Novel Metabolic Markers for theRisk of Diabetes Development inAmerican IndiansDiabetes Care 2015;38:220–227 | DOI: 10.2337/dc14-2033

OBJECTIVE

To identify novel metabolic markers for diabetes development in AmericanIndians.

RESEARCH DESIGN AND METHODS

Using an untargeted high-resolution liquid chromatography–mass spectrometry,we conducted metabolomics analysis of study participants who developed inci-dent diabetes (n = 133) and those who did not (n = 298) from 2,117 normoglycemicAmerican Indians followed for an average of 5.5 years in the Strong Heart FamilyStudy. Relative abundances of metabolites were quantified in baseline fastingplasma of all 431 participants. Prospective association of each metabolite withrisk of developing type 2 diabetes (T2D) was examined using logistic regressionadjusting for established diabetes risk factors.

RESULTS

Seven metabolites (five known and two unknown) significantly predict the risk ofT2D. Notably, one metabolite matching 2-hydroxybiphenyl was significantly as-sociatedwith an increased risk of diabetes, whereas fourmetabolites matching PC(22:6/20:4), (3S)-7-hydroxy-29,39,49,59,8-pentamethoxyisoflavan, or tetrapeptideswere significantly associated with decreased risk of diabetes. Amultimarker scorecomprising all seven metabolites significantly improved risk prediction beyondestablished diabetes risk factors including BMI, fasting glucose, and insulinresistance.

CONCLUSIONS

The findings suggest that these newly detected metabolites may represent novelprognostic markers of T2D in American Indians, a group suffering from a dispro-portionately high rate of T2D.

Type 2 diabetes (T2D) is a metabolic disorder characterized by hyperglycemia re-sulting from impaired insulin secretion and increased insulin resistance (1). Thepathogenesis of T2D is complex, involving both genetic and environmental factors,but the precise mechanisms underlying T2D development remain incompletely un-derstood. Traditional risk factors such as age, sex, obesity, fasting glucose, andinsulin resistance contribute considerably to disease risk and have therefore beenwidely used for routine diagnosis or risk stratification, but most of thesemarkers failto capture the complexity of disease etiology and thus have limitations in detectingearly metabolic abnormalities that may occur years or even decades before theonset/diagnosis of overt T2D. Characterization of metabolic profiles and perturbed

1Department of Epidemiology, Tulane UniversitySchool of Public Health, New Orleans, LA2Department of Biostatistics, School of PublicHealth, University of North Carolina at ChapelHill, Chapel Hill, NC3Division of Pulmonary Medicine, Emory Univer-sity School of Medicine, Atlanta, GA4Department of Biostatistics and Bioinformatics,EmoryUniversity School ofPublicHealth, Atlanta,GA5Center for American Indian Health Research,University of Oklahoma Health Sciences Center,Oklahoma City, OK6Medstar Health Research Institute and George-town and Howard Universities Centers for Trans-lational Sciences, Washington, DC

Corresponding author: Jinying Zhao, jzhao5@tulane.edu.

Received 25 August 2014 and accepted 31October 2014.

This article contains Supplementary Data onlineat http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc14-2033/-/DC1.

© 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.

See accompanying articles, pp. 186,189, 197, 206, 213, and 228.

Jinying Zhao,1 Yun Zhu,1 Noorie Hyun,2

Donglin Zeng,2 Karan Uppal,3

ViLinh T. Tran,3 Tianwei Yu,4 Dean Jones,3

Jiang He,1 Elisa T. Lee,5 and

Barbara V. Howard6

220 Diabetes Care Volume 38, February 2015

HEA

LTHDISPARITIESIN

DIABETES

http://crossmark.crossref.org/dialog/?doi=10.2337/dc14-2033&domain=pdf&date_stamp=2015-01-07mailto:jzhao5@tulane.edumailto:jzhao5@tulane.eduhttp://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc14-2033/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc14-2033/-/DC1

metabolic pathways implicated in T2Ddevelopment will not only provide novelinsights into disease pathophysiologybut also provide instrumental data forrisk prediction and for developing effec-tive therapeutic and preventive strate-gies against diabetes.Metabolomics is an emerging analyt-

ical technology that simultaneouslyquantifies many metabolites in bio-fluids. These metabolites represent theend products of cellular metabolism inresponse to intrinsic and extrinsic stim-uli and thus may reflect the metabolicchanges at earlier stages of disease.Cross-sectional analyses have reportedassociations of altered metabolites withobesity (2), insulin resistance (3), predi-abetes, and overt T2D (4–7). Thesechanges included acylcarnitines (6,8),amino acids (2,8), sugars (5,7), and dif-ferent lipid species (5,8,9). Higherplasma levels of branched-chain aminoacids (BCAAs) and aromatic amino acidswere associated with an increased riskof T2D in the FraminghamOffspring study(10). Another study found that increaseddiacyl-phosphatidylcholines and reducedacyl-alkyl- and lyso-phosphatidylcholinesaswell as sphingomyelinswere associatedwith diabetes in a European population(11). More recently, a-hydroxybutyrateand linoleoylglycerophosphocholinewere also found to predict the develop-ment of dysglycemia and T2D in Euro-peans (12). These findings derived fromEuropean populations, however, maynot represent metabolic alterations inother ethnic groups. Moreover, most ex-isting studies used a targeted metabolo-mics approach by focusing on a subset ofpreselected metabolites and thus mayhave limited ability in discovering noveldisease-related metabolic changes. Theclinical utility of previously detected me-tabolites in risk prediction was either notreported or was minimal over conven-tional clinical factors.The goal of this study is to identify

predictive metabolic markers for futurerisk of T2D in American Indians, a minor-ity group suffering from a disproportion-ately high rate of T2D. Metabolicprofiles of diabetes development wereexamined in normoglycemic partici-pants using fasting plasma samples col-lected prior to disease occurrence. Theutility of novel metabolic markers in riskprediction beyond established diabetesrisk factors was also investigated.

RESEARCH DESIGN AND METHODSStudy PopulationParticipants included in the currentstudy were selected from the StrongHeart Family Study (SHFS), a family-based prospective study designed toidentify genetic factors for cardiovascu-lar disease (CVD), diabetes, and theirrisk factors in American Indians residingin Arizona, North and South Dakota, andOklahoma. A detailed description forthe study design and methods of theSHFS had been reported previously(13,14). In brief, a total of 3,665 tribalmembers (aged 14 years and older)from 94 multiplex families (65 three-generation and 29 two-generation fami-lies, average family size 38) were recruitedand examined in 2001–2003. All livingparticipants were followed and reex-amined between 2006 and 2009. TheSHFS protocol was approved by the in-stitutional review boards from the IndianHealth Service and the participatingstudy centers. All participants gave in-formed consent.

According to the American DiabetesAssociation 2003 criteria (15), diabeteswas defined as fasting plasma glucose$7.0 mmol/L or hypoglycemic medica-tions. Impaired fasting glucose was de-fined as a fasting glucose of 6.1–6.9mmol/L and no hypoglycemicmedications,and normal fasting glucose (NFG) wasdefined as fasting glucose ,6.1 mmol/L.Incident cases of T2D were definedas normal fasting glucose at baseline(2001–2003) and development of newT2D by the end of follow-up (2006–2009).

Participants included in the currentanalysis have to meet the following cri-teria: 1) attended clinical examinationsat both baseline (2001–2003) andfollow-up (2006–2009), 2) had NFG atbaseline, 3) were free of overt CVD andhypoglycemic medications at baseline,and 4) had available fasting plasma sam-ple at baseline for the proposed met-abolomic analysis. Participants withmissing information for fasting glucoseor antidiabetes medication at eitherbaseline or follow-up were also excludedfrom the current analysis.

A total of 2,324 participants free ofovert CVD at baseline attended bothclinical visits and had available fastingplasma samples for the proposed analy-sis. Of these, 2,117 normoglycemic par-ticipants met all of the criteria listed

above. After an average 5.5 years offollow-up, 197 participants (9.3%) de-veloped incident T2D. Among thosewho did not develop T2D (n = 1,920),159 participants (7.5%) progressed toimpaired fasting glucose, whereas theother individuals (n = 1,761) remainedwith stable NFG by the end of follow-up. The current metabolomics analysismeasured metabolite levels in fastingplasma of 431 participants, including133 incident cases randomly selectedfrom participants who developed newT2D (n = 197) and 298 control subjectsrandomly selected from those who didnot develop T2D (n = 1,920). Supple-mentary Table 1 shows the comparisonof baseline clinical characteristics be-tween participants who were selectedand those not selected.

Assessments of Diabetes Risk FactorsFasting plasma glucose, insulin, lipids,lipoproteins, and inflammatory bio-markers were measured by standardlaboratory methods (14,16). BMI wascalculated as body weight in kilogramsdivided by the square of height in me-ters. Hypertension was defined as bloodpressure levels $140/90 mmHg or useof antihypertensive medications. Insulinresistance was assessed using HOMA ac-cording to the following formula: HOMAof insulin resistance (HOMA-IR) = fastingglucose (mg/dL)3 insulin (mU/mL)/405(17). Renal function was assessed usingthe estimated glomerular filtration rate(eGFR) calculated by the MDRD equation(18). For cigarette smoking, subjectswereclassified as current smokers, formersmokers, and nonsmokers. Alcohol con-sumption was determined by self-reported history of alcohol intake, thetype of alcoholic beverages consumed,frequency of alcohol consumption, andaverage quantity consumed per day andper week. Participants are classified ascurrent drinkers, former drinkers, andnever drinkers. Dietary intake was as-sessed using the block food frequencyquestionnaire (19).

Metabolic Profiling by High-Resolution Liquid Chromatography–Mass SpectrometryRelative abundance of fasting plasmametabolites was determined usinghigh-resolution liquid chromatography–mass spectrometry (LC-MS). Detailed lab-oratory protocols have previously beendescribed (20,21). Briefly, 65 mL plasma

care.diabetesjournals.org Zhao and Associates 221

http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc14-2033/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc14-2033/-/DC1http://care.diabetesjournals.org

sample aliquots were treated with ace-tonitrile, spiked with internal standardmix, and centrifuged at 13,000g for10min at 48C to remove proteins. Super-natant (130 mL) was removed andloaded into autosampler vials. Anion ex-change (AE) columns (both C18 and AEcolumns) were equilibrated to the initialcondition for 1.5 min prior to the nextsample injection. Mass spectral datawere collected with a 10-min gradienton a Thermo LTQ-Velos Orbitrap massspectrometer (Thermo Fisher, San Diego,CA) to collect data from mass/charge ra-tio (m/z) 85–2,000 in a positive ionizationmode. Three technical replicates wererun for each sample using a dual-columnchromatography procedure with C18and an AE column. Pooled plasma sam-ples were included in each batch (n = 23)for quality control. Peak extraction, dataalignment, and feature quantificationwere performed using the adaptive pro-cessing software (apLCMS) (22,23), acomputer package designed for high-resolution metabolomics data analysis.Feature and sample quality assessmentwas performed based on coefficient ofvariation (CV) and Pearson correlation,respectively, based on the technical rep-licates using xMSanalyzer (24). Metabo-lites with CV .50% in our samples wereexcluded from further analyses. Potentialmetabolite identitieswere determined byperforming an online search (10 ppmmass accuracy) against the Metlin data-base (25), the Human Metabolomics Da-tabase (26), and the LIPIDMAPS structuredatabase (27). Data filtering, normaliza-tion, diagnostics, and summarizationwere performed using the computerpackage MSPrep (28). Missing data wereimputed using the half of the minimumobserved value within each metaboliteacross all samples. Batch effect was cor-rected using the algorithm ComBat (29)implemented in MSPrep.

Statistical AnalysisPrior to analysis, metabolites data werelog transformed and standardized tounit variance and zero mean (z scores).Continuous variables were also con-verted to standard normal distribu-tions with corresponding mean andSD. Pearson partial correlation coeffi-cients were calculated between identi-fied metabolites and establishedclinical factors, adjusting for age, sex,and study site.

To identify metabolic predictors andto estimate their effects on the risk ofdeveloping T2D, we constructed a Coxproportional hazards frailty model, inwhich time to event was the dependentvariable and the level of each metabo-lite was the independent variable. Thefrailty model was used here to accountfor the relatedness among family mem-bers. The proportional hazards assump-tion was tested using the Schoenfeldresiduals, and it shows that the propor-tionality assumption holds in our data.For estimation of metabolic effects thatare independent of traditional risk fac-tors, the Cox frailty model was adjustedfor age, sex, site, BMI, eGFR, HDL, triglyc-erides, fasting glucose, and insulin re-sistance (assessed by HOMA-IR) atbaseline. Given the potential high corre-lations among detected metabolites, weused the q value method to adjustfor multiple testing (30), and a q value,0.05 was considered statisticallysignificant.

To examine the combined effects ofmetabolites on diabetes risk, weconstructed a multimarker metabolitesscore based on metabolites that are sig-nificantly predictive of diabetes risk byfitting a model according to the follow-ing formula: b1X1 + b2X2+ b3X3, where Xidenotes the z score of the i-th metabo-lite and bi denotes the regression coef-ficient from the logistic regressionmodel containing the indicated metab-olites. The joint predictive ability of me-tabolites was assessed using logisticregression by including all clinical riskfactors (age, sex, study site, BMI, eGFR,HDL, triglycerides, fasting glucose, andHOMA-IR) plus the multimarker metab-olite score compared with the model in-cluding clinical risk factors only. Wecalculated the area under the receiveroperating characteristic curve (AUC),the net reclassification improvement(NRI), and the integrated discriminationimprovement (IDI) to assess the incre-mental value of the metabolic markersfor risk prediction beyond classical riskfactors. Because our analysis was basedon a regression model with no cross-validation or external validation, it islikely that our model could be overfit-ted. To avoid or minimize bias due tooverfitting, we conducted a bootstrapestimation (1,000 reps) for coefficientsby SAS to obtain bias-corrected esti-mates ofmetabolites on risk of diabetes.

To identify metabolic profiles associ-ated with risk of diabetes, we conductedsparse partial least-squares discriminantanalysis (sPLS-DA) using the computerpackage mixOmics implemented in R.The sPLS-DA is a supervised, multivari-ate technique to determine metabolicgroups associated with disease risk.The sPLS-DA analysis included only me-tabolites showing significant associa-tions with risk of diabetes. For ease ofvisualization, we presented a Manhat-tan plot (2log10 P vs. metabolic feature)to show the significance of individualmetabolites according to status of inci-dent cases at follow-up using raw P val-ues obtained from multivariate logisticregression analysis (false discovery rateat q = 0.05 with a horizontal line).

RESULTS

Table 1 presents the characteristics ofthe study participants at baseline(2001–2003) according to diabetes sta-tus at the end of follow-up (2006–2009).The average follow-up period was 5.5years. Compared with participants whodid not develop T2D, those who devel-oped incident T2D had higher levels ofBMI, triglycerides, fasting glucose, fast-ing insulin, and insulin resistance(HOMA-IR) but lower level of HDL atbaseline. We also compared partici-pants who were selected (n = 431) ver-sus those not selected (n = 1,686) forthis study. It shows that, except forBMI and eGFR, selected participantswere not appreciably different fromthose not selected (SupplementaryTable 1).

Our untargeted high-resolution LC-MSdetected 11,628 distinct ions (m/z) withCV #10%, of which 2,093 m/z featuresmatched known compounds in availablemetabolomics databases. Among all11,628 features, altered levels of sevenmetabolites (five matching known metab-olites and two unknown) were signifi-cantly associated with risk of diabetesafter adjustment for clinical factors andmultiple testing. Specially, a metabolitematching 2-hydroxybiphenyl (2HBP)and an unknown chemical (m/z ratio1,178.804 [named X-1178]) were signif-icantly associated with an increased riskof diabetes, whereas five metabo-lites matching phosphatidylcholine (PC22:6/20:4), (3S)-7-hydroxy-29,39,49,59,8-pentamethoxyisoflavan (HPMF), two tet-rapeptides (Met-Glu-Ile-Arg [MEIR] and

222 New Metabolic Markers for Diabetes Risk Diabetes Care Volume 38, February 2015

http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc14-2033/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc14-2033/-/DC1

Leu-Asp-Tyr-Arg [LDYR]), and an un-known metabolite (m/z ratio 490.816[named X-490]) were significantly asso-ciated with a decreased risk of diabetes.These associations are independent ofclinical factors including fasting glucoseand insulin resistance. Per-SD increase inthe log-transformed levels of matching2HBP and X-1178 was associated with80% and 89%, respectively, increasedrisk of T2D. By contrast, per-SD increasein the log-transformed levels of match-ing PC (22:6/20:4), HPMF, tetrapeptides,and X-490 was associated with 32–42%decreased risk of T2D. In themultivariatemodel categorizing metabolites as ter-tiles, participants in the top tertile of2HBP and X-1178 had a hazard ratio(HR) of 2.80 (95% CI 1.19–6.60) and2.87 (95% CI 1.08–7.60) for developingincident T2D, respectively, comparedwith those in the lowest tertile. In con-trast, participants in the top tertile of PC(22:6/20:4), HPMF, MEIR, LDYR, andX-490 had an HR of 0.45 (95% CI 0.21–0.97), 0.38 (95% CI 0.18–0.80), 0.44 (95%CI 0.20–0.96), 0.37 (95% CI 0.16–0.87),and 0.46 (95% CI 0.21–0.97) for develop-ing T2D, respectively, comparedwith thosein the lowest tertile of these metabolites.To estimate the joint effects of metab-

olites on risk of diabetes development,we calculated HRs across tertiles of thecombined metabolites comprising allseven significant metabolites. For thetwo risk metabolites (2HBP and X-1178),the HR for risk of developing incidentT2D by comparing the top with the bot-tom tertiles of the summed metaboliteswas 6.89 (95% CI 2.63–18.08). For thefive protective metabolites (PC [22:6/20:4], HPMF, MEIR, LDYR, and X-490),the HR of the top compared with thebottom tertiles of summed metaboliteswas 0.23 (95% CI 0.10–0.51). Multivari-ate associations of each individual me-tabolite along with their combinedeffects on diabetes risk are shown inTable 2. Of note, regression coefficientslisted in Table 2 were corrected for po-tential overfitting by bootstrapping andthus should represent unbiased esti-mates of metabolic effects on risk ofT2D. For ease of visual inspection,Fig. 1 shows a Manhattan plot (2log10P vs. metabolic feature) of all metabolitesusing raw P values obtained from multi-variate regression analysis. Metabolitessignificantly predictive of diabetes riskare shown at the level of q = 0.05.

To investigate whether these detectedmetabolites improve risk prediction, weadded the weighted multimarker scorecomprising all seven metabolites to thefully adjusted statistical model. Resultsshow that addition of the metabolitescore resulted in significant improve-ment for diabetes risk prediction as as-sessed by all three measures: the AUCvalue increased from 0.763 to 0.822(P = 0.006), the NRI was 0.623 (95% CI0.427–0.819; P, 1025), and the IDI was0.117 (95% CI 0.083–0.151; P , 1025).This indicates that the newly detectedmetabolic markers significantly improve

risk prediction of T2D beyond estab-lished diabetes risk factors. The fivematching known metabolites belong tothe classes of glycerophosphocholine,flavonoids, and polypeptides (Supple-mentary Table 2). Partial correlations ofthese matching metabolites with clinicalrisk factors are shown in SupplementaryTable 3. Apart from some weak correla-tions of 2HBP with fasting insulin or in-sulin resistance, PC (22:6/20:4) with BMI,or LDYR with lipid levels, most metabo-lites were not correlated with estab-lished diabetes factors. The matchingmetabolites HPMF, MEIR, and the

Table 1—Characteristics of the study participants at baseline (2001–2003)

Participants whodeveloped T2D

Participants who didnot develop T2D P*

n 133 298

Age, years 35.45 6 12.2 33.36 6 13.88 0.1208

Female sex, % 67.67 63.42 0.3885

BMI, kg/m2 36.74 6 7.96 31.11 6 8.00 ,0.0001

Current smoker, % 33.83 36.58 0.7266

Current drinker, % 63.16 68.79 0.5034

Systolic blood pressure, mmHg 120.88 6 15.34 118.87 6 12.96 0.1868

Diastolic blood pressure, mmHg 77.39 6 11.80 75.63 6 10.46 0.1222

HDL, mg/dL 47.52 6 14.41 52.44 6 14.63 0.0016

LDL, mg/dL 100.92 6 29.32 96.06 6 28.57 0.1062

Total triglyceride, mg/dL 167.20 6 99.12 132.16 6 65.47 ,0.0001

Total cholesterol, mg/dL 180.70 6 34.16 174.75 6 33.48 0.0923

eGFR, mL/min/1.73 m2 104.56 6 21.41 105.18 6 24.84 0.7917

Fasting glucose, mg/dL 94.30 6 7.81 89.55 6 6.41 ,0.0001

Fasting insulin, mU/mL 20.52 6 13.08 14.14 6 11.47 0.0001

Insulin resistance (HOMA-IR) 4.80 6 3.07 3.15 6 2.60 ,0.0001

Total caloric intake, kcal/day 2,887.59 6 2,079.25 2,812.91 6 2,117.20 0.7409

Total dietary protein, g/day 97.51 6 82.98 94.99 6 81.77 0.7768

Total dietary fat, g/day 126.39 6 99.66 123.71 6 98.08 0.8017

Data are mean6 SD unless otherwise indicated. *Adjusting for family relatedness by generalizedestimating equation.

Table 2—Multivariate association of baseline fasting plasma metabolites with riskof developing T2D in American Indians by Cox proportional hazards frailtymodel‡

Matching metabolitesMetabolite as continuous

variable*Metabolite as categorical

variable†*

Protective metabolitesPC (22:6/20:4) 0.68 (0.52–0.88) 0.45 (0.21–0.97)HPMF 0.58 (0.43–0.79) 0.38 (0.18–0.80)MEIR 0.61 (0.47–0.78) 0.44 (0.20–0.96)LDYR 0.63 (0.47–0.85) 0.37 (0.16–0.87)X-490 0.65 (0.50–0.84) 0.46 (0.21–0.97)Combined protective effects 0.43 (0.31–0.59) 0.23 (0.10–0.51)

Risk metabolites2HBP 1.80 (1.26–2.57) 2.80 (1.19–6.60)X-1178 1.89 (1.29–2.77) 2.87 (1.08–7.60)Combined risk effects 2.56 (1.71–3.84) 6.89 (2.63–18.08)

Data are HR (95% CI). ‡Adjusted for age, sex, site, BMI, eGFR, HDL, triglycerides, fasting glucose,and HOMA-IR. *HR per SD change in log-transformed metabolite level. †Tertile 3 vs. tertile 1.

care.diabetesjournals.org Zhao and Associates 223

http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc14-2033/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc14-2033/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc14-2033/-/DC1http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc14-2033/-/DC1http://care.diabetesjournals.org

unknown compound (X-490) were notcorrelated with any of the known riskfactors for diabetes.To identify metabolic profiles associ-

ated with risk of diabetes development,we conducted sPLS-DA using the sevenmetabolites that were significantly pre-dictive of disease risk. Fig. 2 demonstratesthat participants who developed T2D andthose who did not were separated intotwo distinct groups, suggesting that thesemetabolites could be used as discrimina-tory markers for T2D risk stratification.This observation is consistent with ourresults obtained by risk prediction analy-ses (i.e., AUC, NRI, and IDI). Additionaladjustments for dietary intake of fat, pro-tein, and caloric intake did not attenuatethe observed associations (data notshown).

CONCLUSIONS

In this prospective investigation usingan untargeted high-resolution metabolo-mic approach, we found that seven me-tabolites independently predict futureonset of T2D in American Indians, a groupwith a high rate of diabetes. Of the fivechemicals matching known metabolites,two were lipids in the classes of glycero-phosphocholine (PC) and flavonoid. Itshould be noted that there are manyisobaric lipids, so the precise structuralidentifications will require additional re-search. Theobserved associationwithstoodadjustments for multiple clinical indica-tors including age, sex, study site, BMI,eGFR, HDL, triglycerides, fasting glucose,

and insulin resistance (HOMA-IR). Thecombination of these metabolites signif-icantly improves risk prediction beyondestablished diabetes risk factors. Thesemetabolites have not been reported inprevious studies of European individu-als or other ethnic groups and thusshould represent putative prognostic

markers of diabetes specific to Ameri-can Indians.

We found that a metabolite matching2HBP was associated with 80% in-creased risk of developing T2D inde-pendent of classical risk factors. Themechanism by which this metabolite af-fects diabetes risk is unclear. However,

Figure 1—Manhattan plot of 11,628m/z features comparing participants who developed incident T2D versus those who did not. The negative log Pvalue was plotted against the m/z features. The x-axis represents m/z of the detected features, ordered in increasing value from 85 (left) to 1,800(right). A total of seven metabolites significantly differed between the two groups at the threshold of q = 0.05 (above the horizontal gray line).

Figure 2—Separation of study participants who developed incident T2D and those who did notduring follow-up by sPLS-DA using a multimarker metabolite score comprising all seven metab-olites showing significant associations with incident T2D listed in Table 2.

224 New Metabolic Markers for Diabetes Risk Diabetes Care Volume 38, February 2015

2HBP is known to be an environmentaltoxin that is widely used as industrialantimicrobials, agricultural fungicide,and disinfectants. 2HBP was reportedto be mutagenic in human cells (31)and carcinogenic in animal models(32,33). In addition, hydroxybiphenylchemicals can be degraded by bacteriathrough the biphenyl catabolic pathway(34). It is thus plausible to hypothesizethat, apart from the possible direct toxiceffects of 2HBP on pancreas or periph-eral tissues, 2HBP may also negativelyaffect diabetes through a yet unknownhost-gut microbiota mechanism.Glycerophosphocholines are impor-

tant structural components of plasmalipoproteins and cell membranes withdiverse biological functions. In thisstudy, we found that elevated plasmalevel of matching PC (22:6/20:4) wasassociated with 37% reduced risk ofT2D in our study population. This is inagreement with a previous study dem-onstrating lowerplasmaor serum levels ofPC species in diabetic patients than in con-trol subjects (5). Moreover, reduced levelsof multiple acyl-glycerophosphocholinespecies were highly correlated with in-sulin resistance as measured by the eu-glycemic clamp (35), lending furthersupport for a potential role of PCs in di-abetes etiology. In the current investi-gation, another metabolite matchingknown (3S)-7-hydroxy-29,39,49,59,8-pentamethoxyisoflavan (named HPMF)was also significantly predictive of a de-creased risk of diabetes. This metabolitebelongs to the class of flavonoids thatare known to have a wide range ofbiological and pharmacological activi-ties. Dietary flavonoid intakes havebeen associated with reduced risk ofT2D in both human (36–38) and animalstudies (39). In support of these findings,participants with a higher plasma level(top tertile) of HPMF exhibited over60% reduced risk of T2D compared withthosewith a lower level (bottom tertile) inour analysis. While the precise mecha-nism underlying this association awaitsfurther investigation, it is possible thatHPMFmay decrease diabetes risk throughits potential antioxidant properties (40). Itis also likely that HPMF may exert benefi-cial effects on energy balance and lipidmetabolism (41) or anti-inflammatory ef-fects through the nuclear factor-kB or theAMPK signaling pathways, which play acentral role in the regulation of glucose

and lipid metabolism (42,43). In addition,flavonoids have been shown to have anti-diabetes effects through enhancedpancreatic b-cell function in animal ex-periments (44). The favorable effect ofthis flavonoid chemical has not beenpreviously reported. Its biological prop-erties should be investigated in futureresearch.

In addition to the altered profiles ofPC and flavonoid, elevated levels of twometabolites matching tetrapeptides(MEIR and LDYR) were associated with;40% reduced risk of diabetes. Al-though the mechanisms linking thesepeptides to diabetes remain to be de-termined, peptides are known to be es-sential in regulating lipid metabolism inkey insulin-target tissues and in main-taining energy homeostasis and insulinsensitivity. They may also function aspotent peptide hormones regulatingglucose metabolism in diabetes (45). Inaddition to the five knownmatchingme-tabolites, two unknown compoundswere also significantly predictive of di-abetes development. These unknownchemicals might be not new but merelynot yet identified. The structure andfunction of these unannotated chemicalsshould be examined in future research.

Previous evidence has linked raisedcirculating levels of BCAAs with insulinresistance (2,46,47) or diabetes (10,47).Our study, however, did not find a sig-nificant association of BCAAs with risk ofT2D development. This lack of replica-tion may not necessarily representtrue negative findings because our anal-ysis accounted for multiple testing of.11,000 m/z features with a stringentcriterion, which could result in inappro-priate exclusion for a large number ofmetabolites (false negatives). The dis-crepancy could also represent genuinedifference between American Indiansand other ethnic groups included in pre-vious studies because the unique char-acteristics of American Indians, e.g.,genetic background and lifestyle, couldpotentially lead to population-specificmetabolic signatures. Future large-scalemetabolomics studies should addressthis discrepancy.

In search of the origin of the interin-dividual variation, we calculated partialcorrelations of metabolite relativeabundance with standard risk indicatorsof diabetes, expecting that, for example,higher BMI or fasting glucose should

correspondwith higher levels of riskme-tabolites or lower levels of protectivemetabolites. However, in our study co-hort, most of the detected matchingmetabolites were not correlated withclassical risk factors, such as BMI, fastingglucose, and insulin resistance, but thecombination of thesemetabolites signif-icantly improved risk prediction beyondstandard risk factors. This is importantbecause the fundamental task of risk pre-diction is to identify predictive markersthat are sufficiently uncorrelated with es-tablished risk factors so that they can beused to improve risk prediction over andabove conventional clinical factors. Thesenewly detected metabolic markers willprovide valuable information regardingthe pathophysiology of diabetes develop-ment and also potential therapeutic tar-gets for novel treatment options.

Our study has several limitations. First,although our high-resolution LC-MS de-tected .11,000 distinct features, itshould be noted that only 18% of thecompounds detected had a match in thecurrent metabolomics database. Thesecompounds were unable to be pursuedowing to the large number of possibleisomers and a lack of available standards.However, these currently unannotatedmetabolites may represent dietary,microbiome-related, or environmentalchemicals associated with diabetes.With the advancement of metabolomicresearch, we expect that the majority ofthese unidentified chemicals will ulti-mately be annotated and their associa-tions with disease will be determined.Additionally, manym/z features matchedto therapeutic drugs and nutritional sup-plements, but owing to their wide use bydiabetic patients, wewere unable to eval-uate their contributions to the alteredmetabolic profiles. Second, althoughhighly correlated, relative abundancesbut not absolute concentrations wereused as a surrogate for plasmametabolitelevels. Third, although we were able tocontrol many of the known risk factors,the possibility of potential confoundingby other factors such as diet and gut mi-crobiota cannot be entirely excluded.Fourth, participants in the current studyare young to middle-aged American Indi-ans who may have a high propensity forthe development of T2D; therefore, gen-eralization of our findings to other popu-lations should be approached cautiously.However, given the rising tide of T2D in

care.diabetesjournals.org Zhao and Associates 225

http://care.diabetesjournals.org

almost all ethnic groups worldwide, webelieve that our results could be applica-ble to other populations. Finally, our re-sults need to be replicated in large-scale,prospective metabolomic analysis ofAmerican Indians and other ethnicgroups.Nonetheless, this is the first prospec-

tive study to report novel predictive met-abolic markers and altered metabolicprofiles associated with development ofT2D in American Indians, aminority groupsuffering from a disproportionately highrate of T2D. The SHFS has phenotypic lon-gitudinal data available that allowed usto accurately classify participants as inci-dent cases of diabetes. The untargetedhigh-resolution metabolomics approachallowed us to identify previously unde-scribed metabolic markers that may bespecific to the population of American In-dians, whose genetic makeup and/or life-style could be distinct from that ofindividuals of European ancestry.In summary, this study identified sig-

nificant metabolic predictors of T2D inAmerican Indians above and over estab-lished diabetes indicators. Targeting bi-ological pathways that involve thesenewly detected metabolites wouldhelp to develop early preventive andtherapeutic strategies tailored to Amer-ican Indians, an ethnically important buttraditionally understudied minoritypopulation.

Acknowledgments. The authors thank theSHFS participants, Indian Health Service facili-ties, and participating tribal communities fortheir extraordinary cooperation and involve-ment, which has contributed to the success ofthe SHFS.Funding. This study was supported by NationalInstitutes of Health grants R01DK091369,K01AG034259, and R21HL092363 and coopera-tive agreement grants U01HL65520, U01HL41642,U01HL41652, U01HL41654, and U01HL65521.

The views expressed in this article are those ofthe authors and do not necessarily reflect thoseof the Indian Health Service.Duality of Interest. No potential conflicts ofinterest relevant to this article were reported.Author Contributions. J.Z. conceived thestudy, supervised the statistical analyses, andwrote the manuscript. Y.Z., N.H., and D.Z.conducted statistical analyses. K.U. and V.T.T.collected LC-MS data and conducted metabolo-mic analyses. T.Y. and D.J. supervised metab-olomic data analyses. J.H., E.T.L., and B.V.H.contributed to study design, data interpreta-tion, and discussion and reviewed and editedthe manuscript. J.Z. is the guarantor of thiswork and, as such, had full access to all of thedata in the study and takes responsibility for

the integrity of the data and the accuracy of thedata analysis.

References1. Defronzo RA. Banting Lecture. From the tri-umvirate to the ominous octet: a new paradigmfor the treatment of type 2 diabetes mellitus.Diabetes 2009;58:773–7952. Newgard CB, An J, Bain JR, et al. Abranched-chain amino acid-related metabolicsignature that differentiates obese and lean hu-mans and contributes to insulin resistance. CellMetab 2009;9:311–3263. Würtz P, Mäkinen VP, Soininen P, et al.Metabolic signatures of insulin resistance in7,098 young adults. Diabetes 2012;61:1372–13804. Wopereis S, Rubingh CM, van Erk MJ, et al.Metabolic profiling of the response to an oralglucose tolerance test detects subtle metabolicchanges. PLoS ONE 2009;4:e45255. Suhre K, Meisinger C, Döring A, et al. Met-abolic footprint of diabetes: a multiplatformmetabolomics study in an epidemiological set-ting. PLoS ONE 2010;5:e139536. Adams SH, Hoppel CL, Lok KH, et al. Plasmaacylcarnitine profiles suggest incomplete long-chain fatty acid beta-oxidation and altered tri-carboxylic acid cycle activity in type 2 diabeticAfrican-American women. J Nutr 2009;139:1073–10817. Fiehn O, Garvey WT, Newman JW, Lok KH,Hoppel CL, Adams SH. Plasma metabolomicprofiles reflective of glucose homeostasis innon-diabetic and type 2 diabetic obese African-American women. PLoS ONE 2010;5:e152348. Wang-Sattler R, Yu Z, Herder C, et al. Novelbiomarkers for pre-diabetes identified by me-tabolomics. Mol Syst Biol 2012;8:6159. Menni C, Fauman E, Erte I, et al. Biomarkersfor type 2 diabetes and impaired fasting glucoseusing a nontargeted metabolomics approach.Diabetes 2013;62:4270–427610. Wang TJ, Larson MG, Vasan RS, et al.Metabolite profiles and the risk of developingdiabetes. Nat Med 2011;17:448–45311. Floegel A, Stefan N, Yu Z, et al. Identifi-cation of serummetabolites associatedwith riskof type 2 diabetes using a targetedmetabolomicapproach. Diabetes 2013;62:639–64812. Ferrannini E, Natali A, Camastra S, et al.Early metabolic markers of the development ofdysglycemia and type 2 diabetes and their phys-iological significance. Diabetes 2013;62:1730–173713. North KE, Howard BV, Welty TK, et al.Genetic and environmental contributions tocardiovascular disease risk in American Indians:the strong heart family study. Am J Epidemiol2003;157:303–31414. Lee ET, Welty TK, Fabsitz R, et al. TheStrong Heart Study. A study of cardiovasculardisease in American Indians: design and meth-ods. Am J Epidemiol 1990;132:1141–115515. Expert Committee on the Diagnosis andClassification of Diabetes Mellitus. Report ofthe Expert Committee on the Diagnosis andClassification of Diabetes Mellitus. DiabetesCare 2003;26(Suppl. 1):S5–S2016. Clauss A. Rapid physiological coagulationmethod in determination of fibrinogen. ActaHaematol 1957;17:237–246 [in German]

17. Matthews DR, Hosker JP, Rudenski AS,Naylor BA, Treacher DF, Turner RC. Homeostasismodel assessment: insulin resistance and beta-cell function from fasting plasma glucose and in-sulin concentrations in man. Diabetologia 1985;28:412–41918. Levey AS, Bosch JP, Lewis JB, Greene T,Rogers N, Roth D; Modification of Diet in RenalDisease Study Group. A more accurate methodto estimate glomerular filtration rate from se-rum creatinine: a new prediction equation. AnnIntern Med 1999;130:461–47019. Willett W. Food frequency methods. InNutritional Epidemiology. 2nd ed. New York,Oxford University Press, 1998, p. 74–9120. Osborn MP, Park Y, Parks MB, et al.Metabolome-wide association study of neovas-cular age-related macular degeneration. PLoSONE 2013;8:e7273721. Roede JR, Uppal K, Park Y, et al. Serummetabolomics of slow vs. rapid motor progres-sion Parkinson’s disease: a pilot study. PLoSONE2013;8:e7762922. Yu T, Park Y, Johnson JM, Jones DP.apLCMSdadaptive processing of high-resolutionLC/MS data. Bioinformatics 2009;25:1930–193623. Yu T, Park Y, Li S, Jones DP. Hybrid featuredetection and information accumulation usinghigh-resolution LC-MS metabolomics data.J Proteome Res 2013;12:1419–142724. Uppal K, Soltow QA, Strobel FH, et al.xMSanalyzer: automated pipeline for improvedfeature detection and downstream analysis oflarge-scale, non-targeted metabolomics data.BMC Bioinformatics 2013;14:1525. Smith CA, O’Maille G, Want EJ, et al.METLIN: a metabolite mass spectral database.Ther Drug Monit 2005;27:747–75126. Wishart DS, Jewison T, Guo AC, et al.HMDB 3.0–The Human Metabolome Databasein 2013. Nucleic Acids Res 2013;41:D801–D80727. Sud M, Fahy E, Cotter D, et al. LMSD: LIPIDMAPS structure database. Nucleic Acids Res2007;35:D527–D53228. Hughes G, Cruickshank-Quinn C, Reisdorph R,et al. MSPrep–summarization, normalization anddiagnostics for processing of mass spectrometry-based metabolomic data. Bioinformatics 2014;30:133–13429. Johnson WE, Li C, Rabinovic A. Adjustingbatch effects in microarray expression data us-ing empirical Bayes methods. Biostatistics 2007;8:118–12730. Storey JD. A direct approach to false dis-covery rates. J R Stat Soc Series B Stat Methodol2002;64:187–20531. Suzuki H, Suzuki N, Sasaki M, Hiraga K.Orthophenylphenol mutagenicity in a humancell strain. Mutat Res 1985;156:123–12732. Brusick D. Analysis of genotoxicity and the car-cinogenic mode of action for ortho-phenylphenol.Environ Mol Mutagen 2005;45:460–48133. HagiwaraA, ShibataM,HiroseM, FukushimaS,Ito N. Long-term toxicity and carcinogenicitystudy of sodium o-phenylphenate in B6C3F1mice. Food Chem Toxicol 1984; 22:809–81434. Sondossi M, Sylvestre M, Ahmad D,Masse R.Metabolism of hydroxybiphenyl and choloro-hydroxybiphenyl by biphenyl/cholorobiphenyl de-gradign Pseudomonas testosteroni, strain B-356.J Ind Microbiol 1991;6:77–88

226 New Metabolic Markers for Diabetes Risk Diabetes Care Volume 38, February 2015

35. Gall WE, Beebe K, Lawton KA, et al.; RISCStudy Group. alpha-Hydroxybutyrate is an earlybiomarker of insulin resistance and glucose in-tolerance in a nondiabetic population. PLoSONE 2010;5:e1088336. vanDam RM,Naidoo N, Landberg R. Dietaryflavonoids and the development of type 2 diabe-tes and cardiovascular diseases: review of recentfindings. Curr Opin Lipidol 2013;24:25–3337. Wedick NM, Pan A, Cassidy A, et al. Dietaryflavonoid intakes and risk of type 2 diabetes in USmen and women. Am J Clin Nutr 2012;95:925–93338. Zamora-Ros R, Forouhi NG, Sharp SJ, et al.The association between dietary flavonoid andlignan intakes and incident type 2 diabetes inEuropean populations: the EPIC-InterAct study.Diabetes Care 2013;36:3961–397039. Chen YK, Cheung C, Reuhl KR, et al. Effects ofgreen tea polyphenol (-)-epigallocatechin-3-gallate

on newly developed high-fat/Western-stylediet-induced obesity and metabolic syndromein mice. J Agric Food Chem 2011;59:11862–1187140. Lotito SB, Zhang WJ,Yang CS,Crozier A, Frei B.Metabolic conversion of dietary flavonoids al-ters their anti-inflammatory and antioxidantproperties. Free Radic BiolMed 2011;51:454–46341. Friedrich M, Petzke KJ, Raederstorff D,Wolfram S, Klaus S. Acute effects of epigalloca-techin gallate from green tea on oxidation andtissue incorporation of dietary lipids in mice fed ahigh-fat diet. Int J Obes (Lond) 2012;36:735–74342. Salminen A, Hyttinen JM, Kaarniranta K.AMP-activated protein kinase inhibits NF-kBsignaling and inflammation: impact on health-span and lifespan. J Mol Med (Berl) 2011;89:667–67643. Leiherer A, Mündlein A, Drexel H. Phyto-chemicals and their impact on adipose tissue

inflammation and diabetes. Vascul Pharmacol2013;58:3–2044. Ortsäter H,Grankvist N,Wolfram S, Kuehn N,Sjöholm A. Diet supplementation with green teaextract epigallocatechin gallate prevents progres-sion to glucose intolerance in db/db mice. NutrMetab (Lond) 2012;9:1145. Todd JF, Bloom SR. Incretins and other pep-tides in the treatment of diabetes. Diabet Med2007;24:223–23246. Huffman KM, Shah SH, Stevens RD, et al.Relationships between circulating metabolic in-termediates and insulin action in overweight toobese, inactive men and women. Diabetes Care2009;32:1678–168347. Stancáková A, Civelek M, Saleem NK, et al.Hyperglycemia and a common variant of GCKRare associated with the levels of eight aminoacids in 9,369 Finnish men. Diabetes 2012;61:1895–1902

care.diabetesjournals.org Zhao and Associates 227

http://care.diabetesjournals.org