Circulating Modified Metabolitesand a Risk of ESRD in PatientsWith Type 1 Diabetes and ChronicKidney DiseaseDiabetes Care 2017;40:383–390 | DOI: 10.2337/dc16-0173

OBJECTIVE

Patients with type 1 diabetes (T1D) with impaired renal function are at increasedrisk for end-stage renal disease (ESRD). Although the rate of progression varies,determinants and mechanisms of this variation are unknown.

RESEARCH DESIGN AND METHODS

We examined serummetabolomic profiles associated with variation in renal func-tion decline in participants with T1D (the Joslin Kidney Study prospective cohort).One hundred fifty-eight patients with proteinuria and chronic kidney diseasestage 3were followed for amedian of 11 years to determine estimated glomerularfiltration rate slopes from serial measurements of serum creatinine and to ascer-tain time to onset of ESRD. Baseline serum samples were subjected to globalmetabolomic profiling.

RESULTS

One hundred ten amino acids and purine and pyrimidine metabolites were de-tected in at least 80% of participants. Serum levels of seven modified metabolites(C-glycosyltryptophan, pseudouridine, O-sulfotyrosine, N-acetylthreonine,N-acetylserine, N6-carbamoylthreonyladenosine, and N6-acetyllysine) were asso-ciated with renal function decline and time to ESRD (P < 0.001) independent of therelevant clinical covariates. The significant metabolites correlated with one an-other and with the indices of tubular injury.

CONCLUSIONS

This prospective cohort study in participants with T1D, proteinuria, and impairedrenal function at baseline demonstrated that patients with increased circulatinglevels of certain modified metabolites experience faster renal function decline,leading to ESRD. Whether some of these candidate metabolites are risk factors orjust prognostic biomarkers of progression to ESRD in T1D needs to be determined.

Progressive renal decline is the clinical manifestation of the disease process underlyingthe development of diabetic nephropathy (1,2). Although patients with chronic kidneydisease (CKD) are at high risk for end-stage renal disease (ESRD), this risk varies tre-mendously. Some patients with CKD progress rapidly, whereas others may progressover a decade or more. In the absence of diabetic animal models that develop renalfailure, observational studies in humans are valuable to search for newpathways to thedevelopment of drugs that would slow renal function decline and postpone ESRD.

1Research Division, Joslin Diabetes Center, Bos-ton, MA2Department of Medicine, Harvard MedicalSchool, Boston, MA3Division of Nephrology, Department of InternalMedicine, University of Michigan, Ann Arbor, MI4Renal Division, Brigham andWomen’s Hospital,Boston, MA

Corresponding author: Monika A. Niewczas,monika.niewczas@joslin.harvard.edu.

Received 27 January 2016 and accepted 17December 2016.

This article contains Supplementary Data onlineat http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc16-0173/-/DC1.

© 2017 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 thework is notaltered.More infor-mation is available at http://www.diabetesjournals.org/content/license.

Monika A. Niewczas,1,2 Anna V. Mathew,3

Stephanie Croall,1 Jaeman Byun,3

Melissa Major,1 Venkatta S. Sabisetti,4

Adam Smiles,1 Joseph V. Bonventre,2,4

Subramaniam Pennathur,3 and

Andrzej S. Krolewski1,2

Diabetes Care Volume 40, March 2017 383

PATH

OPHYSIO

LOGY/COMPLIC

ATIO

NS

The mechanisms responsible for thisvariation in patients with diabetes arestudied with the use of newly developedhigh throughput omics technologies(3–5). Among these, metabolomic plat-forms have been used to search for newmetabolic pathways (3,6–9). Our globalprofiling study in type 2 diabetes (T2D)demonstrated that certain metabolitescan be predictors of ESRD that occurs10 years later independent of legacymarkers such as albuminuria and baselinerenal function (3). Todate, no study toourknowledge has examined metabolomicprofiles of patients with type 1 diabetes(T1D) who progress to ESRD, the ultimateoutcome of diabetic nephropathy.

RESEARCH DESIGN AND METHODS

The Committee on Human Subjects ofthe Joslin Diabetes Center approved theinformed consent and recruitment andexamination procedures for this study.

Study GroupApproximately 3,500 adult patients withT1D diagnosed before age 40 receivecare at the Joslin clinic. Between1991 and 2004, we monitored this pop-ulation for the occurrence of persistentproteinuria by using routinely measuredalbumin-to-creatinine ratios (ACRs). Pa-tients with persistent proteinuria wereinvited to participate in a follow-upstudy on the genetics of diabetic ne-phropathy (2,10). Eligibility criteria in-cluded residence in New England andage at enrollment between 21 and54 years. In that interval, persistent pro-teinuriawas diagnosed in 790 eligible pa-tients, and 564 (71%) participated in theJoslin Proteinuria Cohort Study. At studyenrollment, a standardized baseline ex-amination was performed by trained re-cruiters who administered a structuredinterview and brief physical examina-tion. Blood and urine specimens col-lected at the baseline and follow-upexaminations were stored at285°C untilanalysis (2,10).Enrolled patients were followed until

2012. Blood and urine specimens werecollected at least every 2 years. Collec-tion of research specimens occurredduring routine clinic visits. Patientswith less frequent clinic visits were ex-amined at their homes.Among the patients enrolled in the

Joslin Proteinuria Cohort Study, 199 hadCKD of whom 158 were included in the

current study. Of note, selection of thestudy participants was based purely ontheir baseline exposures andwas indepen-dent from their follow-up characteristics.

Assessment of Abnormalities inUrinary Albumin ExcretionIn the Joslin clinical laboratory, albuminconcentrations in spot urines are rou-tinely measured at least once a year.Between 1991 and 2009, urinary albuminconcentrations were measured by immu-nonephelometry on a BN ProSpec Sys-tem (Dade Behring, Newark, DE) withN-albumin kits. Creatininemeasurementsin urine were assayed by the Jaffe modi-fied picrate method on a Ciba Corning550 Express Plus Chemistry Analyzer.For each patient, the geometric meanACR for the preceding 2-year interval ofthe baseline examination was deter-mined to assign an albuminuria status:normoalbuminuria (ACR ,30 mg/g cre-atinine), microalbuminuria (ACR 30–300 mg/g creatinine), and proteinuria(.300 mg/g creatinine) (11).

Assessment of Renal FunctionFrom 2011 to 2014, serum specimensobtained for research purposes at base-line (n = 520) and during follow-up (n =1,560) in participants in the Joslin Pro-teinuria Cohort Study were used tomeasure creatinine concentrations.The measurements were performed inthe Advanced Research and DiagnosticLaboratory at the University of Minne-sota by using the Roche enzymatic assay(product no. 11775685) on a Roche/Hitache Mod P analyzer. This method iscalibrated to be traceable to an isotopedilution mass spectrometry referenceassay and was verified by measuringNational Institute of Standards andTechnology SRM 967.

In addition, we retrieved 2,900 mea-surements of baseline serum creatinineconcentrations (n = 520) from Joslin clin-ical laboratory records. These mea-surements were assayed by the Jaffemodified picrate method as describedabove. For 633 samples, we used dupli-cate measurements to calibrate theJoslin clinical measurements to the assaymethod used at the University of Minne-sota laboratory. The CKD EpidemiologyCollaboration equation was used todetermine the estimated glomerular fil-tration rate (eGFR) (12). For each partici-pant, we used all available serial eGFR

determinations performed during follow-up to determine eGFR slopes by generallinear model procedure.

Ascertainment of Onset of ESRDAll participants from the Joslin Protein-uria Cohort Study included in the cur-rent study were queried against rostersof the U.S. Renal Data System (USRDS)and the National Death Index (NDI), cov-ering all events up to the end of 2012.USRDS maintains a roster of U.S. pa-tients who received renal replacementtherapy, which includes dates of dialysisand transplantation (13). The NDI is acomprehensive roster of dates andcauses of deaths in the U.S. (14). ESRDwas defined by a match with the USRDSroster or a listing of renal failure amongthe causes of death on an NDI deathcertificate.

Global Metabolomic ProfilingBaseline serum specimens were sub-jected to global profiling (Metabolon,Durham, NC). The Metabolon platformuses gas chromatography and liquidchromatography mass spectrometry inpositive and negative modes (15,16).The liquid chromatography mass spec-trometry portion of the platformincorporates a Waters ACQUITY UPLCsystem and a Finnigan LTQ mass spec-trometer (Thermo Fisher Scientific), in-cluding an electrospray ionizationsource and linear ion trap mass analyzer.The gas chromatography column is 5%phenyl dimethyl silicone, and the tem-perature ramp is from 40°C to 300°Cover 16 min. All samples were thenanalyzed on a Finnigan TRACE DSQfast-scanning single-quadrupole massspectrometer (Thermo Fisher Scientific)using electron impact ionization. Afterpeak identification and quality controlfiltering, the metabolite-relative con-centrations were obtained from me-dian-scaled day-block normalized datafor each compound. Samples in thisstudy were run in batches balanced bycase status.

Repeated Metabolomic MeasurementsOver TimeIn the study group, we identified20 participants with a stable diabeticnephropathy phenotype over time(eGFR slope ,2.5 mL/min/1.73 m2 peryear, ACR 6200 mg/g creatinine peryear) for whom we performed globalmetabolomics measurements in serum

384 Modified Metabolites and ESRD Risk in Diabetes Diabetes Care Volume 40, March 2017

samples taken at baseline and 3–8 yearslater. Global profiling was performedaccording to the protocol describedabove. Samples from the same individu-als were run on the same batch.

Comparison of Findings in OtherPublished Reports With Our GlobalMetabolomics DataWe analyzed amino acids and purineand pyrimidine metabolites reportedto be associated with diabetic nephrop-athy or eGFR decline that were also welldetected in our main study by globalmetabolomic profiling (3,6,8,9,17). Weevaluated their associations with eGFRslope and with progression to ESRD inthe main study.

Measurements of Glycemic Controland Urinary Glomerular and TubularProtein MarkersHbA1c wasmeasured in the Joslin clinicallaboratory according to themethod cali-brated to Diabetes Complications andControl standards. In addition to ACR,three other proteins were measured inbaseline urine specimens. IgG2 wasmeasured in the Joslin research labora-tory as previously described (18). Neu-trophil gelatinase-associated lipocalin(NGAL) was measured in the Joslin re-search laboratory by Quantikine ELISA(DLCN20; R&D Systems, Minneapolis,MN) according to the manufacturer’sprotocol. KIM-1 was measured in the

laboratory of J.V.B. (Brigham andWomen’sHospital) (19,20). Measurements ofprotein-based markers (IgG2, KIM-1,and NGAL) were available for 91 studyparticipants. The interassay coefficientof variation was ,20% for all analytes.All the urine specimens for marker mea-surements had the same number offreeze-thaw cycles.

Auxiliary Study of Renal ClearanceAmong 365 Joslin Proteinuria Cohort Studyparticipants who had normal renal functionat baseline (eGFR .60 mL/min/1.73 m2),21 who had progressed to ESRD were se-lected and matched by age, sex, and base-line renal functionwith 32participantswhohadaneGFR,23.3mL/min/1.73m2 (non-progressors) (2,11). Measurements of themetabolites were performed according toa similar protocol, the global profiling usedin themain study. Mass spectra were com-pared with the library of ;1,000 purifiedstandards for their analytical identification,whereas measurements performed in themain study were compared with thelibrary of 2,400 standards.

Data AnalysisDifferences in clinical characteristics weretested byANOVA for continuous variablesand x2 test for categorical variables. Me-tabolites detected by global profiling in atleast 80% of the study participants wereconsidered in further analysis. Correla-tions of the metabolites with baseline

eGFR, other baseline measures, and sub-sequent eGFR slopes were expressed asSpearman rank correlation coefficients.Adjustment for multiple comparisonswas performed with a positive false dis-covery rate (q-value ,0.05 for signifi-cance). An independent effect of ametabolite for the progression to ESRDwas estimated by using a Cox propor-tional hazards regression model. Aftertransformation of themetabolite concen-tration to normally distributed ranks, itseffect on risk of progression was ex-pressed as the hazard ratio (HR) per1 SD difference (3). Discrimination abil-ities of the risk prediction models wereevaluated with C statistics (21). Principalcomponent (PC) analysis was used as adata reduction approach to create a me-tabolite index. Reproducibility of metab-olomic measurements over time wasdetermined by intraclass correlation coef-ficients (ICCs), which represent ratios ofbetween-person variance to the sum ofthe within- and between-person vari-ances (modified ICC_SAS macro). Dataanalysis was performed with SAS 9.4and JMP Pro 12.0.0 software (SAS Insti-tute, Cary, NC).

RESULTS

Characteristics of the Study GroupFive hundred sixty-four patients attend-ing the Joslin clinic were recruited intothe Joslin Proteinuria Cohort Study. Of

Table 1—Clinical characteristics of the study group stratified by the rate of the renal function decline

Renal Function Decline

CharacteristicMinimal(n = 53)

Moderate(n = 53)

Fast(n = 52) P value

At baselineMale [n (%)] 25 (47) 28 (53) 29 (46) 0.38Age (years) 45 6 10 42 6 8 40 6 10 0.003Duration of diabetes (years) 30 6 8 29 6 8 27 6 9 0.007HbA1c (%) 8.5 6 1.3 9.3 6 1.6 9.6 6 1.7 0.005HbA1c (mmol/mol) 69 6 14 78 6 18 81 6 19 0.005Serum cholesterol (mg/dL) 191 6 43 219 6 63 219 6 71 0.03Systolic blood pressure (mmHg) 134 6 20 136 6 19 137 6 24 0.57Diastolic blood pressure (mmHg) 75 6 11 79 6 11 80 6 11 0.07Antihypertensive/renoprotective treatment (%) 93 92 80 0.07ACR (mg/g creatinine) 670 (403, 1,009) 1,007 (680, 1,934) 1,533 (748, 3,526) 0.0004eGFR (mL/min/1.73 m2) 50 6 12 48 6 13 44 6 13 0.02

Follow-upLength (years) 14.4 (9.7, 18.4) 11.8 (7.7, 13.8) 9.2 (5.6, 12.1) ,0.001Annual eGFR slope (mL/min/1.73 m2) 21.9 (21.0 to 22.5) 25.0 (23.9 to 25.9) 210.9 (28.5 to 215.6) By designIncidence of ESRDCases (n) 16 35 48 ,0.001Rate per 1,000 person-years* 1.1 (0.6, 1.9) 3.7 (2.6, 5.1) 10.5 (7.7, 13.9) ,0.001

Data are mean 6 SD or median (25th, 75th percentile) unless otherwise indicated. *Incidence rate of ESRD is accompanied by 95% CI.

care.diabetesjournals.org Niewczas and Associates 385

these patients, 158 with eGFR 30–60 mL/min/1.73 m2 at enrollmentwere selected for this study. Medianfollow-up for this study was 11.5 years.Follow-up for participants in whomESRD did not develop was 4.4–21 yearsand was understandably shorter (2.4–17.3 years) for participants in whomESRD did develop. During this time,ESRD developed in 99 participants(63%). Median eGFR slope was 25.0(28.4 to 22.5) mL/min/1.73 m2 peryear. Participants with more-rapideGFR decline were younger and had di-abetes for a shorter period, worse gly-cemic control, and higher albuminuria.They differed slightly with regard totheir baseline eGFR. Clinical character-istics of the study group according totertiles of eGFR slope distribution aresummarized in Table 1.

Metabolomic Profiles Associated WithVariation in eGFR Slope

Multivariate Analysis

Global metabolomic profiling resulted inthe detection of mass spectra of 123 me-tabolites. Of those, 110 were detected inat least 80% of the participants (87 aminoacids and 23 purine and pyrimidine me-tabolites), and we designated them as

commonly detected. Eleven metaboliteswere significantly associated with eGFRslope in the analysis adjusted for multipletesting, and nine of them (81%) weremodified metabolites. Overall, the me-tabolomic platform’s coverage included39 modified metabolites featuringfive types of modifications: acylation,C-glycosylation, carbamoylation, sulfa-tion, and methylation. Metabolites ofall types of modifications, except formethylation, contained numerous bio-chemicals associated nominally withnephropathy progression. Nine modifiedmetabolites remained significant in theanalysis adjusted for multiple testing.C-glycosyltryptophan, N6-carbamoyl-threonyladenosine, and O-sulfotyrosinewere the top-three metabolites, of whichhigher levels were associated with amore-rapid eGFR slope. All nine signifi-cant modified metabolites correlatedwith baseline eGFR at a nominal signifi-cance level (P , 0.05) (Fig. 1A).

A PC analysis of all well-detected me-tabolites (n = 114) explained only asmall proportion of the overall vari-ance. PC1all explained 15.5% of the var-iance, and PC2all explained only 7.1%.Score plots of all PCs with eigen-values .1 were examined and did not

separate well decliners from nonde-cliners (data not shown). Of note, toploadings contributing to the PC1all corre-sponded with the nine significant modi-fied metabolites identified in the globalmatrix analysis. In the further PC analysisthat considered only these nine metab-olites, PC19 was the only componentwith an eigenvalue .1 and explained68.3% of the variance. Contributions toPC19 were comparable among thesemetabolites, except for N6-acetyllysineand phenol sulfate, for which the load-ings were lower. The related PC scoreplot decently separated declinersfrom nondecliners (Fig. 1B). PC19 cor-related significantly with eGFR slope(r = 20.28).

The multivariate analysis did notreveal a predictive value of essentialamino acids or branched chain aminoacid–related metabolites for renal func-tion decline. Creatinine was the top me-tabolite associated with baseline eGFR,but only moderately correlated with thesubsequent eGFR slope. Analysis of ratiosof modified metabolites (products) totheir respective substrates did not im-prove risk predictions over the effects ofthe sole modified metabolites (data notshown). Associations of all detected

Figure 1—Associations of metabolites with subsequent renal function decline. A: Global matrix of correlation coefficients of all commonly detectedamino acids and nucleotides (n = 114) with baseline eGFR (x-axis) and with a subsequent eGFR slope (y-axis). Each mark represents a correlationcoefficient of an individual metabolite. Modified metabolites are represented as open shapes (blue,, C-glycosylated; red4, sulfated; orange○,acylated; green◇, carbamylated; purple bars, methylated). All the remaining metabolites are presented asC. B: Score plot of PC1 and PC2 basedon nine significant candidate metabolites identified in the global matrix analysis. Marks represent individual study participants with renal functiondecline in the bottom tertile of the eGFR slope distribution (decliners) and those in the top tertile of the eGFR slope distribution (nondecliners). Thelabels of the x- and y-axes include the explained variance. CGST, C-glycosyltryptophan; CRE, creatinine; N6CA, N6-carbamoylthreonyladenosine;NATHR, N-acetylthreonine; OST, O-sulfotyrosine; PSD, pseudouridine.

386 Modified Metabolites and ESRD Risk in Diabetes Diabetes Care Volume 40, March 2017

metabolites with eGFR slope and theirinterrelationships expressed as loadingsare listed in Supplementary Table 1.

Modified Metabolites as CandidateBiomarker Predictors of Progressionto ESRD

Multivariable Analysis

We used a Cox proportional hazardsregression model to examine the in-dependence of associations of themodified metabolites with risk of pro-gression to ESRD. All nine modified me-tabolites associated with eGFR slopewere also associated significantly withprogression to ESRD in the Cox propor-tional hazards regression model aftercontrolling for the covariates HbA1c,ACR, and eGFR. Seven of the nine me-tabolites (except for N-acetylalanineand phenol sulfate) remained signif-icant in the fully adjusted model,controlled for an expanded panel ofclinical covariates (Fig. 2). Discrim-ination ability of the clinical modelmeasured by C-index was 0.706. Theaddition of each significant metabolite(one at a time) increased the C-indexfrom 0.708 to 0.722. Seven significantmetabolites correlated among each,with nonparametric correlation coeffi-cients ranging from r = 0.45 to r = 0.89(P , 0.0001 for each) (Fig. 3). TheC-index of the Cox model containing ametabolite index derived from PC19

and controlled for clinical covariates in-creased to 0.728, but an improvement inthe discrimination ability did not reachstatistical significance.

Within-Individual Reproducibility

Over Time

Repeatedmeasurements ofwell-detectablemetabolites over a median of 5 yearsdemonstrated that 36% of amino acids(31 of 87) and 13% of purine and pyrim-idine metabolites (3 of 23) had at leastfair reproducibility over time (ICC .0.3)(global data not shown). Among the ninemodified metabolites significant in themultivariate analysis, pseudouridinetracked very well over time (ICC 0.74[95% CI 0.48, 0.88]). Reproducibility ofC-glycosyltryptophan, O-sulfotyrosine,and N-acetylthreonine was fairly good(ICC 0.34–0.38) and was poor for theremaining five metabolites (Fig. 2).

Metabolite Levels in the Study Participants

Compared With Subjects Without

Nephropathy

We compared levels of the nine signif-icant modified metabolites in the mul-tivariate analysis between the studyparticipants with T1D, proteinuria,and CKD (main study) and 20 subjectswith T1D, normoalbuminuria, and me-dian diabetes duration of 24 years. Ra-tios of metabolites in the main studyparticipants compared with that in sub-jects with normoalbuminuria ranged

from 1.21 to 1.94 (P , 0.05 for eachcomparison) (data not shown).

Candidate Metabolite Biomarkers Across

Independent Cohort Studies

We identified 13 metabolites measuredin the current study that were formerlyreported to be associated with ne-phropathy phenotype in other indepen-dent metabolomics studies (3,6,8,9,17).C-glycosyltryptophan and pseudouri-dine, O-sulfotyrosine, and tryptophanmetabolites were reproducibly associ-ated with renal outcome in this andat least one other independent study(Supplementary Table 2).

Modified Metabolites as PotentialContributors to Diabetic NephropathyProgression

Modified Metabolites Versus Glycemic

Control and Kidney Injury Markers At

Baseline

We created a matrix of the correlationcoefficients between the nine sig-nificant modified metabolites in themultivariate analysis and baseline mea-surements of HbA1c (glycemic control),markers of glomerular damage (urinaryexcretion of albumin and IgG2), andmarkers of tubular damage (urinaryexcretion of KIM-1 and NGAL). Allnine modified metabolites were alsocorrelated, to at least a moderate de-gree, with baseline eGFR (r = 20.48to 20.3). The pattern of associations

Figure 2—Nine modified metabolites significant in the global multivariate analysis as candidate biomarkers for progression to ESRD. Reproducibilityover time is expressed as ICCs. HRs are presented per 1 SD of themetabolite. Partially adjusted HRs are controlled for baseline HbA1c, ACR, and eGFR.Fully adjusted HRs are controlled for blood pressure, BMI, smoking status, HbA1c, ACR, eGFR, uric acid levels, treatment with renin-angiotensinsystem inhibitors, other antihypertensive treatment, and statins.

care.diabetesjournals.org Niewczas and Associates 387

with markers of tubular injury (KIM-1or NGAL) mirrored associations ob-served with a baseline eGFR, except forN-acetylthreonine, N6-acetyllysine, andphenol sulfate, whose associations werenotably weaker. Associations of the me-tabolites with ACR and IgG2 were absentor weak (Fig. 3).

Renal Clearance of the Modified

Metabolites and Risk of Progression

to ESRD

To elucidate whether an increase in me-tabolites in participants with proteinuriaand CKD is not due to impaired clearance,we conducted an auxiliary study of sub-jects with comparable proteinuria butnormal renal function for whom weevaluated metabolite levels in plasmaand urine according to the nephropa-thy risk. Clinical characteristics of theauxiliary study subjects are presented inSupplementary Table 3.Median follow-upfor the auxiliary study subjects was 12.3years. Follow-up for nonprogressors was6.9–21 years and 2.4–19 years for ESRDprogressors. The library of the metabolo-mic platform in the auxiliary study had lesscomprehensive coverage than the mainstudy because only four significant modi-fiedmetabolites weremeasured. Risk pat-terns remained similar to that observed

in the main study. The auxiliary studywas small; therefore, confidence inter-vals around HRs were understandablybroader. Of note, in the excretion patternsanalysis,metabolitesmeasured in urine ormeasured as fractional excretions werenot significantly associated with progres-sion to ESRD (Supplementary Fig. 1).

We also compared free acetyllysinewith circulating proteomic content ofacetyllysine in a subset of subjects. Wedid not observe a significant correlationbetween the two. These results suggestthat circulating acetyllysine residues donot likely reflect proteolysis of circulat-ing proteins (Supplementary Data).

CONCLUSIONS

This study was conducted in a cohort ofpatients with T1D, proteinuria, and CKDstage 3 at baselinewhowere followed forapproximately a decade to assess eGFRtrajectories and identify time to onsetof ESRD. To our knowledge, this and ourformer study in T2D (3) are the only todate that evaluated the associations ofcirculating metabolomic profiles withprogression to the ultimate outcome ofdiabetic nephropathy: ESRD.

In the current study, global metabo-lomic profiling revealed nine candidate

metabolite biomarkers as poten-tial risk factors of progression toESRD in T1D. These metaboites wereC-glycosyltryptophan, pseudouridine,O-sulfotyrosine, N-acetylthreonine, N-acetylserine, N6-carbamoylthreonylade-nosine, N6-acetyllysine, N-acetylalanine,and phenol sulfate. The effect of thesemetabolites on risk of ESRD remainedindependent after adjustment for glyce-mic control, renal function, and albumin-uria. Seven of these nine (except forN-acetylalanine and phenol sulfate) re-mained significant in multivariable mod-els adjusted for a comprehensive selectionof clinical covariates. C-indices of modelscontaining added biomarkers were higherthan theC-indexof the clinicalmodel alone,but improvements in the discriminationabilities of models were not significant,likely reflecting that the study may nothave had sufficient power because of thenumber of applied adjustments.

Do these findings have the potentialfor generalization? This report and twoothers (Joslin Kidney Study in T2D [3]and Cooperative Health Research in theRegion of Augsburg [KORA] F4 [9]) havereported that C-glycosyltryptophan,pseudouridine, and O-sulfotyrosine areamong the top risk metabolites that

Figure 3—Matrix of correlation coefficients among and between modified metabolites and baseline indices of diabetic kidney injury: glycemiccontrol (HbA1c), glomerular markers (ACR, IgG2), and tubular markers (KIM-1, NGAL). Cell color indicates strengths and directions of the correlationsfrom red (positive correlation) to white (no correlation) to blue (negative correlation).

388 Modified Metabolites and ESRD Risk in Diabetes Diabetes Care Volume 40, March 2017

predict GFR-based renal outcome. KORAF4 evaluated incident CKD in the generalpopulation, and all three studies usedthe same metabolomic platform. TheFramingham cohort study used a differentplatform (6) wherein the authors identifiedtryptophan derivatives as determinants ofincident CKD. Tryptophan-derived metabo-lites also reached nominal significance inthe current study.Do the candidatemetabolites actually

contribute to disease progression? Thisstudy demonstrates that significantmodified metabolites markedly corre-late with one another and further withbaseline indices of kidney tubular injury.We hypothesize that a common, cur-rently unknown mechanism leads to anincrease in these modified metabolitesin circulation and that thesemetabolitesmay be toxic to kidney tubules, contrib-uting to subsequent eGFR loss. Func-tional studies are needed to validatesuch a hypothesis.The presence of impaired renal clear-

ance at baseline in the current studyraises concerns about whether reten-tion can account for increased serumlevels of metabolites (9,22–24). Never-theless, effects of metabolites were in-dependent from baseline renal functionin the multivariable models examined.Second, we provide limited evidencethat renal clearances of the selectedmodified metabolites were unalteredwith regard to ESRD risk.One may speculate that an increase in

significant metabolites may also be due tothe increased posttranslational modifi-cations in tissues of patients at risk.Increased glycosylation, sulfation, acetyla-tion, and carbamoylation of amino acidsand proteins have been reported in dia-betic and uremic tissues (25–28). The im-portance of the circulating metabolitesrather than their urinary levels suggestssources other than the kidney, such as liverand muscle. No studies to our knowledgehave systematically examined the contentof modified metabolites across tissues inthe presence of diabetes. However, basicresearch studies have demonstrated thatthe modification of metabolites identi-fied in the current study have been de-termined in many tissues of subjectswithout diabetes, including kidney, liver,and muscle (27–35). A number of studiesalso have highlighted the importance ofmodifications of metabolites in the pres-ence of chronic diabetic complications

(25–28,34,36–38). Different modifica-tions are involved in a variety of biologicalprocesses, such as inflammation (36,37),glucose metabolism (34,35), and epige-netic regulation (38), among others.

Finally, wemust consider the strengthsand limitations of the current study. Thestudy was conducted in Caucasians, so itmay not be generalizable to populationsof other ethnic ancestry. Our study isnovel because the modified metaboliteswere studied as risk factors for subse-quent progression of diabetic nephropa-thy; on the contrary, in past cross-sectionalcomparisons, posttranslational modifi-cations in diabetic complications wereevaluated as features associated withan already-present disease. The currentstudy includes a well-characterized pop-ulation with a long follow-up and ascer-tainment of ESRD together with eGFRslopes. Associations of the metabolitesalsowere concordant between two renaloutcomes and in other studies in T2Dand the general population (3,6,9). Fu-ture replication in independent prospec-tive cohorts of T1D is needed to validatethese predictive associations in patientswith this disease phenotype. An im-portant advantage of the nontargetedmetabolomic approach utilized in ourstudy is that it examines metabolitescomprehensively and assigns a rankingto identified metabolic alterations. Forexample, not all modifications wereassociated with disease progression(acetylation was, whereas methylationwas not).

In summary, the metabolomic ap-proach is an attractive line of investigationin diabetes because of the metabolicnature of the disease. It is relevant inkidney disease because the kidney is in-volved in handling numerous metabolites,and the metabolites themselves may con-tribute further to kidney injury. We reportassociations ofmetabolomic determinantswith risk of ESRD in T1D. Future directionsshould involve elucidating mechanismsthat underlie increased acetylation orC-glycosylation of metabolites in patientsat increased risk for renal disease,whereasindependent replication studies will ex-pand our knowledge about the potentialgeneralizability of the current findings.

Funding. This study was supported by grantsfrom JDRF (career development award 5-CDA-2015-89-A-B to M.A.N. and 17-2013-311 to A.S.K.)and from the National Institute of Diabetes and

Digestive and Kidney Diseases (DK-072381to J.V.B., DK-082841 to S.P., and DK-41526 toA.S.K.).Dualityof Interest. J.V.B. is co-inventoronKIM-1patents, which have been licensed by PartnersHealthcare to several companies. He has receivedroyalties from Partners Healthcare and grantfunding from Novo Nordisk. No other potentialconflicts relevant to this article were reported.Author Contributions. M.A.N. designed thestudy, supervised experimental data collection,analyzed and interpreted data, and wrote themanuscript. A.V.M. supervised measurementsof acetyllysine, analyzed the data, and edited themanuscript. S.C. contributed to the data collec-tion and performed the relevant laboratorymeasurements. J.B. performed the measure-ments of acetyllysine and analyzed the data.M.M. examined patients and contributed to thedata collection. V.S.S. performedmeasurementsof urinary KIM-1 and analyzed the data. A.S.supervised phenotypic data collection and con-tributed to the data analysis. J.V.B. contributedto the data interpretation and edited the man-uscript. S.P. contributed to the study design anddata interpretation and edited the manuscript.A.S.K. designed the study, contributed to thedata interpretation, and edited the manuscript.M.A.N. is the guarantor of thiswork and, as such,had full access to all the data in the study andtakes responsibility for the integrity of data andthe accuracy of the data analysis.

References1. Krolewski AS. Progressive renal decline: thenew paradigm of diabetic nephropathy intype 1 diabetes. Diabetes Care 2015;38:954–9622. Skupien J, Warram JH, Smiles AM, et al. Theearly decline in renal function in patients withtype 1 diabetes and proteinuria predicts the riskof end-stage renal disease. Kidney Int 2012;82:589–5973. Niewczas MA, Sirich TL, Mathew AV, et al.Uremic solutes and risk of end-stage renal dis-ease in type 2 diabetes: metabolomic study.Kidney Int 2014;85:1214–12244. Merchant ML, Perkins BA, Boratyn GM, et al.Urinary peptidome may predict renal functiondecline in type 1 diabetes and microalbuminu-ria. J Am Soc Nephrol 2009;20:2065–20745. Pezzolesi MG, Satake E,McDonnell KP, MajorM, Smiles AM, Krolewski AS. Circulating TGF-b1-regulated miRNAs and the risk of rapid progres-sion to ESRD in type 1 diabetes. Diabetes 2015;64:3285–32936. Rhee EP, Clish CB, Ghorbani A, et al. A com-bined epidemiologic and metabolomic ap-proach improves CKD prediction. J Am SocNephrol 2013;24:1330–13387. Sas KM, Karnovsky A, Michailidis G,Pennathur S. Metabolomics and diabetes: ana-lytical and computational approaches. Diabetes2015;64:718–7328. Sharma K, Karl B, Mathew AV, et al. Metab-olomics reveals signature of mitochondrial dys-function in diabetic kidney disease. J Am SocNephrol 2013;24:1901–19129. Sekula P, Goek ON, Quaye L, et al. A metab-olome-wide association study of kidney func-tion and disease in the general population.J Am Soc Nephrol 2015;27:1175–1188

care.diabetesjournals.org Niewczas and Associates 389

10. Rosolowsky ET, Skupien J, Smiles AM, et al.Risk for ESRD in type 1 diabetes remains highdespite renoprotection. J Am Soc Nephrol 2011;22:545–55311. Warram JH, Gearin G, Laffel L, Krolewski AS.Effect of duration of type I diabetes on the prev-alence of stages of diabetic nephropathy de-fined by urinary albumin/creatinine ratio. J AmSoc Nephrol 1996;7:930–93712. Inker LA, Schmid CH, Tighiouart H, et al.;CKD-EPI Investigators. Estimating glomerular fil-tration rate from serum creatinine and cystatinC. N Engl J Med 2012;367:20–2913. U.S. Renal Data System.USRDS 2010AnnualData Report: Atlas of Chronic Kidney Diseaseand End-Stage Renal Disease in the UnitedStates. Bethesda, MD, National Institutes ofHealth, National Institute of Diabetes and Diges-tive and Kidney Diseases, 201314. Center for Disease Control and Prevention;National Center for Health Statistics. National DeathIndex [Internet]. Available from https://www.cdc.gov/nchs/ndi. Accessed 1 September 201315. Dehaven CD, Evans AM, Dai H, Lawton KA.Organization of GC/MS and LC/MS metabolomicsdata into chemical libraries. J Cheminform2010;2:916. Ohta T, Masutomi N, Tsutsui N, et al. Un-targeted metabolomic profiling as an evaluativetool of fenofibrate-induced toxicology in Fischer344 male rats. Toxicol Pathol 2009;37:521–53517. LookerHC, ColomboM,Hess S, et al.; SUMMITInvestigators. Biomarkers of rapid chronickidney disease progression in type 2 diabetes.Kidney Int 2015;88:888–89618. Gohda T, Walker WH, Wolkow P, et al. Ele-vated urinary excretion of immunoglobulins innonproteinuric patients with type 1 diabetes.Am J Physiol Renal Physiol 2012;303:F157–F162

19. Sabbisetti VS, Waikar SS, Antoine DJ, et al.Blood kidney injury molecule-1 is a biomarker ofacute and chronic kidney injury and predictsprogression to ESRD in type I diabetes. J AmSoc Nephrol 2014;25:2177–218620. Vaidya VS, NiewczasMA, Ficociello LH, et al.Regression of microalbuminuria in type 1 diabe-tes is associated with lower levels of urinarytubular injury biomarkers, kidney injury mole-cule-1, and N-acetyl-b-D-glucosaminidase. Kid-ney Int 2011;79:464–47021. Pencina MJ, D’Agostino RB Sr, D’AgostinoRB Jr, Vasan RS. Evaluating the added predictiveability of a new marker: from area under theROC curve to reclassification and beyond. StatMed 2008;27:157–172; discussion 207–21222. Duranton F, Cohen G, De Smet R, et al.;European Uremic Toxin Work Group. Normaland pathologic concentrations of uremic toxins.J Am Soc Nephrol 2012;23:1258–127023. Rhee EP, Souza A, Farrell L, et al. Metaboliteprofiling identifies markers of uremia. J Am SocNephrol 2010;21:1041–105124. Tanaka H, Sirich TL, Plummer NS, WeaverDS, Meyer TW. An enlarged profile of uremicsolutes. PLoS One 2015;10:e013565725. Ihara Y,Manabe S, KandaM, et al. Increasedexpression of protein C-mannosylation in theaortic vessels of diabetic Zucker rats. Glycobiol-ogy 2005;15:383–39226. Kanan Y, Brobst D, Han Z, Naash MI, Al-Ubaidi MR. Fibulin 2, a tyrosine O-sulfatedprotein, is up-regulated following retinal de-tachment. J Biol Chem 2014;289:13419–1343327. Kosanam H, Thai K, Zhang Y, et al. Diabetesinduces lysine acetylation of intermediary me-tabolism enzymes in the kidney. Diabetes 2014;63:2432–2439

28. Kalim S, Karumanchi SA, Thadhani RI, BergAH. Protein carbamylation in kidney disease:pathogenesis and clinical implications. Am JKidney Dis 2014;64:793–80329. Hille A, Rosa P, Huttner WB. Tyrosine sulfa-tion: a post-translational modification of pro-teins destined for secretion? FEBS Lett 1984;177:129–13430. Moore KL. The biology and enzymology ofprotein tyrosine O-sulfation. J Biol Chem 2003;278:24243–2424631. Furmanek A, Hofsteenge J. ProteinC-mannosylation: facts and questions. ActaBiochim Pol 2000;47:781–78932. Charette M, Gray MW. Pseudouridine inRNA: what, where, how, and why. IUBMB Life2000;49:341–35133. Chow CS, Lamichhane TN, Mahto SK. Ex-panding the nucleotide repertoire of the ribo-some with post-transcriptional modifications.ACS Chem Biol 2007;2:610–61934. Liu R, Zhong Y, Li X, et al. Role of transcrip-tion factor acetylation in diabetic kidney dis-ease. Diabetes 2014;63:2440–245335. Zhao S, Xu W, Jiang W, et al. Regulation ofcellular metabolism by protein lysine acetyla-tion. Science 2010;327:1000–100436. Wang Z, Nicholls SJ, Rodriguez ER, et al. Pro-tein carbamylation links inflammation, smoking,uremia and atherogenesis. Nat Med 2007;13:1176–118437. Farzan M, Mirzabekov T, Kolchinsky P, et al.Tyrosine sulfation of the amino terminus ofCCR5 facilitates HIV-1 entry. Cell 1999;96:667–67638. Kato M, Natarajan R. Diabetic nephropathydemerging epigenetic mechanisms. Nat RevNephrol 2014;10:517–530

390 Modified Metabolites and ESRD Risk in Diabetes Diabetes Care Volume 40, March 2017