included in the statistical analysesin these two replication studies.

Initially, we had contacted a third po-tential replication study, the EstonianGenome Center of the University ofTartu (EGCUT). However, only two mt-T2D participants with available meta-bolomics data were available in thiscohort; results from the EGCUT studyare therefore not shown.

Figure 1—Flowchart of the study design.

1860 Metformin, Metabolic Profiles, and LDL-C Diabetes Care Volume 38, October 2015

Blood SamplingIn the KORA cohort study, blood wasdrawn into S-Monovette serum tubes(Sarstedt AG & Co., Nümbrecht, Ger-many) in the morning between 8:00 A.M.and 10:30 A.M. after at least 8 h of fasting.Tubes were gently inverted twice,followed by 30 min resting at room tem-perature to obtain complete coagulation.For serum collection, blood was centri-fuged at 2,750g at 158C for 10min. Serumwas filled into synthetic straws, whichwere stored in liquid nitrogen (21968C)until the metabolomics analyses (9,23).In the ERF and NTR, the overnight

fasting serum samples were drawnfor metabolite profiling. Details aboutthe sampling in these two cohorts weredescribed in previous publications(28,32).

Metabolomics MeasurementThe serum samples from participants inthe baseline KORA S4 and follow-upKORA F4 study were measured with theAbsoluteIDQp180 and AbsoluteIDQp150kits (Biocrates Life Sciences AG, Innsbruck,Austria), respectively. The assay procedures

were previously described in detail (27).For KORA S4 and F4, identical quality-control procedures (9,13), which areexplained in details in our previous pub-lications, were used. In KORA F4, 131metabolites of the initially targeted 163metabolites passed all quality-controlcriteria: hexose (H1), 24 acylcarnitines,14 amino acids, 13 sphingomyelines, 34phosphatidylcholines (PCs), diacyl (aa),37 PCs acyl-alkyl (ae), and 8 lysoPCs. Intotal, 124 metabolites overlapped be-tween KORA S4 and F4, including H1,21 acylcarnitines, 14 amino acids, 13sphingomyelines, 33 PC aas, and 34 PCaes, as well as 8 lysoPCs.

The metabolite measurements forboth replication studies (ERF and NTR)were performed using the same platform(AbsoluteIDQp150 kit) as in the KORA F4study. Additionally, in ERF, PC ae C36:4,PC ae C38:5, and PC ae C38:6 were mea-sured in the full set of serum samples by atargeted liquid chromatography–massspectrometry method. The measure-ment is performed on a UPLC-ESI-Q-TOF(Agilent 6530; Agilent Technologies, SanJose, CA) mass spectrometer using

reference mass correction. Chromato-graphic separation was achieved on anACQUITY UPLC HSS T3 column (1.8 mm,2.1 * 100 mm) with a flow of 0.4 mL/minover a 16-min gradient. The metaboliteswere detected in full scan in the positive-ion mode. The raw data were processedusing Agilent MassHunter QuantitativeAnalysis software (version B.04.00; AgilentTechnologies).

Measured concentration values of allanalyzed metabolites are reported inmicromolar (mM) and were natural-logtransformed, and the distributions weresubsequently standardized with meanof zero and an SD of 1 for all analysesunless otherwise indicated.

Single Nucleotide PolymorphismGenotyping, Imputation, and GenesIn KORA F4, we carried out genotypingusing the Affymetrix 6.0 GeneChip array(Affymetrix, Santa Clara, CA). Imputationwas performed with Impute (http://mathgen.stats.ox.ac.uk/impute/), version0.4.2 (referenceHapMap phase 2, release22). We only used autosomal single nu-cleotide polymorphisms (SNPs) with a

Table 1—Characteristics of the KORA F4 cross-sectional study population

Clinical parameters NGT IGT ndt-T2D mt-T2D it-T2D

n 2,129 375 169 90 24

Age, years 52.8 (12.6) 63.9 (11.0) 66.3 (9.7) 66.8 (8.7) 69.2 (9.8)

Male 46 49 62 59 54

BMI, kg/m2 26.6 (4.3) 29.7 (4.9) 30.8 (4.4) 31.7 (5.4) 32.2 (5.9)

Waist, cm 90.5 (12.9) 99.7 (14.3) 104.6 (11.4) 106.3 (1.27) 107.2 (12.4)

Physical activity, .1 h per week 58 50 47 33 17

High alcohol intake† 17 17 18 20 8

Smoker 21 8 12 13 8

Systolic BP, mmHg 119.1 (17.4) 127.5 (18.5) 133.7 (18.6) 131.3 (18.9) 135.6 (22.7)

HDL-C, mg/dL 57.6 (14.4) 54.1 (14.0) 47.8 (12.1) 50.6 (10.5) 48.0 (9.6)

LDL-C, mg/dL 134.9 (34.3) 143.7 (35.4) 138.5 (36.5) 122.9 (29.0) 120.0 (31.6)

Triglycerides, mg/dL 110.6 (73.0) 146.0 (86.2) 175.1 (127.0) 174.4 (132.2) 142.1 (73.2)

HbA1c, % 5.4 (0.3) 5.6 (0.3) 6.3 (0.9) 6.9 (1.1) 7.3 (1.1)

HbA1c, mmol/mol 36 (3.3) 38 (3.3) 45 (9.8) 52 (12.0) 56 (12.0)

Fasting glucose, mg/dL 91.7 (7.6) 100.1 (10.6) 125.7 (29.1) 144.1 (37.1) 141.9 (39.0)

2-h postglucose load, mg/dL 97.7 (20.8) 161.7 (17.1) 214.5 (50.7)U d dTime since diagnosis, years d d 1.0 (3.1)# 7.7 (7.1) 16.7 (7.4)

Insulin, mlU/mL 6.9 (25.9) 13.1 (64.0) 16.6 (30.1) 10.4 (10.4) 32.2 (77.8)

Statin usage 8 16 24 38 33

b-Blocker usage 12 31 43 41 63

ACE inhibitor usage 8 21 31 43 58

ARB usage 6 9 15 13 8

Metformin usage 0 0 0 100 0

Insulin therapy 0 0 0 0 100

Percentages of individuals or means (SD) are shown for each variable and each group (NGT, IGT, ndt-T2D, mt-T2D, and it-T2D). †$20 g/day forwomen; $40 g/day for men. Un = 121. #For newly diagnosed T2D patients (n = 112), years since T2D diagnosis was defined as 0.

care.diabetesjournals.org Xu and Associates 1861

http://mathgen.stats.ox.ac.uk/impute/http://mathgen.stats.ox.ac.uk/impute/http://care.diabetesjournals.org

minor allele frequency .5%, call rate.95%, and imputation quality .0.4. Forthe phenotype set enrichment analysis(PSEA), we only mapped those SNPs to agene that were either in its transcribedregion or in its flanking region (110 kbupstream, 40 kb downstream). Gene in-formation was downloaded from theUCSC (University of California, SantaCruz) genome browser (http://genome.ucsc.edu). The SNP gene mapping wasdescribed in detail previously (25). In to-tal, 20,801 genes were analyzed.

Statistical AnalysisTo evaluate the effect ofmetformin treat-ment on metabolites, we used multivari-able linear regression models with themetabolite concentration values as out-come and the grouping variable as pre-dictor. Each metabolite was assessedindividually. To include potential con-founders, we adjusted for two sets of co-variates: 1) age and sex as the crudemodel and 2) age, sex, BMI, physical ac-tivity, alcohol intake, smoking, systolicblood pressure (BP), levels of HDL choles-terol (HDL-C), triglycerides, HbA1c, andfasting glucose as well as the use of sta-tins, b-blockers, ACE inhibitors, and an-giotensin receptor blockers (ARBs) asthe full model (Table 1). To account formultiple testing, we used Bonferroni cor-rection and considered only thosemetab-olites with a P, 0.05/131 = 3.8E-04 to bestatistically significantly different in KORAF4.Meta-analysis of the three studieswasperformed using random effect models,using a restricted maximum-likelihoodestimator.In the KORA S4 to F4 longitudinal

study, we used linear mixed-effect mod-els. We adjusted for the two sets ofcovariates as described above whileassigning a random offset to each ofthe individual participant in the longitu-dinal study. Additionally, using linear re-gression models on the KORA data set,including two time points (S4 n = 1,335and F4 n = 2,763) (9), we calculated theresidues of the metabolite concentra-tions adjusted for age, sex, BMI, physicalactivity, alcohol intake, smoking, systolicBP, HDL-C, triglyceride, fasting glucose,and HbA1c. The significance of thechanges in the metabolite concentra-tions between the two time points (S4and F4) was tested using a linear mixed-effect model with the covariates at twotime points.

PSEA is a gene-based approach to an-alyze the associations of genome-wideSNP data with multiple phenotypesin a combined way (25). The significanceof enrichment was calculated based on10,000 permutations (limited by compu-tational restrictions), while setting thesignificance level at P, 1.0E-04 (lowestpossible P value owing to the permuta-tion number).

Mediation analysis (33) was con-ducted to model the identified metabo-lites as mediators for the associationbetween metformin treatment andLDL-C and total cholesterol in the longi-tudinal KORA data. The mediation ef-fects of each single metabolite and theirsummed concentration were tested withcrude and fully adjusted multivariablelinear regression models.

All statistical analyses were per-formed in R (version 3.0.1 [http://cran.r-project.org/]).

Pathway AnalysisWith use of a bioinformatical approach, anetwork was constructed by retrievingpairwise connections between candidatemetabolites, PSEA-identified genes, inter-mediate proteins, and known metformintarget genes (9,31). Information on protein-protein interactions was extracted fromSTITCH (30). Known metformin targetgenes were retrieved from the DrugBank(4). In our network, we only consideredthe shortest paths (allowing one interme-diate protein, confidence score .0.7)connecting the protein encoded by thegenes identified in PSEAwith themetformintarget genes.

RESULTS

Metabolite Profiles in Three CohortsWe quantified.130metabolites in fast-ing serum samples from the KORA S4and F4, ERF, and NTR studies (Fig. 1).The discovery study, KORA F4, includes2,129 NGT, 375 IGT, 169 ndt-T2D, 90mt-T2D, and 24 it-T2D subjects (charac-teristics shown in Table 1). In the longi-tudinal study, we used samples from912 participants without metformintreatment at baseline (KORA S4); 43 ofthem were treated with metformin atfollow-up (KORA F4 [Supplementary Ta-ble 1]). In reference to the two replica-tion cohorts, ERF contained 29 ndt-T2Dand 32 mt-T2D patients (characteristicsshown in Supplementary Table 2), whileNTR included 73 ndt-T2D and 29 mt-T2D

patients (characteristics shown in Sup-plementary Table 3).

In general, patients with T2D in thethree studies were older and more fre-quently men, with higher BMI, and tookmore nonantihyperglycemic medica-tions than the participants without di-abetes. Among the five groups in KORAF4, people on statin treatment had sig-nificantly lower LDL-C levels than thosewho were not taking statins (Sup-plementary Fig. 1). When comparingmt-T2D with ndt-T2D, lower levels ofLDL-C were observed both in the cross-sectional (KORA F4, ERF, and NTR) and inthe longitudinal KORA studies. Follow-ing the 43 patients, who started metfor-min treatment after the baseline, we didnot observe significant changes in thelevels of HbA1c and fasting glucose butobserved significant changes for LDL-Cand total cholesterol (SupplementaryTable 1).

Metabolites Associated WithMetformin TreatmentWe found six metabolites includingthree acyl-alkyl PCs, two diacyl (aa)PCs, and one amino acid to have signif-icantly lower concentrations in the 90mt-T2D patients compared with the169 ndt-T2D individuals in KORA F4 (Ta-ble 2). For example, for the metabolitePC ae C36:4, we observed that the fullyadjusted effect estimate was 20.66with P = 4.92E-07; i.e., the PC ae C36:4level in the mt-T2D group was 0.66 SDlower than the ndt-T2D group.

We further investigated whether theobserved differences are specifically formetformin treatment or just reflect theprogress of T2D in general. The concen-trations of the six metabolites are signif-icantly lower in mt-T2D than in the NGTand IGT groups (Supplementary Table4). In contrast, none of the six metabo-lites showed a significantly differentconcentration in the pairwise compari-sons among the four groups withoutmetformin treatment, i.e., NGT, IGT,ndt-T2D, and it-T2D (SupplementaryTable 4).

For sensitivity analysis, we tested theassociations of the six metabolites afteradding the duration of T2D to the fullyadjusted model. The three acyl-alkyl PCs(PC ae C36:4, PC ae C38:5, and PC aeC38:6), which are composed of at leastone polyunsaturated fatty acid (PUFA),remained significantly different in the

1862 Metformin, Metabolic Profiles, and LDL-C Diabetes Care Volume 38, October 2015

http://genome.ucsc.eduhttp://genome.ucsc.eduhttp://cran.r-project.org/http://cran.r-project.org/http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1

comparison between mt-T2D and ndt-T2D (P , 3.8E-04) (Fig. 2A), whereasthe other three metabolites were notsignificantly different anymore (Supple-mentary Table 5). After adjustment ofthe full model for 1) waist, 2) LDL-C,and 3) the combination of LDL-C andinsulin, the effect estimates of the sixmetabolites were almost unchanged(Supplementary Table 5).

Replication and Meta-analysisFor the three acyl-alkyl PCs, we ob-served consistent results in both repli-cation studies (ERF and NTR); i.e.,significantly lower levels were observedin mt-T2D patients compared with ndt-T2D individuals (P , 0.05) (Table 2 andFig. 2B and C). Additionally, a meta-analysis of the three studies (KORA F4,ERF, and NTR) yielded significant resultsfor the three replicated metabolites(P , 3.8E-04) (Table 2). We refer tothese three highly intercorrelated me-tabolites, which are not associated withfasting glucose or HbA1c, as metforminassociated in the following paragraphs(Supplementary Fig. 2).

In the longitudinal examination, wefound significantly decreased levels ofthe three metformin-associated metab-olites in patients who underwent met-formin treatment during the follow-up(P , 3.8E-04 using the fully adjustedmodel) (Supplementary Table 6). Con-sistent results for the three acyl-alkylPCs were observed in a sensitivity anal-ysis with a subgroup of 55 ndt-T2D pa-tients at the KORA S4, of whom 19 werendt-T2D patients and 36 were mt-T2Dpatients in KORA F4 in the fully adjustedmodel (P , 0.05) (Fig. 2D and Supple-mentary Table 6). These prospectivefindings confirmed our observations inthe cross-sectional study.

Relationship Between MetforminTreatment, the Three Metabolites, andLDL-C LevelsTo investigate a potentially mediatingeffect of the three acyl-alkyl PCs on theassociations between metformin treat-ment and lipid profiles, we explored theprospective data of 912 KORA partici-pants (Supplementary Table 1). Wefound that metformin treatment ac-counts for a significant decrease ofLDL-C and total cholesterol levels, whileits influence on HDL-C and triglycerideswas not significant in both crude and

Table

2—Metabolite

sasso

ciatedwith

metfo

rmin

treatm

entin

patie

nts

with

T2D

Metab

olite

Disco

veryKORA

Rep

licationER

FRep

licationNTR

Meta-an

alysis

Effectestim

ate(95%

CI)

PEffect

estimate

(95%CI)

PEffect

estimate

(95%CI)

PEffect

estimate

(95%CI)

P

Crudemodel

PCae

C36:4

20.66

(20.90,2

0.43)1.07E-07

21.03

(21.36,2

0.69)2.65E-08

20.97

(21.38,2

0.56)9.21E-06

20.86

(21.11,2

0.60)5.24E-11

PCae

C38:5

20.65

(20.90,2

0.41)2.92E-07

21.04

(21.37,2

0.71)1.94E-08

20.72

(21.12,2

0.32)5.75E-04

20.79

(21.04,2

0.55)2.86E-10

PCae

C38:6

20.58

(20.82,2

0.35)2.39E-06

20.76

(21.11,2

0.42)3.53E-05

20.61

(21.02,2

0.19)4.44E-03

20.63

(20.81,2

0.45)4.11E-12

PCaa

C36:0

20.50

(20.75,2

0.25)9.53E-05

20.30

(20.78,0.18)

2.31E-0120.54

(21.08,0.01)

5.43E-0220.47

(20.68,2

0.26)7.48E-06

PCaa

C38:0

20.73

(20.97,2

0.49)1.48E-08

20.31

(20.88,0.27)

2.98E-0120.56

(21.12,2

0.01)4.49E-02

20.64

(20.87,2

0.40)8.22E-08

Ornith

ine

20.57

(20.80,2

0.33)4.42E-06

20.03

(20.18,0.11)

6.49E-010.01

(20.45,0.46)

9.78E-0120.21

(20.59,0.17)

2.72E-01

Fullm

odel

PCae

C36:4

20.66

(20.91,2

0.41)4.92E-07

21.08

(21.33,2

0.83)8.37E-05

20.84

(21.27,2

0.42)1.67E-04

20.79

(21.02,2

0.55)5.27E-11

PCae

C38:5

20.62

(20.88,2

0.36)3.81E-06

21.11

(21.36,2

0.85)6.18E-05

20.62

(21.04,2

0.20)4.65E-03

20.73

(20.99,2

0.47)3.73E-08

PCae

C38:6

20.58

(20.82,2

0.34)4.94E-06

20.74

(21.00,2

0.48)6.15E-03

20.49

(20.91,2

0.07)2.15E-02

20.58

(20.78,2

0.39)2.96E-09

PCaa

C36:0

20.57

(20.84,2

0.30)4.25E-05

20.37

(20.62,2

0.12)1.41E-01

20.31

(20.86,0.24)

2.68E-0120.49

(20.71,2

0.27)9.79E-06

PCaa

C38:0

20.68

(20.94,2

0.42)5.25E-07

20.35

(20.64,2

0.05)2.47E-01

20.34

(20.90,0.22)

2.32E-0120.56

(20.82,2

0.30)1.85E-05

Ornith

ine

20.58

(20.80,2

0.19)3.82E-05

20.01

(20.09,0.07)

8.80E-0120.13

(20.62,0.36)

6.02E-0120.25

(20.62,0.13)

2.00E-01

Effectestim

atesandPvalu

esforthecompariso

nbetw

eenmt-T2D

andndt-T2D

were

calculated

usin

gmultivariab

lelinear

regressionanalysis

adjusted

forcru

demodel(age

andsex)

andfullm

odel,

which

inclu

des

age,sex,BMI,p

hysicalactivity,alco

holin

take,smokin

g,systolic

BP,H

DL-C

,triglycerides,H

bA1c ,fastin

gglu

cose,an

duse

ofstatin

s,b-blockers,A

CEinhibito

rs,andARBs.Lip

idsid

echain

compositio

nis

abbreviated

asCx:y,w

here

xden

otes

thenumber

ofcarb

onsin

thesid

echain

andythenumber

ofdouble

bonds.Statistically

significan

tPvalu

esappear

inboldface.

care.diabetesjournals.org Xu and Associates 1863

http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org

fully adjusted models (P, 0.05) (Fig. 2Eand Supplementary Table 7). In particu-lar, metformin was associated with a de-crease in LDL-C levels of 11.83 mg/dL.We therefore focused on the analysisof LDL-C and total cholesterol.After adding the three metabolites to

the full model, the direct association be-tween metformin treatment and LDL-Clevels was not significant anymore

(P = 0.25) (Fig. 2F and SupplementaryTable 8A). Based on longitudinal analy-sis, we found consistent results asreported above (Table 2); i.e., signifi-cantly reduced levels of the three me-tabolites in the metformin-treatedpatients were observed (e.g., for thesummed metabolite concentration P =2.16E-05) (Fig. 2F and SupplementaryTable 8B). Furthermore, we found

significant positive associations be-tween LDL-C and each of the three me-tabolites as well as their summedconcentration after adjusting for met-formin treatment (e.g., for the summedmetabolite concentration P = 6.87E-12)(Fig. 2F and Supplementary Table 8C).This means that these associations ofthe metformin-associated metaboliteswith LDL-C are independent of metfor-min treatment. Finally, for each of thethree metformin-associated metabo-lites (and their summed concentra-tion), the mediation effects on theassociation between metformin treat-ment and the LDL-C levels were signif-icant in both models (Table 3). Forinstance, the summed concentrationof the metabolites mediates 3.43 mg/dLreduction in LDL-C level, which accountsfor 29% of the total effect of metforminon LDL-C (Table 3).

To rule out the potential effect of sta-tin intake, we performed a sensitivityanalysis by excluding individuals tak-ing statin at baseline KORA S4 and/orfollow-up F4. The mediation effects ofthe summed concentration were alsosignificant for the associations betweenmetformin and LDL-C level (Supplemen-tary Table 9A). However, although thecrude and full model showed similarlysignificant mediation effects for totalcholesterol (Supplementary Table 8Dand E and Table 3), after excluding statinusers from the analysis, the effects ontotal cholesterol were not significantanymore with respect to the fully ad-justed model (P, 0.05) (SupplementaryTable 9B).

Seventeen Genes Are Linked toMetformin-Associated Metabolitesand Pathway AnalysisTo identify genes associated with thethree metabolites, we applied PSEA onthese metabolites in a subset of KORAF4 individuals (n = 1,809) with availablegenotyping data and metabolite profiles.We found 17 geneswith an enrichmentofSNPs in their transcribed or flanking re-gion (P, 1.0E-04) (Supplementary Table10). These genes belong to five clusters,one of them containing 12 genes locatedon chromosome 11. A literature searchrevealed disease phenotypes associatedwith these 17 genes. Six genes, namely,FADS1, FADS2, FADS3, MYRF, BEST1 andRAB3IL1, are associated with T2D or itscomorbidities, including retinopathy and

Figure 2—Differences in metabolite concentrations, mediation effect, and organ-specific path-ways. Mean residuals of the concentrations (with SEs) of three identified acyl-alkyl PC metab-olites for the NGT, IGT, ndt-T2D, mt-T2D, and it-T2D groups derived in cross-sectional analysis ofthe KORA F4 are shown in A. The mean residuals of the same metabolites in ERF are illustratedfor the NGT, ndt-T2D, and mt-T2D groups in B and in NTR for the non-T2D, ndt-T2D, and mt-T2Dgroups in C, respectively. D refers to the longitudinal setting of the KORA study and shows themean residuals of the concentrations (with SEs) of the threemetabolites with respect to changeswithin the 7 years between baseline and follow-up study when people were treated withmetformin. Residuals were calculated from linear regression model with the full adjustment.E: The association betweenmetformin and LDL-C without consideration of the threemetformin-associated metabolites. F: The results of the mediation analysis; the red cross indicates that thedirect association between metformin and LDL-C is not significant anymore. G: An overview ofthe involved pathways. The connections indicated by liver, hypothalamus, muscle, and bloodshow organ specificity between genes, pathway-related proteins, and metformin drug targets aswell as metformin. The metabolites (ellipses) were connected to metformin treatment (straightside hexagons) through genes (rounded rectangles), proteins (hexagons), andmetformin targets(rectangles). The activation or inhibition is indicated. Plus or minus symbol next to the lineindicates positive or negative association. For further information, see Table 3 and Supplemen-tary Tables 8 and 12.

1864 Metformin, Metabolic Profiles, and LDL-C Diabetes Care Volume 38, October 2015

http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1

coronary artery diseases (for references,see Supplementary Table 10).To explore potentially related path-

ways, we used a bioinformatics ap-proach, integrating the 17 identifiedgenes with 6 known metformin targetgenes (4) into a protein-protein interac-tion network (9,30,31). For 3 of the 17genes, there was no record for Homosapiens in the STITCH (30); therefore,we investigated the interaction of theremaining 14 genes with the 6 metfor-min targets (Supplementary Table 11).AMPK was found to be linked to FADS1and FADS2 through interacting proteins(leptin and sterol regulatory element–binding protein 1c [SREBP1c]). A man-ual evaluation of these interactionsin a literature research showed organspecificity, mainly referring to liver andhypothalamus (Fig. 2G). The AMPK com-plex is inhibited by leptin and metforminin the hypothalamus, whereas it is acti-vated by metformin and leptin in theliver. (References for each interactionare provided in the SupplementaryTable 12).

CONCLUSIONS

We found significant concentration dif-ferences for three metabolites (PC aeC36:4, PC ae C38:5, and PC ae C38:6) inthe blood of patients with T2D undermetformin treatment and replicatedthem in two independent studies. Weidentified SNP variations in 17 genes (in-cluding FADS1 and FADS2) that were as-sociated with the three metabolites.Based on these genes, we built an in-teraction network to investigate the

underlying mechanisms of metformintreatment and identified the organ-specificAMPK pathway. We further found thatthe reduced LDL-C levels in metformin-treated patients with T2D were medi-ated partially by the three acyl-alkylPCs. Sensitivity analyses were performedto consider the duration of diabetes andstatin use.

The levels of metabolites depend onmultiple modifiable factors, such as life-style and environment (9–11,13–16).We therefore considered a number ofconfounding effects, e.g., physiologicalparameters (age, sex, BMI, and systolicBP), lifestyle (physical activity, alcoholintake, and smoking), glucose levels(HbA1c and fasting glucose), lipid levels(HDL-C and triglycerides), and medi-cation usage (statins, b-blockers, ACEinhibitors, and ARBs). Additionally, in-termediates or end products of metab-olism are influenced by underlyinggenetic factors (23,24). In our study,phenotypes and genotypes are availablefor each person (n = 1,809); we thusused phenotype set enrichment analysis(25). Our combined analysis of geneticand metabolomic data enabled us toidentify genes associated with the threemetabolites and supported the identifi-cation of an organ-specific pathway. Theobservation of significantly lower levelsof the three metformin-associated me-tabolites (polyunsaturated acyl-alkylPCs) in the mt-T2D patients can be ex-plained by metformin’s effects on AMPKin the liver (Fig. 2G and SupplementaryTable 12). In the hepatocyte, metforminincreases the AMP-to-ATP ratio and thus

leads to the activation of AMPK. Acti-vated AMPK blocks SREBP1c, a tran-scription factor controlling enzymesinvolved in the fatty acid synthesis andinhibiting the synthesis of FADS1 andFADS2 (22). This results in a reducedsynthesis of unsaturated fatty acidsand consequently lower acyl-alkyl PCconcentrations. Leptin occupies a cen-tral position in the network (Fig. 2G)and affects the FADS complex via threedifferent interactions. In the liver, leptinnot only activates AMPK, thereby sup-pressing SREBP1c and downregulatingFADS1 and FADS2, but can also directlyinhibit both SREBP1c and FADS2 (34).Metformin and leptin exert opposite ef-fects in the hypothalamus and in theliver (for references, see SupplementaryTable 12), but further studies are re-quired to better understand the organ-specific metformin effects in humans.

Recently, clinical practice guidelineshave recommended the usage of met-formin as first-line therapy in T2D pa-tients with heart failure (1,2). Ourobservation of lower blood levels ofLDL-C in metformin-treated patientspoints toward a beneficial effect of met-formin for the prevention of CVD. Ameta-analysis of randomized clinical tri-als shows that metformin treatment re-sults in lowered LDL-C levels in newlydiagnosed T2D patients (8). Similar re-sults were also reported in patientswithout T2D in an epidemiological study(7). Here, we observed that metformintreatment leads to lowered LDL-C levels,an effect mediated most likely throughmetformin-mediated reduction of FADS

Table 3—Mediation effects of the three metabolites for the association between metformin treatment and reduction of LDL-Cand total cholesterol

Crude model Full model

Effect estimate(95% CI) P Explained effect (%)

Effect estimate(95% CI) P Explained effect (%)

LDL-CPC ae C36:4 23.05 (24.38, 21.71) 2.21E-04 25.74 22.51 (23.71, 21.31) 1.33E-03 21.22PC ae C38:5 22.94 (24.21, 21.67) 2.65E-04 24.82 22.36 (23.50, 21.22) 1.94E-03 19.97PC ae C38:6 25.25 (28.11, 22.40) 1.34E-05 44.40 24.02 (26.59, 21.44) 4.55E-04 33.95Summed concentration† 24.37 (26.37, 22.37) 1.52E-05 36.92 23.43 (25.19, 21.67) 2.95E-04 28.99

Total cholesterolPC ae C36:4 25.00 (27.77, 22.23) 2.63E-05 26.1 23.08 (24.72, 21.45) 7.42E-04 26.5PC ae C38:5 24.99 (27.71, 22.26) 2.33E-05 26.0 22.77 (24.25, 21.30) 1.38E-03 23.8PC ae C38:6 26.99 (211.86, 22.13) 8.96E-06 36.4 24.16 (26.98, 21.35) 5.11E-04 35.8Summed concentration† 26.63 (210.55, 22.71) 2.76E-06 34.6 23.87 (26.05, 21.69) 2.41E-04 33.3

The estimates of the mediation effects and P values were calculated using the longitudinal (KORA S4→F4) mediation analysis adjusted for the crudeand full model. The mediation effects for the three metformin-associated metabolites and the summed concentration are shown. †The summedconcentration refers to the overall concentration of the three metabolites (PC ae C36:4, PC ae C38:5, and PC ae C38:6).

care.diabetesjournals.org Xu and Associates 1865

http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-0658/-/DC1http://care.diabetesjournals.org

activity and consequently reduction ofthe levels of PUFA, namely, arachidonicacid (35). It has been suggested thatlower levels of arachidonic acid leadsto an increased membrane fluidity,thus increasing LDL-C receptor recycling(35). This hypothesis is especially strong,given that genetic variants assigned tolower activity of FADS1 and -2 were sig-nificantly associated with lower LDL-Clevels (36). While certain PCs can indeedexert antidiabetic effects (37), furthermechanistic studies are required totest whether lowering of these circulatinglipids contributes directly to the preven-tion of CVDs or merely by its antidiabeticeffect (1).Beyond its common antihyperglycemic

action and its effect in lowering LDL-C,metformin can potentially reduce therisk of cancer mortality and diminish theprogression of cancer (38). In the currentstudy,wehave found the threemetformin-associated metabolites significantly as-sociated with two genes, FEN1 andC20orf94, which are involved in DNA re-pair (39,40). This may partly explain thatmetformin has been shown to influencethe prevalence of different types of car-cinoma, such as gastrointestinal cancers(39) and leukemia (40).The strength of our study is that we

used three independent cohort studiesto discover and replicate our observa-tions. Importantly, all results presentedin this study were independent of phys-iological parameters, lifestyle, glucoselevels, lipid levels, and medication. Wecombined metabolomics and genomicsdata, broad literature research, and organ-specific information from animal studies todeepen the insight into the underlyingmechanisms.Our findings are limited by the obser-

vational nature of cohort studies, andthe applied methods, such as the medi-ation analysis, are of purely statisticalcharacter, but they offer the opportu-nity to raise new questions for experi-mental confirmation studies, such asrandomized controlled clinical trials toinvestigate, for instance, the effect ofmetformin on blood lipid levels of pa-tients without diabetes.In the present studies (KORA, ERF, NTR),

the duration of T2D is based on self-reported information. Moreover, neitherdata on the dosage nor data on durationand compliance of the metformin treat-ment were available. Furthermore, it has

to be mentioned that the degree of dia-betes severity presumably discriminatesthe different groups of patients (ndt-T2D,mt-T2D, and it-T2D), which is reflected bytheir HbA1c and fasting glucose values(Table 1). Although the investigated me-tabolite panel does not represent thewhole human metabolome, the compre-hensive analysis of .130 metabolitesfrom different classes represents a con-siderable improvement compared withprevious technologies.

We found threemetformin-associatedmetabolites, which showed no overlapwith the findings of previous studies(17–21). This is likely to result from the useof different sampling matrices (plasmavs. serum), unmeasured metabolites(asymmetric dimethylarginine), or studydesign (glipizide treatment). Additional,our study considered considerably morepotential cofounding effects in a compa-rably larger number of individuals thanprevious studies (17–21).

In conclusion, we observed thatmetfor-min treatment reduced levels of the threeacyl-alkyl PC metabolites in patients withT2D. This change in the metabolic profilesmay mediate lowered blood levels of LDL-C.The underlying mechanism is most likelythe metformin-induced activation ofAMPK and the consequent suppressionof SREBP1c and FADS, which leads to re-duced levels of PUFA and LDL-C. Our find-ings suggest a pharmaco-epidemiologicmechanismbywhichmetforminmayexertbeneficial effects to prevent CVD. Moreimportantly, our study suggests a novelapproach to identify pleiotropic effects ofmedication using multilevel omics data.

Acknowledgments. The authors express theirappreciation to all KORA study participants fordonating their blood and time. The authorsthank the field staff in Augsburg conducting theKORA studies. The authors thank the staff fromthe Institute of Epidemiology at the HelmholtzZentrum München and the Genome AnalysisCenter Metabolomics Platform, who helped inthe sample logistics, data and straw collection,and metabolomic measurements, and especiallyJ. Scarpa, K. Faschinger, F. Scharl, N. Lindemann,H. Chavez, A. Sabunchi, A. Schneider, A. Ludolph,S. Jelic, and B. Langer. The authors are grateful toall general practitioners involved in ERF for theircontributions. The authors thank all participantsin the NTR.Funding. The KORA study was initiated andfinanced by the Helmholtz Zentrum München–German Research Center for EnvironmentalHealth, which is funded by the German FederalMinistry of Education and Research (BMBF) and

by the Free State of Bavaria. Furthermore,KORA research was supported within the Mu-nich Center of Health Sciences (MC-Health),Ludwig-Maximilians-Universität, as part ofLMUinnovativ. Part of this project was sup-ported by the European Community’s SeventhFramework Programme grants HEALTH-2009-2.2.1-3/242114 (Project OPTiMiS) and HEALTH-2013-2.4.2-1/602936 (Project CarTarDis). The Ger-man Diabetes Center is funded by the GermanFederal Ministry of Health (Berlin, Germany)and the Ministry of Innovation, Science and Re-search of the State of North Rhine-Westphalia(Düsseldorf, Germany). The diabetes part of theKORA F4 study was funded by a grant from theGerman Research Foundation (DFG) RA 459/3-1.This study was supported in part by a grant fromthe BMBF to the German Center for DiabetesResearch (DZD e. V.). W.R.-M. is funded by theGerman Federal Ministry of Education and Re-search grant 03IS2061B (project Gani_Med).K.Su. is supported by Biomedical Research Pro-gram funds at Weill Cornell Medical College inQatar, a program funded by theQatar Foundation.The EGCUT received support from the EuropeanCommunity’s Seventh Framework Programmegrant BBMRI-LPC 313010, targeted financingfrom Estonian Government IUT20-60 and IUT24-6, Estonian Research Roadmap through the Esto-nian Ministry of Education and Research(3.2.0304.11-0312), theCenter of Excellence inGe-nomics (EXCEGEN), and Development Fund fromthe University of Tartu (SP1GVARENG), and froman EFSD New Horizons grant. This work was alsosupported by the U.S. National Institutes ofHealth (R01DK075787). The ERF was supportedby grants from the Netherlands Organisation forScientific Research (NWO) and Erasmus MC andthe Centre for Medical Systems Biology (CMSB).Telomere length assessment was supportedthrough funds fromtheEuropeanCommunity’sSev-enth Framework Programme (FP7/2007-2013),grant agreement HEALTH-F4-2007-201413(ENGAGE). Research for the NTR was fundedby the Netherlands Organisation for ScientificResearch (NWO) (MagW/ZonMW grants 904-61-090, 985-10-002, 904-61-1, 480-04-004,400-05-717; Addiction-31160008 Middelgroot-911-09-032; and Spinozapremie 56-464-14192),the Center for Medical Systems Biology (CSMB,NWO Genomics), NBIC/BioAssist/RK(2008.024),Biobanking and Biomolecular Resources Re-search Infrastructure (BBMRI –NL) (184z021z007),the VU University’s Institute for Health and CareResearch (EMGO+), and the European Com-munity’s Seventh Framework Program (FP7/2007-2013), grant HEALTH-F4-2007- 201413(ENGAGE).Duality of Interest. No potential conflicts ofinterest relevant to this article were reported.Author Contributions. T.X., S.B., A.C.M.,C.He., H.H.M.D., A.D., Z.Y., J.S.R., and T.Hal.analyzed the data and interpreted the results.T.X., S.B., A.C.M., C.He., M.Rod., T.M., and R.W.-S.wrote the manuscript. M.H., R.S., C.Ho., J.Adam,S.C., M.Rot., T.P., Y.H., G.K., C.M., B.T., A.R.,M.H.d.A., Y.L., H.-E.W., B.S., A.M., C.G., K.St.,and W.R. assisted in manuscript generation andrevision. M.C., G.F., A.S.Q., G.W., R.P., K.K., andK.W.v.D. interpreted the results. C.P., W.R.-M.,K.Su., T.Han., J.Adams., and T.I. performed themetabolic profiling. T.I., A.P., C.M.v.D., D.I.B.,

1866 Metformin, Metabolic Profiles, and LDL-C Diabetes Care Volume 38, October 2015

T.M., and R.W.-S. conceived and designed thestudy. R.W.-S. is the guarantor of thiswork and, assuch, had full access to all the data in the studyand takes responsibility for the integrity of thedata and the accuracy of the data analysis.

References1. Foretz M, Guigas B, Bertrand L, Pollak M,Viollet B. Metformin: from mechanisms of ac-tion to therapies. Cell Metab 2014;20:953–9662. ADA. Executive summary: Standards of med-ical care in diabetes–2013. Diabetes Care 2013;36(Suppl. 1):S4–S103. Madiraju AK, Erion DM, Rahimi Y, et al. Met-formin suppresses gluconeogenesis by inhibitingmitochondrial glycerophosphate dehydrogenase.Nature 2014;510:542–5464. Law V, Knox C, Djoumbou Y, et al. DrugBank4.0: shedding new light on drug metabolism.Nucleic Acids Res 2014;42:D1091–D10975. Decensi A, Puntoni M, Goodwin P, et al. Met-formin and cancer risk in diabetic patients: a sys-tematic review and meta-analysis. Cancer PrevRes (Phila) 2010;3:1451–14616. Norwood DK, Chilipko AA, Amin SM, MachariaD, Still KL. Evaluating the potential benefits ofmet-formin in patients with cardiovascular disease andheart failure. Consult Pharm 2013;28:579–5837. Glueck CJ, Fontaine RN, Wang P, et al. Met-formin reducesweight, centripetal obesity, insulin,leptin, and low-density lipoprotein cholesterol innondiabetic, morbidly obese subjects with bodymass index greater than 30. Metabolism 2001;50:856–8618. Wulffelé MG,KooyA, deZeeuwD,StehouwerCD,Gansevoort RT. The effect of metformin on bloodpressure, plasma cholesterol and triglycerides intype 2 diabetes mellitus: a systematic review. JIntern Med 2004;256:1–149. Wang-Sattler R, Yu Z, Herder C, et al. Novelbiomarkers for pre-diabetes identified by me-tabolomics. Mol Syst Biol 2012;8:61510. Wang TJ, Larson MG, Vasan RS, et al. Me-tabolite profiles and the risk of developing di-abetes. Nat Med 2011;17:448–45311. Floegel A, Stefan N, Yu Z, et al. Identificationof serum metabolites associated with risk oftype 2 diabetes using a targeted metabolomicapproach. Diabetes 2013;62:639–64812. Newgard CB, An J, Bain JR, et al. A branched-chain amino acid-related metabolic signature thatdifferentiates obese and lean humans and con-tributes to insulin resistance. Cell Metab 2009;9:311–326

13. Yu Z, Zhai G, Singmann P, et al. Human se-rum metabolic profiles are age dependent. Ag-ing Cell 2012;11:960–96714. Mittelstrass K, Ried JS, Yu Z, et al. Discoveryof sexual dimorphisms in metabolic and geneticbiomarkers. PLoS Genet 2011;7:e100221515. Xu T, Holzapfel C, Dong X, et al. Effects ofsmoking and smoking cessation on human se-rum metabolite profile: results from the KORAcohort study. BMC Med 2013;11:6016. JaremekM, Yu Z, ManginoM, et al. Alcohol-induced metabolomic differences in humans.Transl Psychiatry 2013;3:e27617. Marchetti P, Masiello P, Benzi L, et al. Ef-fects of metformin therapy on plasma aminoacid pattern in patients with maturity-onset di-abetes. Drugs Exp Clin Res 1989;15:565–57018. Cai S, Huo T, Li N, Xiong Z, Li F. Lysophos-phatidylcholine–biomarker of Metformin ac-tion: studied using UPLC/MS/MS. BiomedChromatogr 2009;23:782–78619. Huo T, Cai S, Lu X, Sha Y, YuM, Li F. Metabo-nomic study of biochemical changes in the se-rum of type 2 diabetes mellitus patients afterthe treatment of metformin hydrochloride. JPharm Biomed Anal 2009;49:976–98220. Asagami T, Abbasi F, StuelingerM, et al. Met-formin treatment lowers asymmetric dimethylar-ginine concentrations in patients with type 2diabetes. Metabolism 2002;51:843–84621. Zhang Y, Hu C, Hong J, et al. Lipid profilingreveals different therapeutic effects of metfor-min and glipizide in patients with type 2 diabe-tes and coronary artery disease. Diabetes Care2014;37:2804–281222. Gieger C, Geistlinger L, Altmaier E, et al.Genetics meets metabolomics: a genome-wideassociation study of metabolite profiles in hu-man serum. PLoS Genet 2008;4:e100028223. Illig T, Gieger C, Zhai G, et al. A genome-wide perspective of genetic variation in humanmetabolism. Nat Genet 2010;42:137–14124. Suhre K, Shin SY, Petersen AK, et al.;CARDIoGRAM. Human metabolic individualityin biomedical and pharmaceutical research.Nature 2011;477:54–6025. Ried JS, Döring A, Oexle K, et al. PSEA: Phe-notype Set Enrichment Analysis–a new methodfor analysis of multiple phenotypes. Genet Epi-demiol 2012;36:244–25226. Rathmann W, Kowall B, Heier M, et al.Prediction models for incident type 2 diabe-tes mellitusin the older population: KORA S4/F4 cohort study. Diabet Med 2010;27:1116–1123

27. Jourdan C, Petersen AK, Gieger C, et al.Body fat free mass is associated with the serummetabolite profile in a population-based study.PLoS One 2012;7:e4000928. Aulchenko YS, Heutink P, Mackay I, et al.Linkage disequilibrium in young genetically iso-lated Dutch population. Eur J Hum Genet 2004;12:527–53429. Boomsma DI, de Geus EJ, Vink JM, et al.Netherlands Twin Register: from twins to twinfamilies. Twin Res 2006;9:849–85730. Kuhn M, Szklarczyk D, Pletscher-Frankild S,et al. STITCH 4: integration of protein-chemicalinteractions with user data. Nucleic Acids Res2014;42:D401–D40731. Lechner M, Höhn V, Brauner B, et al. CIDeR:multifactorial interaction networks in humandiseases. Genome Biol 2012;13:R6232. WillemsenG, deGeus EJ, BartelsM, et al. TheNetherlands Twin Register biobank: a resourcefor genetic epidemiological studies. Twin Res2010;13:231–24533. Bauer DJ, Preacher KJ, Gil KM. Conceptual-izing and testing random indirect effects andmoderated mediation in multilevel models:new procedures and recommendations. PsycholMethods 2006;11:142–16334. Lim CT, Kola B, Korbonits M. AMPK as amediator of hormonal signalling. J Mol Endocri-nol 2010;44:87–9735. Sone Y, Kido T, Ainuki T, et al. Genetic var-iants of the fatty acid desaturase gene clusterare associated with plasma LDL cholesterollevels in Japanese males. J Nutr Sci Vitaminol(Tokyo) 2013;59:325–33536. Global Lipids Genetics Consortium, WillerCJ, Schmidt EM, Sengupta S, et al. Discoveryand refinement of loci associated with lipid lev-els. Nat Genet 2013;45:1274–128337. Lee JM, Lee YK, Mamrosh JL, et al.Anuclear-receptor-dependentphosphatidylcholinepathway with antidiabetic effects. Nature 2011;474:506–51038. Pernicova I, KorbonitsM.Metformin–mode ofaction and clinical implications for diabetes andcancer. Nat Rev Endocrinol 2014;10:143–15639. Liu L, Zhou C, Zhou L, et al. Functional FEN1genetic variants contribute to risk of hepatocellularcarcinoma, esophageal cancer, gastric cancer andcolorectal cancer. Carcinogenesis 2012;33:119–12340. Meissner B, Bartram T, Eckert C, et al. Fre-quent and sex-biased deletion of SLX4IP by ille-gitimate V(D)J-mediated recombination inchildhood acute lymphoblastic leukemia. HumMol Genet 2014;23:590–601

care.diabetesjournals.org Xu and Associates 1867

http://care.diabetesjournals.org